Files White Papers Protected Veeam Be2010 Test Openbench-jack-fegreus

Customer Value

VMware Backup and Recovery

openBench Labs

PERFORMANCE

Analysis:

DataCenterManagement

Commissioned by

Veeam Backup & Replication v5vs.

Symantec Backup Exec 2010 R2

Author: Jack Fegreus, Ph.D.Managing DirectoropenBench Labs

http://www.openBench.comApril 21, 2011

PERFORMANCE

Analysis:

Jack Fegreus is Managing Director of openBench Labs and consults throughRidgetop Research. He also contributes to InfoStor, Virtual Strategy Magazine,and Open Magazine, and serves as CTO of Strategic Communications.Previously he was Editor in Chief of Open Magazine, Data Storage, BackOfficeCTO, Client/Server Today, and Digital Review. Jack also served as a consultantto Demax Software and was IT Director at Riley Stoker Corp. Jack holds a Ph.D.in Mathematics and worked on the application of computers to symbolic logic.

VMware Backup and RecoveryVeeam Backup & Replication v5

vs.Symantec Backup Exec 2010 R2

Table of Contents

03

Table of Contents

Test Brief 04Real Issues With Virtual Machine Data Protection 04Recovery-Centric Test Plan 05

Infrastructure Capabilities 08Performance Foundation 08Isolating I/O Bottlenecks 08

Backup Performance 10Test 1: Agentless Backup 13Test 2: Backup File Reduction 15Test 3: Single Execution of All Options 17

Recovery Performance 21Test 4: Direct VM Recovery From a Backup File 22Test 5: Automated Recovery Verification 26Test 6: Multi-OS File-Level Recovery 29Test 7: Application-Independent Item-Level Recovery 30Test 8: Point-in-Time VM Replication 32

Test Summary 34

REAL ISSUES WITH VIRTUAL MACHINE DATA PROTECTIONIn a 2010 Symantec datacenter

survey, IT sites without a virtualinfrastructure project planned or inprogress represented just 18% of the1,780 sites surveyed. On the otherhand, roughly half of the sites with avirtual operating environment wereplanning to expand the use of virtualmachines (VMs) by running mission-critical database-driven applications.

Nonetheless, Symantec’s 2010Disaster Recovery (DR) Plan surveyindicates that IT regularly backs uponly 56% of the applications and datarunning on VMs. What’s more, DRplans typically exclude 60% of virtualservers at a site. As for data andapplications backed up under theumbrella of a DR plan, those backupfiles fail to comply with eitherRecovery Time Objectives (RTOs) orRecovery Point Objectives (RPOs) on40% of IT’s tests to recover criticaldata and applications.

Despite what seems to be a ratherlax concern for VM data protection,CIOs and Line of Business Executives

alike cite the strength of a virtual environment as a business continuity solution. Whenasked to cite key reasons for adoption of virtual infrastructure, IT decision makers listthe ability to rapidly restart, non-disruptive move, and spawn new instances of VMs.With high availability, disaster recovery and data protection core components of any

Test Configuration

Test Brief

“The gulf between IT’s expectations of the benefits that can be derivedfrom a virtual infrastructure and the reality of the lagging data

protection measures put into place highlights the fact that a highly flexiblehost infrastructure does not solve the knotty issues IT faces trying toprotect the applications and data running on VMs.”

UNDER TEST: VM DATA-PROTECTION FEATURESVeeam Backup & Replication v5 Software

Symantec Net Backup 2010 R21) Agent-less Backup Configuration: Test basic backup performance andfunctionality along with any dependencies on the need to license,manually deploy, or manage agents on either VMs or vSphere hosts.

2) Image File Reduction via Data Compression and Deduplication: Bothdata compression and data deduplication can be applied to imagebackups to reduce backup file requirements.

3) Single Execution for Multiple Backup and Recovery Options: Easy-to-implement backup and recovery operations with all recovery options,including the whole VM, guest VM files and VM application items,available from a single backup.

4) Direct VM Recovery from a Backup File: Instantly start VM recoveryfrom a backup file without rehydrating data, and complete the processwith Storage vMotion, VM cloning, or Veeam’s VM Copy wizard, whichuses Veeam’s FastSCP.

5) Automated Recovery Verification: Perform a test restore for everybackup file as a standard backup validation test.

6) File-Level Recovery: VM image-level backups allow any VM to berecovered in full or as a collection of guest OS data files.

7) Application-Item-level Recovery: Quick recovery of individual objectsfrom any virtualized application, such as an Oracle Database, MicrosoftExchange Server, or Microsoft SharePoint.

8) Point-in-Time VM Replication: Replicate and fail over VMs using low-impact synthetic backups for near continuous data protection (CDP) thatminimizes RPO for any VM.

04

business continuity strategy, CIOs view virtual infrastructure as the best way to assuagedown-time fears of corporate executives within budget constraints that precludeimplementing costly (DR) components, such as fault-tolerant servers, redundanthardware resources, and additional software licenses.

The gulf between IT’s expectations ofthe benefits that can be derived from avirtual infrastructure and the reality ofthe lagging data protection measures putinto place highlights the fact that ahighly flexible host infrastructure doesnot solve the knotty issues IT facestrying to protect the applications anddata running on VMs. Moreimportantly, when surveyed about theproblems related to protecting VM data,a plurality of the top issues cited by ITwere related to the recovery of datarather than the backing up of data. Inparticular the two reasons most oftencited to explain why IT organizationsinfrequently test DR plans centered onthe costs associated with provisioningthe resources necessary to stage a

recovery test, and the disruption that staging such a test would cause for staff.

RECOVERY-CENTRIC TEST PLAN

To reflect growing concerns for data recovery within the overall data protection process,openBench Labs set up an eight point test plan with five key tests focused on data recoveryissues. We applied this test plan to five VM workflows. These workflows represented:

1) A corporate email scenario, with Exchange Server 2010 and a Domain Controller,2) A 64-bit database-driven business process, with SQL Server 2008 R2, and a

Windows level 2008 Domain Controller, 3) A 32-bit database-driven business process, with SQL Server 2005, and a

Windows level 2003 Domain Controller, and a Windows 7 workstation,4) A standalone Linux server, 5) A standalone Windows workstation running Windows 7 Enterprise.

We tested the ability to validate backup files, including the ability to verify that abackup file is recoverable. We then extended this line of testing to assess the time andresources required to fully restore a VM from any backup file. In addition, we focused onthe ability to quickly perform fine-grained restore processes from a single backup file ofa VM. In these tests, we examined the ability to easily recover guest files and application-level items from VMs running multiple versions of Windows- and Linux-based guestoperating systems.

Test Configuration

05

PRINCIPAL VM DATA PROTECTION CHALLENGES1) Faster Data Recovery: In surveys of IT sites by both Symantec andVeeam, the top data protection challenge was the time necessary forrestoration, which was cited with roughly 10% greater frequency thanthe time to run a backup process.

2) IT Budget Resources Required to Test DR VM Restoration: Whenasked about more frequent testing of restore operations, 60% of thesites in the Symantec DR survey listed budgetary constraints onresource provisioning as the biggest impediment.

3) Employee Disruption Staging DR VM Tests: In a virtual dead heatwith resource provisioning issues, 59% of datacenters reported thedisruption to employees as limiting their ability to test DR restoration ofbackup files.

4) Fine Grained Restoration of VM Data: Entwined with the need forrestore processes in the Symantec survey was the need for moregranular data restoration in the forms of VM data files and applicationitems, such as individual databases.

Test Configuration

HOW WE TESTED KEY VM DATA PROTECTION FEATURESBACKUP SERVER:Dell PowerEdge 1900• Quad core CPU• 4GB RAM• QLogic QLE2462 4Gbps HBA• Quantum DLT-S4 Tape Drive• Windows 2008 Server R2• Veeam Backup & Replication v5• Symantec Backup Exec 2010 R2• vCenter Server 4.1• up.time 5 • Iometer Benchmark

VSPHERE 4.1 HOSTS:Dell PowerEdge 1900 • Quad core CPU• 8GB RAM• QLogic QLE2462 4Gbps HBA• VMware ESX 4.1HP ProLiant DL580 • Quad processor CPU• 8GB RAM• QLogic 2462 4Gbps HBA• VMware ESX 4.1

SAN INFRASTRUCTURER:QLogic SANbox 9002 FC SwitchXiotech Emprise 5000 ISE• (2) Balanced 4.2TB DataPacs

VM WORKFLOWS:Windows Server 2008 R2 VM• Active Directory, DNS, DHCPWindows Server 2008 R2 VM• Exchange Server 2010 • 500 mailboxes (400MB per mailbox)

Windows Server 2003 R2 VM• Active Directory, DNS, DHCPWindows Server 2003 R2 VM• SQL Server 2005Windows Server 2008 R2 VM• SQL Server 2008 2

SUSE Linux Enterprise Server 11 • oblDisk Benchmark

Windows 7 VM• MS Office 2010• Iometer Benchmark

06

On front-end backup processes, we concentrated our tests on functions to reduce thetime needed to backup a workflow and the volume of storage that a workflow wouldconsume. To minimize storage requirements, we tested the ability to invoke datacompression and data deduplication. To minimize backup window time, we testedsupport for incremental backups and the ability to generate full-featured syntheticbackups for use in any restore process.

We concluded our tests with an examination of VM replication. In particular, wetested the ability to apply backup optimization techniques to achieve near continuousdata protection (near-CDP) for VMs that can meet aggressive RTO and RPO demands.

07

Test Configuration

PERFORMANCE FOUNDATION

Before starting our evaluation of Veeam Backup & Replication v5 with Backup Exec2010 R2, we first ran hardware benchmarks on our test configuration to determine allhardware limitations that could affect our data protection testing. We used two serversrunning the VMware vSphere ESX hypervisor to host our vSphere 4.1 test scenarios. Allbackup and restore tests using Veeam Backup & Replication v5 and Symantec BackupExec 2010 R2 were performed from a Dell PowerEdge 1900 Server with a Quad-core,4GB RAM, and a dual-port 4Gbps QLogic QLE 2462 FC HBA.

We provisioned logical disks on our backup server, toprovide storage for both Veeam Backup & Replication andBackup Exec image files, as well as datastores for our vSpherehosts on a Xiotech Emprise 5000 ISE. To facilitate server andstorage monitoring during our backup and restore testing, weinstalled up.time 5 service level monitoring on our backupserver. The up.time 5 software is able to monitor and reporton all virtual and physical servers, and any other resourceswith an SNMP interface. As a result, we would be able tocompare process loads on all systems during a backup, whichis a growing concern at IT sites.

To frame the maximum disk throughput that ourinfrastructure was capable of delivering in a backup process,we ran Iometer on our Dell PowerEdge 1900 backup server.To simulate a backup process, we streamed large-block(128KB) reads and writes simultaneously to two logicaldrives from the Emprise 5000. We measured total read andwrite throughput at 950 MB per second. These results peggedour ability to support a straight-through end-to-end backupprocess without data compression or data deduplication ataround 450MB per second.

ISOLATING I/O BOTTLENECKS

As part of our virtual infrastructure configuration, we set up a test domain thatincluded two VMs. One VM took on the role of the domain controller and ran WindowsServer 2008 R2 with Active Directory (AD), DNS, and DHCP. The other VM played the

Infrastructure Capabilities

08


“Using Iometer to benchmark I/O throughput for our backup server,we supported two simultaneous I/O streams: one read and one

write that ranged from 450MB to 475MB per second each using large128KB blocks similar to backup applications.”

SIMULATED BACKUP THROUGHPUT

Using Iometer, two simultaneous I/O streams, onereading and one writing data , each sustained450MB to 475MB per second using 128KB datablocks, similar to backup applications.

role of an email server running Exchange Server 2010 on Windows Server 2008 R2.

We configured this server with three additional logical diskvolumes to support 500 email accounts. To balance emailtransactions, we used two 125GB logical volumes to distributeour 500 mailboxes across two mailbox databases. Each databasecontained 250 mail boxes, each of which held an average of400MB of message data. The third volume was used to storeExchange log files associated with mailbox databasetransactions.

With 500 active mailboxes, our configuration was typical of asingle Exchange server at a small to medium enterprise (SME)site or a distributed Exchange server used to scale up a largeenterprise site by distributing the load and improving resiliency.Next, we configured the Jetstress benchmark to place atransaction load on our Exchange Server to keep approximately100% of our mailboxes busy processing 1 transaction persecond. When we measured actual performance with up.time 5,Jetstress transactions were engaging all of our mailboxes at a rateof 1.6 transactions per second.

To put that level of performance in perspective, over an 8-hour business day, each mailbox would process over, 46,000transactions. The value of being able to support such aperformance level rests in the ability to easily measure theoverhead placed on a VM with direct recovery from a backup

file and the transition to a SAN-based datastore via vMotion.


09

VM EMAIL TRANSACTION LOAD TEST

We set up Jetstress to generate an emailtransaction load that would keep all mailboxesbusy. With up.time 5, we measured theJetstress load as actually averaging 800 emailand 340 log transactions per second.

BACKUP SOFTWARE ARCHITECTURE DIFFERENCES

We installed both Veeam Backup & Replication 5 and Symantec Backup Exec 2010 R2on the same Dell PowerEdge server. A central issue in comparing the use of Backup Exec2010 R2 with Veeam Backup & Replication v5 in a vSphere environment is the scope ofthe two products. Veeam Backup & Replication is designed explicitly for a VMwareenvironment. Symantec Backup Exec, on the other hand, is designed to support an entiresite, with both physical and virtual servers. As a result, most Backup Exec 2010 R2modules apply to both physical and virtual servers, which complicate a comparison ofSymantec’s Backup Exec 2010 Agent for VMware Virtual Infrastructure with VeeamBackup & replication v5.

Comparing the functionality ofVeeam Backup & Replication v5with Backup Exec 2010 R2’s Agentfor VMware Virtual Infrastructureis dependent on the modules thatare installed in the base BackupExec 2010 R2 package. Manycross-architecture modules,including a Shared Storage Optionfor a SAN and a DeduplicationOption, played important roles inour vSphere environment and arerequired to match the functionality

provided by Veeam Backup & Replication v5.

These cross-architecture modules, however, significantly drive up the total cost of aBackup Exec 2010 R2 installation. What’s more, the need for all Backup Exec 2010 R2modules to work in a physical server environment frequently adds overhead to a vSphereenvironment. For example, Backup Exec 2010 R2 optimizes integration with multiplephysical and virtual tape libraries. As a result, Backup Exec 2010 R2 implements anersatz tape library structure to perform disk-to-disk (D2D) backups.

Implementation of an enterprise-class data deduplication solution complicatesBackup Exec 2010 R2’s use of a simulated tape library structure for D2D backups. In ageneric enterprise-class deduplication scheme, backup files that feature deduplicateed,

Backup Performance

Backup Performance

“The ability of Veeam Backup & Replication to implement datadeduplication and compression with both full and incremental

backup files opens the door for IT to support extremely aggressive RTO and RPO goals in an SLA for business continuity.”

10

VM BACKUP PERFORMANCE TESTS:1) Agentless VM Backup: Test basic backup performance and functionalityalong with any dependencies on the need to license, manually deploy, ormanage agents on either VMs or vSphere hosts.

2) Image File Reduction via Data Compression and Deduplication: Bothdata compression and data deduplication can be applied to image backupsto reduce backup file requirements.

3) Single Execution for Multiple Backup and Recovery Options: Easy-to-implement backup operations for both full and incremental backups with allrecovery options, including the whole VM, guest VM files and VM applicationitems, available for any backup.

single-instance data are not self-contained. These files contain pointers to a repository ofsingle-instance data representations. In turn, that necessitates the creation of a speciallibrary within Backup Exec 2010 R2 to handle the media used in D2D backups with datadeduplication.

For our tests, creating two distinct D2D devices for backups with and without datadeduplication meant that we had to create two distinct devices: a Backup-to-Disk folderand a Deduplication Storage folder. To simplify device management, we assigned adistinct virtual volume to each of these folders. Backup Exec then created media files inboth of these directories and wrote backup data into the respective media files.

ALONG THE DATA DEDUPLICATION DIVIDE

For D2D backups, the dual-device structure of Backup Exec 2010 R2 aligned with amajor backup distinction between Veeam Backup & Replication v5 and Backup Exec2010 R2 with respect to compression and deduplication. While Veeam Backup &Replication v5 allows all data reduction techniques, including data deduplication, datacompression, and incremental backup, to be used in a single job, Backup Exec 2010 R2,must apply data deduplication exclusively and only to a full backup.

Veeam Backup & Replication v5implements a simple datadeduplication scheme that comparesonly the data contained within anactive job stream and makes eachdeduplicateed backup file completelyself-contained. Since a Veeam backup file contains all of the information needed forrehydration, any backup file can be restored from any server running Veeam Backup &Replication v5.

For a VM containing multiple versions of data files and multiple structures such asdatabase tables, Veeam’s active stream deduplication often provided data reduction onthe order of 40% to 50%. Moreover, with self-contained backup files, Veeam allows ITadministrators to combine data deduplication with other data reduction techniques,including data compression. With both techniques in play, we typically experienced totaldata reduction rates between 15-to-1 and 20-to-1.

What’s more, if a company has standardized on Symantec Backup Exec for physicalservers, it can leverage this capability to compound the benefits of using Veeam forVMware virtualized servers. The self-contained structure of Veeam’s backup filesprovides IT with an option to add tape archiving to Veeam Backup & Replication v5 viathe backup of its host server. In this way, Backup Exec can be used to meet any off siterequirements for backups of VMs.

On the other hand, the PureDisk service, which handles all data deduplication forBackup Exec 2010 R2, takes a holistic time-centric perspective to VM data, which istypical of most enterprise-level data-deduplication schemes designed to deduplicate data

Backup Performance

11

“If a company has standardized on SymantecBackup Exec for physical servers, it can

leverage this capability to compound the benefitsfrom using Veeam for VMware virtualized servers.”

from all sorts of generic sources. The goal of these deduplication schemes is to discoverdata segments that are backed up repeatedly over time and replace them with pointers toa single-instance of the data. To achieve this goal each enterprise-class data deduplicationscheme develops arcane algorithms to scan and break a data stream it into multiplesegments in a way that improves the likelihood of making as many matches as possible ina way that is as efficient as possible.

In addition to the algorithms used to identify redundant data, a fundamentalconstruct of any deduplication scheme is how the pointers work. Inline datadeduplication schemes, such as PureDisk, use backwards referencing which to replaceredundant data in the current backup with a pointer to a previous instance of that datawhile the backup process is running. Comparing data blocks and substituting matchingblocks with pointers adds very significant overhead to the backup process and evenchanges the characteristics of I/O operations.

In the PureDisk scheme, a Backup Exec 2010 R2 media server maintains a catalog ofhash-coded data fingerprints that belong to unique segments of data discovered duringfull backups over time. This makes storage reduction for a current backup job dependenton the rate of cache hits for the job’s data segments with respect to fingerprints in themedia server’s catalog. What’s more, the caching of backup segment fingerprints createsan affinity between a VM backup processes and a physical media server. The segmentcatalog belonging to the media server that last backed up a VM is very likely to providethe highest number of cache hits on the next backup of that VM.

As a media server’s segment fingerprint catalog grows over time, cache hits increaseand the sizes of backup files decrease; however, growing the amount of data in thesegment fingerprint catalog also increases the number of potential comparisons that canbe made during a backup and the time needed to complete the process. That makes itimportant for PureDisk to grow segment catalogs in a way that increases cache hitswithout burdening search time.

As a result, Backup Exec 2010 R2 prohibits the use of software compression, whichalters the structure of data segments, with deduplication, and data deduplication cannotbe used with either incremental or differential backups, which by definition copy onlynew data segments. On the other hand, the ability of Veeam Backup & Replication toimplement data deduplication and compression with both full and incremental backupfiles opens the door for IT to support extremely aggressive RTO and RPO goals in anSLA for business continuity.

Backup Performance

12

AGENT SYNCHRONIZATION

We began our assessment by examining functionality with respect to dependencies onlicensing, deploying, or managing agents on either individual VMs or vSphere hosts. Inaddition, we also tested basic backup performance as a baseline for all testing.

With Veeam Backup &Replication v5, we were able toperform all backup operationson any VM—whether Linux-or Windows-based —withoutfirst installing an agent. Weneeded to install and maintaina Backup Exec 2010 R2 RemoteAgent on any Windows-basedsystem—virtual or physical—for which we intended torestore individual files or item-level data.

For IT, the installation ofmultiple Remote Backup Exec2010 R2 Agents on backup-clients can be easily pusheddown from a media serverusing a wizard. Nonetheless,updating clients is typicallynecessary after every update ofa media server t. Clientsoftware must be updated toleverage fixes and functionalitychanges to the media server.

What’s more, with no agentsoftware available for a clientrunning a Linux OS, we could

not perform file-based restores directly from a backup of our VM running SUSE LinuxEnterprise v11. In order to restore user files using Backup Exec 2010 R2, we needed torun a separate Windows-network-based backup of directories exposed using Samba onour Linux-based VM.

Test 1: Agentless Backup

13

“We needed to install and maintain a Backup Exec 2010 R2Remote Agent on any Windows-based system—virtual

or physical—for which we intended to restore individual files oritem-level data.”

BACKUP EXEC AGENT

Using Backup Exec 2010 R2, remote agent software—available only for Windows-basedclients—was required to restore file-level data from a VM backup. We were also able torestore item-level data from a VM running either Active Directory, Exchange Server, or SQLServer with Granular Recovery Technology enabled. While a wizard on the media servermade it possible to update multiple clients with a single push process, we often found itnecessary to reboot clients individually after some updates.

Test 1: Agentless VM Backup

In our initial tests, all backups were performed withoutdata deduplication or data compression. We measured theelapsed process time of the backup window and the size ofthe resulting backup file. In addition we examined the CPUprocess load and the disk throughput load imposed on thebackup server during the backup window via up.time 5.

Not only were results similar across VMs, the patternestablished continued with more sophisticated processingoptions. Veeam consistently leveraged the highest level ofavailable CPU resources to complete backup processes asquickly as possible. As a result, Veeam Backup & Replicationv5 wrote data faster and completed jobs faster than BackupExec 2010 R2.

More importantly, Veeam Backup & Replication v5removes backup verification from the backup window. Byfocusing only on data protection for VMs, Veeam has beenable to introduce new technologies that redefine alltraditional IT notions of system recovery.

For our VM acting as a primary domain controller with11.9 GB of active data, the backup process took just over 2minutes with Veeam Backup & Replication and 5 minutesBackup Exec 2010 R2. To verify the backup data, the fullbackup window took just over 7 minutes. Similarly, backing

up our VM running Exchange Server 2010 took nearly 21 minutes with Veeam Backup &Replication v5 and just over 49 minutes with Backup Exec 2010 R2, which required overan hour and 6 minutes in completing the entire process.

Test 1: Agentless Backup

14

VM Server Backup PerformanceFull backup, No Data Deduplication or Compression

VM ParametersBackupSoftware

ActiveData

BackupFile

Backup Window(hh:mm:ss)

Applications: Active Directory,DNS Server, DHCPServer OS: Windows Server 2008 R2

Veeam 11.9GB 11.5GB 0:02:090:02:09

BackupExec

11.9GB 12.1GB 0:05:090:05:090:07:07 0:07:07 with verify

Application: SQL Server 2008 R2OS: Windows Server 2008 R2

Veeam 20.8GB 19.3GB 0:03:290:03:29

BackupExec

20.8GB 20.3GB 0:04:360:04:360:07:00 0:07:00 with verify

Applications: Jetstress,Exchange Server 2010OS: Windows Server 2008 R2

Veeam 228.8GB 229.0GB 0:20:410:20:41

BackupExec

228.8GB 234.7GB 0:49:050:49:051:06:38 1:06:38 with verify

BACKUP SERVER CPU &DISK LOAD

Backing up a primary domain controller VM, VeeamBackup & Replication v5 used up to 85% of aprocessor on a quad-core server, while Backup Exec2010 R2 peaked at 75%. Writing the backup files,Veeam reached 215MB per second, while BackupExec peaked at 185MB per second. Backup Execalso kept the backup window open to verify backupdata by reading the backup file (265MB per second)and calculating a checksum.

THE DATA DEDUPLICATION DIVIDE

With backup files for our VM running Exchange Server consuming approximately230GB with either Veeam Backup & Replication or Backup Exec 2010 R2, the need toreduce backup-file size is undisputable. In our second set of backup tests, we examinedthe ability to apply multiple data reduction techniques to backup files.

While there is a strong consensus that the technologiesneeded for backup file reduction are data deduplication andcompression, there are wide differences in the application ofthese technologies. In fact, some of the biggest performancedifferences between Veeam Backup & Replication v5 andBackup Exec 2010 R2 arise from differences in the way thateach package approaches data deduplication.

Veeam Backup & Replication v5 does not store or comparedata beyond the context of a running job. As a result, a VMbackup can utilize both data deduplication and compression.Moreover, since a Veeam backup file is self-contained, the filehas no dependencies on physical servers or support files. Thatmeans a Veeam backup file can be backed up to tape as aWindows Server file, trasnported to any Windows Serverrunning Veeam Backup & Replication v5 to recover the VM.

On the other hand, Backup Exec 2010 R2 uses an inlineenterprise-class data deduplication scheme that employs acatalog of hash-codes—often dubbed ‘fingerprints’—of singleinstance data segments. The catalog is used to simplify the useof backwards-referencing pointers to unique instances ofpreviously encountered data segments. When a data segmentin the current backup is discovered to match an existing datasegment, the segment in the backup file is replaced by a copyof the pointer in the catalog.

This method creates interesting dependencies on time andplace. Frequent backups of a VM on a server will populatethat server’s catalog with the most accurate up-to-datecollection of data segment fingerprints to boost the dtatdeduplication cache hit rate. When we repeated a backup of aVM on a media server without changing any user data, thedata deduplication rate jumped from 76% to 90%.Nonetheless, when we repeated that backup process a third

Test 2: Backup File Reduction

BACKUP FILE REDUCTION

While backing up a 20.8GB VM running WindowsServer 2008 R2 and SQL Server 2008 R2, weapplied both data compression and datadeduplication with Veeam Backup & Replicationv5. Veeam completed the backup in less than 6minutes and reduced the backup file to 8.5 GB.Using Backup Express 2010 R2, a backup withsoftware compression took nearly 21 minutes andcreated an 11.4GB backup file. On the other hand,a backup with data deduplication, took close to 18minutes and created a backup file with just a4.2GB footprint.

“Full backups configured for data reduction took up tonine times longer to complete with Backup Exec 2010 R2,

compared to Veeam Backup & Replication v5.”Test 2: Backup File Reduction

15

time after defragmenting the VM’s logical disk, the deduplication rate fell back to81%.percent.

Inline data deduplication requires significant CPU and memeory resources, especiallyif data reduction rates of around 90% or higher are to be reached. In terms of leveragingavailable CPU resources, Veeam Backup & Replication v5, which implements a relativelysimple data depuplication scheme, typically utilized all available CPU cycles on ourquad-core server to minimize backup widows.

In contrast, BackupExec 2010 R2, whichruns a far moresophisticateddeduplication service,typically utilized onlytwo processor cores. As aresult, backupsconfigured for datareduction took up tonine times longer tocomplete using BackupExec 2010 R2, whencompared to VeeamBackup & Replication v5.

Even with datadeduplication combinedwith data compression,Veeam Backup &Replication v5 oftenprovided around a 5-to-1advantage in wall clock

time for a backup with respect to Backup Exec 2010 R2. What’s more, Veeam also createdconsistently smaller back files compared to using software compression with BackupExec 2010 R2. These advantages were increased by packaging multiple servers in a singlebackup, which increased the ability for Veeam to apply data deduplication. In particular,by backing up our SQL Server 2008 R2 and a domain controller VMs as a workflow withVeeam, we cut the backup file foot print by approximately 30% and still retained allrecovery options for each server.

Test 2: Backup File Reduction

VM Server Backup PerformanceFull backup: Data Deduplication/Compression and Verification


ActiveData

BackupFile

Backup Window(hh:mm:ss)

Applications: Active Directory,DNS Server, DHCPServer OS: Windows Server 2008 R2

Veeam 11.9GB 4.5GB 0:03:070:03:07

BackupExec

11.9GB6.3GB1.2GB

0:16:30 0:16:30 Compression0:14:05 0:14:05 Deduplication

Application: SQL Server 2008 R2OS: Windows Server 2008 R2

Veeam 20.8GB 8.5GB 0:05:400:05:40

BackupExec

20.8GB12.2GB4.0GB

0:20:450:20:45 Compression

0:17:580:17:58 Deduplication

Applications: Active Directory,DNS Server, DHCPServer SQL Server 2008 R2OS: (2)Windows Server 2008 R2

Veeam 32.7GB 10.2GB 0:09:230:09:23

BackupExec

32.7GB19.2GB2.6GB




Veeam 228.8GB 13.5GB 0:19:590:19:59

BackupExec

228.8GB21.0GB8.4GB



16

INCREMENTAL EQUALITY

In our third series oftests, we examined theability of each package tosimplify operations for ITadministrators. Inparticular, we focused onthe ability to implementall backup options,including full andincremental, from a singleGUI command. We alsoexamined the ability toapply all recovery options,including VM guest OSfiles and data items, toeach backup file. Whilevirtualization’s reasond’être is simplification ofresource management,data center managerscontinue struggling withthe complexity ofmeeting RTO and RPOdata protection targets ina virtual environment.

We were able to runincremental and fullbackups interchangeablyusing Veeam Backup &Replication v5, which bydefault uses an initial fullbackup followed byincremental backups. Tomake the most recentbackup the fastest torestore, we used reversed

Test 3: Single Execution for All Options

17

“With the ability to apply any backup or restoreoption to an incremental backup, even IT at a

small SMB shop can leverage just the local backupcapabilities of Veeam Backup and Replication v5 to

support a business continuity SLA with RPO and RTO requirements of less than two minutes.”


INCREMENTAL BACKUP: VEEAM BACKUP & REPLICATION V5

We set the default backup mode for our PDC VM running Active Directory to run reversedincremental backups continuously. To reduce the size of backup files we applied both datadeduplication and compression . As a result, an incremental backup of our VM with 12GB of activedata took about 90 seconds and generated a 25MB rollback file. More importantly, we were ableto leverage all backup and recovery features:

incremental backups: The incremental backup file is used to create a new synthetic fullbackup from the previous full backup and a rollback file generated to recreate theprevious full backup to maintain that restore point. What’s more, we leveraged the shorttime to run an incremental backup and the small size of the rollback file to schedulecontinuous incremental backups—as a backup completed, a new backup started—andthereby maintain an RPO of less than two minutes.

In addition toapplying all of theoptions available for afull backup, whilerunning anincremental backup,we were able to runevery restore optionwith each incrementalbackup file, whetherthe backup file was arollup or rollback file.Using a wizard, wewere able to pick anyrecovery pointassociated with eithera full or incrementalbackup and run all ofthe restore options.

In particular, wewere able to utilizeour incrementalbackup file as arestore point with theVeeam Instant Restorefeature, which allows aVM to run directlyfrom a backup file andenables IT to complywith an RTO of lessthan two minutes.With the ability toapply any backup orrestore option to anincremental backup,even IT at a small

SMB shop can leverage just the local backup capabilities of Veeam Backup andReplication v5 to support a business continuity SLA with RPO and RTO requirements of

18

INCREMENTAL BACKUP: BACKUP EXEC 2010 R2

Using Backup Exec 2010 R2, we had to create a special policy to run an incremental backup. Inaddition, we were not able to implement data deduplication or Granular Recovery Technology with anincremental backup. As a result, restore wizard allowed us to recover the vSphere vmdk filesassociated with a VM backed up with an incremental backup; however, the wizard did not make any ofthe local Exchange Server 2010 files on our VM available from an incremental backup.


less than two minutes. For an IT administrator, the difference between an incrementaland a full backup are measured in time and storage volume.

On the other hand, Backup Exec 2010 R2, with its reliance on agents, theincompatibilities of differing D2D backups, and the overhead of an inline deduplicationscheme that uses a disk segment cache, is not able to provide all restore options to everystandard backup of a VM. In particular, IT administrators can perform a file-level restoreonly on a full backup of a VM that runs Windows and has a Backup Exec agent installed.Moreover, incremental and differential backups for VMs require a special backup policyin order to execute and cannot utilize data deduplication.

INLINE BLOCK POINTING GAMBIT

In a vSphere environment, there is a bigger question concerning the need for aninline enterprise-class data deduplication scheme. Can data segments repeated over timebe discovered and removed from a VM backup file with significantly less overhead? Theanswer can be found in the vStorage APIs for Data Protection (VADP) and ChangedBlock Tracking (CBT).

When scanning image backup data, an enterprise-class data deduplication schemetargets the discovery of ordered sequences of physical data blocks that have been foundin a previous backups. In essence, the software looks for sequences of blocks that havenot changed over time. That’s why defragmentation of a logical volume absent explicitchanges to data will degrade the cache hit rate of a backup that uses backwardsreferencing pointers. On the other hand, excluding data blocks that have not changedfrom the last backup of a vSphere VM is a trivial process using CBT in conjunction withan incremental backup.

To test our ability to maximize segment caching, we ran two successive full backupsof our VM running Exchange Server 2010 using Backup Exec 2010 R2 with the PureDiskdeduplication service. On the second run, the deduplication rate was 98%, which wasmanifested in an 8.4GB backup file. Nonetheless, the backup window required a full 3hours and 2 minutes.

Next we applied an incremental backup with Veeam Backup & Replication v5. Withboth data deduplication and compression turned on, the incremental backup filegenerated with vStorage and CBT started at 926.4MB. Modest internal deduplication anda great deal of compression reduced the size of that backup file to 260.6MB. Moreimportantly the entire process took just 1minute and 52 seconds.


19

VPOWER TO MEET RECOVERY-CENTRIC SLAS

For IT, backup has long been a necessary daily activity centered on backing up data,applications, and OS files in a minimal amount of time, which is dubbed the backupwindow. From the perspective of a Line of Business (LoB) executive, however, the valueof IT data-protection operations rests entirely in recovery processes. That’s why data-protection Service Level Agreements (SLAs) between IT and LoB units focus on RPOand RTO targets: Backup is simply a means to provide a recovery point.

With the introduction of new vPowertechnology, Veeam Backup &Replication v5 revolutionizes IT’s abilityto perform fast VM backup operations,test the recoverability of 100% ofbackups outside of the backup window,and meet strict RPO and RTOrequirements. With traditional backupsoftware, backup validation consists ofchecking that the bits in the backup filematch the bits in the original VM. Thisamounts to rereading all of the data andcalculating a checksum.

Traditional checksum validationextends the time needed for a backupand only ensures that the VM and theVM backup are duplicates. This

validation method cannot check if the original VM has not been corrupted to the pointthat it is no longer capable of being rebooted. To verify that a backed system will actuallyrun, Veeam Backup & Replication v5 has introduces a number of key technologies underthe banner of vPower.

A key vPower technology is the ability to launch a VM directly from a backup file:This capability entirely changes the construct of VM recovery for IT operations. From aprocess perspective, booting a VM directly from a backup file begins with placingpointers to the files contained within the backup file inside a special directory on theVeeam Backup & Replication v5 server. That directory is exported to the vSphere hostvia NFS as a network attached datastore with read-only files.

Recovery Performance

VM RECOVERY PERFORMANCE TESTS:4) Direct VM Recovery from a Backup File: Instantly start VMrecovery from a backup file without rehydrating data, and completethe process with Storage vMotion, VM cloning, or Veeam’s VM Copywizard, which uses Veeam’s FastSCP.

5) Automated Recovery Verification: Perform a test restore for everybackup file as a standard backup validation test.

6) Multi-OS VM File-Level Recovery: VM image-level backups allowany VM to be recovered in full or as as a collection of logical guestOS data files.

7) Application-Independent Item-level Recovery: Quick recovery ofindividual objects from any virtualized application, such as an OracleDatabase, Microsoft Exchange Server, or Microsoft Sharepoint.

8) Point-in-Time VM Replication: Replicate and fail over VMs usinglow-impact synthetic backups for near continuous data protection(CDP) that minimizes RPO for any VM.


“Akey vPower technology is the ability to launch a VM directly froma backup file: This capability entirely changes the construct of VM

recovery for IT operations.”

20

With all of the VM virtual disks represented as read-only files via pointers, writingnew data is handled through cache files that Veeam adds to the vPower NFS datastore bydefault. To improve write performance, new data can be redirected to a differentdatastore. If this option is implemented, Veeam Instant Recovery triggers a snapshot ofthe VM and redirects the snapshot and CBT files to a VeeamIR directory in thedesignated datastore. In this case the vSphere snapshot handles the new data and cachefiles are not created.

While a VM running from a backup file cansustain a distinctly higher IOPS load with dataredirection, there is an operational drawbackwhen consolidating the original and new dataon a production datastore. If data redirectionhas not been utilized, vMotion can be used to consolidate that data on a productiondatastore while the VM remains running online. On the other hand, if data redirectionhas been used to put new data in a snapshot on a second datastore, vMotion cannot beutilized. An IT administrator will need to hot-clone the VM, power down the VM, andpower on the cloned copy.

Another vPower construct is a Virtual Lab, which creates an isolated environment,into which administrators can boot and test backed-up VMs. In addition to providing awizard to easily configure a virtual lab, Veeam provides test scripts to run on VMs toensure service roles on domain controllers, database servers, and email servers arerunning properly. In a matter of minutes, test scripts can verify that a VM as backed upwith no configuration changes boots properly and its key applications run correctly.What’s more these tests are run outside of the backup window while the original VMcontinues to run in the production environment.

Nonetheless, the lion’s share of restore operations for IT remain focused on retrievingindividual files inadvertently lost or corrupted in day-to-day use. In a virtual infrastructure,this translates into the need to easily restore user files under the VM’s guest OS. To thisend, Veeam Backup & Replication v5 provides a wizard that works directly withdepuplicated and compressed backup files to restore any file associated with any guest OS.What’s more, Veeam Backup & Replication v5 extends the construct of file-level datarestore for any guest OS to object-level data for any application on any guest OS.


“Veeam Backup & Replication v5extends the construct of file-level

data restore for any guest OS to object-leveldata for any application on any guest OS.”

21

RECONCILING RTO WITH REALITY

For LoB executives, RTO represents the maximum amount of time before anorganization is negatively impacted by the interruption of a core business process. As aresult, an SLA that targets business continuity must be driven by the needs of thebusiness unit. As such, a business continuity SLA needs to satisfy a worst possible timefor an interruption to occur. More importantly, the SLA should not degenerate into areflection of IT capabilities.

For IT administrators,recovering a VM to aworking state thatcorresponds to a previouspoint in time oftenrequires several hours ofwork. Symantec’s 2010DR survey pegs the totalamount of time spent inrecovering a system toaverage about five hours.

We began recovery testing by assessing the time to do a traditional restore or our VMrunning Exchange Server. In particular, we measured the time period needed to extractand rehydrate VM data files from a backup file and then write the data to a vSphere hostand register the VM. This time period did not include any storage provisioning activitiesbefore instituting the restore process or any post-process testing of VM functionality. Justthe restore process for our VM running Exchange Server 2010 took 2hours, 43 minutes,and 35 seconds, when we used Backup Exec 2010 R2 with its data deduplication option.Even our best traditional restore time ran 1 hour, 28 minutes, and 35 seconds withVeeam Backup & Replication v5.

With traditional system restore techniques measured in hours, RTO issues havebecome serious problems for CIOs as more LoB executives link computer downtime tomore than lost sales revenue. Many LoB executives have upped the ante by equatingcomputer outages with potential losses in customer confidence and market share. As aresult, senior LoB executives expect IT to meet a recovery time objective (RTO) that ismeasured in hours and in some cases even minutes. The challenge for many IT decisionmakers is how to assuage the business continuity fears of corporate executives withinbudget constraints that exclude costly components, such as fault-tolerant servers,redundant hardware and software, and standby network bandwidth.

Traditional VM Restore PerformanceFull Restoration with Rehydration of Deduplicated/Compressed Data


RestoredData

BackupFile

Restore Window(hh:mm:ss)


Veeam 235.4GB 13.0GB 1:28:351:28:35

BackupExec

235.4GB22.1GB8.8GB

2:26:00 2:26:00 Compression2:43:35 2:43:35 Deduplication

“Veeam takes VM backup verification fromsimple byte comparisons to validating VM

backup files boot, internal VM applications work,and external function correctly.”

Test 4: Direct VM Recovery from a Backup File

22


Unique to Veeam Backup & Replication v5 is the ability to start and run a VMdirectly from a backup file without first going through provisioning and full restoreprocesses. In just 22 seconds we were able to boot and publish a VM running ExchangeServer into our vSphere production environment. The tradeoff for blazing recoveryspeed comes with lower sustained IOPS rates. Nonetheless, it is possible to bring a VMrunning from a backup file back to full performance levels with less than 30 seconds oftotal server down time.

INSTANT RECOVERY START UP

We used the Instant Recovery wizard to start our VM running Exchange Server 2010 directly from a backup file. Within 5 seconds, aVeeam server directory was populated with pointers to the contents of the VM backup file and exported to the vSphere server as avPower NFS datastore. By default, Veeam added cache files to handle new data written to each of the VM’s logical disks. When wechose to redirect new data updates for Instant Recovery, Veeam triggered a VM snapshot and redirected the snapshot and CBT files toa directory dubbed VeeamIR on the datastore of our choice. With the snapshot now handling data updates, there was no need forspecial cache files.

23


We ran a number of tests with the Veeam Instant Restore feature. In these tests weconcentrated on RTO and performance requirements for SLAs. In particular, we ranJetstress with all data directed to the vPower NFS datastore and with updates redirectedto another datastore using a redirected snapshot. In addition we tested IOPS rates whileconsolidating data for both cases.

In our initial test, we did not redirect data updates from the vPower NFS datastore. Asa result, writes were captured in local cache files and we were able to leverage vMotion.

INSTANT RECOVERY WITH VMOTION

Using Veeam Instant Recovery, our VM running Exchange Server was published in vSphere within 22 seconds from an incrementalbackup file. Within another 67 seconds we were able to log into the VM and launch Jetstress. At the same time, we started a vMotionmigration of the VM from the vPower NFS database to a SAN-based datastore. The vMotion process took 2 hours and 50 minutes. Theincremental Veeam backup and the Veeam Instant Recovery process took a total of just 2 minutes. That put the total time to backup andfully recover our VM at 2 hours and 52 minutes with VM downtime less than 30 seconds.

24


By redirecting update data for the logicaldisks to another datastore, we were able todouble the level of Jetstress transactions thatwe could sustain. What’s more, this relativedifference was carried over to theconsolidation processes for these InstantRecovery scenarios. During the hot-cloneprocess with the backup VM running withdata redirection, we sustained an IOPS loadvery similar to the IOPS load that we hadsustained by the default Instant Recoveryconfiguration, which did not use dataredirection.

In particular, we sustained approximately150 email and 50 log transactions per secondwhile consolidating data by hot-cloning thebackup VM with data redirection. Incomparison, without data redirection we wereable to sustain roughly 60 email transactionsand 25 log transactions per second whileimplementing vMotion to consolidate data forour backup VM. That IOPS level represents afairly typical transaction processing load formany mission critical applications. Inparticular, it equates to 3,456 emailtransactions—reads, writes, and deletes—peruser mailbox over an 8-hour work period,

Consolidation with hot cloning tookroughly 12% more wall clock time. While thevMotion process completed in 2 hours and 50

minutes, hot cloning took 3 hours and 10 minutes to complete. We then needed to shutdown the backup VM and start the cloned VM manually. In addition, the consolidationprocess can be performed by invoking the VM Copy wizard , whch uses Veeam FastSCP.

In both cases we had the option to start consolidating the recovered VM at any pointin time. We could easily wait to run vMotion or hot-cloning during an off-peakprocessing period. More importantly, using Veeam Backup & Replication v5, we wereable to perform an incremental backup and then recover our VM running ExchangeServer 2010 with Veeam Backup & Replication v5 in less time than it took just to run abackup using data deduplication with Backup Exec 2010 R2.

25

INSTANT RECOVERY IOPS PERFORMANCE

With the Exchange Server running from a backup file with dataredirection, we sustained significantly more email transactions runningthe Jetstress benchmark. With data updates redirected via a snapshot,we averaged 290 email and 105 log transactions per second. Withoutredirection, we averaged 140 email and 38 log transactions per second.


FULL RECOVERY VALIDATION

Using Backup Exec, backup validation consists of checkingthat the bytes in a backup file match the bytes in the VM. Thisamounts to rereading the data in the backup file, calculating achecksum, and then comparing that checksum against acheckuum calculated while backing up the original VM.

This method significantly extends the backup windowand does not ensure that the backed up version actuallyworks. If the VM has has had critical OS files or applicationfiles corrupted but not identified, the backup file will be anexact duplicate of the original problem. Depending on theseverity of the problem, if the origial VM is shut down,neither the original nor the backed-up VM may be bootable.

To verify that a backed-up VM and its applications willrun correctly, Veeam Backup & Replication v5 introduced anew feature, dubbed SureBackup Recovery Verification,which non disruptively verifies the full recovery of a VM.Using the vPower NFS datastore construct utilized withInstant Recovery, SureBackup starts a backed-up VMassociated with a recovery point directly from its backup file.SureBackup Recovery Verificaton, however, does not publishthe backed-up VM to the production environment.

In contrast to Instant Recovery, SureBackup creates aprotected Virtual Lab environment, in which backups ofproduction VMs can run while the original VMs continue tooperate. IT administrators create virtual lab environments

using a wizard that configures a private network, a vSwitch, and a gateway appliance thatmasquerades recovered VMs with dynamic IP addresses outside the production network.As a result, administrators can test VMs recovered from backup files using the VM’soriginal IP address and node name without creating conflicts with production VMs.

An IT administrator can also statically configure the appliance to masquerade aspecific private-network address with a specific production-network address. Using thistechnique, we discovered and monitored VMs in the virtual lab with up.time 5. Anotherconfiguration option for each virtual lab is the choice of a datastore for redo logs, whichare automatically assigned to recovered VMs running within the lab environment.Unlike Instant Recovery, which uses cache files or redirects data with a snapshot, updates

Test 5: Automated Recovery Verification

“Veeam takes VM backup verification from simplebyte comparisons to validating that VM backup

files boot, internal VM applications work, and externaldependencies on other VMs function correctly.”


26

TRADITIONAL BACKUP VALIDATION

The default backup scenario with Backup Exec2010 R2 includes the verification of backup dataimmediately after writing a backup file. In our tests,the time to reread the backup file was dependent onthe data reduction method used in the backup. Whiledata compression consistently added more time tobackup processes, data deduplication consistentlyadded more time to verification processes.

to virtual disks are handled by redo logs.

Beyond validating a backup file of a single VM, SureBackup Recovery Verificationsupports the construct of an application group, which corresponds to an LoB workflow.Wihin a group, VMs are placed in an ordered boot-sequence based on role andapplication dependencies. Then each VM is assigned a series of tests to verify that all keyroles are functioning properly. As a result, Veeam takes VM backup verification fromsimple byte comparisons to validating that VM backup files boot, internal VMapplications work, and external dependencies on other VMs function correctly.

Another important aspect of SureBackup is the ability to leave a virtual labenvironment running after verification tests complete. In addition, an IT administratorcan statically configure the appliance to masquerade a specific private-network address

27


SUREBACKUP RECOVERY VERIFICATION

We tested Veeam’s SureBackup Recovery Verification feature by testing the recoverability of a backup of a VM acting as a primarydomain controller. We started our backed up VM directly from a backup file using the vPower NFS datastore and updates to the virtualdisks files of the VM were handled by redo logs. Once the virtual lab and virtual lab network were started, SureBackup published the VMin the lab network and tested that the VM was running. Finally, scripts were run to test the VM’s roles as an Active Directory domaincontroller, a global catalog server, and a DNS server. The entire process took 9 minutes and 13 seconds.

with a specific adress in the production network. Using this technique, we used up.time 5to discover and monitor backed-up VMs running in a virtual lab. We also ran RDPsessions on production systems that connected to the backed-up VM.

By initiating a session with the backed-up VM from its corresponding productionVM, we were able to restore object-level data using application tools. Using thistechnique, we successfully leveraged a SureBackup virtual lab environment with the mostcurrent versions of Microsoft server infrastructure, including Exchange Server 2010 andSQL Server 2008 R2, to access application items captured in backup files. As a result,there is no need to wait for a backup software upgrade when a new version of a guestVM’s OS or application is introduced.

28


RESTORATION EX MACHINA

Using Backup Exec 2010 R2, we were not able to implement a file-level recovery onany backup file of VM that did not run Windows and that did not have a Backup Exec2010 R2 agent installed. In contrast, any backup file created with Veeam Backup &Replication v5 can be restored on a logical, file-level basis for any VM OS.

Using the Veeam Backup& Replication v5 restorewizard, we were able torecover files from a backupfile of a VM running SUSELinux Enterprise v11. In thisprocess, Veeam Backup &Replication leveraged thevPower NFS datastoreconstruct used with InstantRecovery and SureBackup tolaunch a File Level Recovery(FLR). appliance thatpresented the files in thetarget backup file.

The restore wizard usesthe FLR appliance, whichruns on the vPower NFSdatastore, to present a logicalVM file-level view of thecontents of the target backupfile. The Veeam FLRappliance and restore wizardwork independently of theVM’s OS, which can beLinux, Unix, or the Mac OS.As a result, IT administrators

can easily drill down on the directory structure of a logical VM in order to restore anyfile.

29

“The Veeam File Level Recovery (FLR) applianceand restore wizard work independently of the

VM’s OS, which can be Linux, Unix, or the Mac OS.”Test 6: Multi-OS VM File-Level Recovery

FILE LEVEL RECOVERY

We were able to perform a file level restore on a VM running SUSE Linux Enterprise Server11 with Veeam Backup & Replication. In particular, we used a wizard, which launched anFLR appliance on a vSphere host in order to present the contents of a backup file that wasmade available via a vPower NFS datastore.

Test 6: File-Level Recovery

OBJECT FOCUS

With the sophistication of VMapplications continuing toincrease, there is also a growingneed for solutions to handle fine-grained problems associated withindividual objects from VMapplications, such as an OracleDatabase, Microsoft ExchangeServer, or Microsoft SharePoint.Restoring object-level data fromapplications is a much morecomplicated task than file-levelrestoration for a guest OS. Whilesome object-level data restorationtasks can be orchestrated byexploiting explicit file structureswithin the guest OS,comprehensive restoration ofobject-level data often requires theability to run one or more utilityprograms related to the associatedapplication.

The need to leverage runningapplications to restore object-leveldata aligns precisely with thecapabilities of the Veeam’sSureBackup feature. As a result,Veeam’s vPower technology, whichprovides the ability to run a VMdirectly from a backup file, alongwith the SureBackup virtual Labconstruct, which provides a secureenvironment to run a backup VMin parallel with the originalproduction VM are at the

foundation of Veeam’s Universal Application-Item Recovery (U-AIR).

Test 7: Item-Level Recovery

“Since IT administrators can use anapplication’s own tools to recover

item-level data, there is no need to wait foran upgrade to Veeam Backup &

Replication, when new versions of a guest VM’s OS or application software are introduced.”

Test 7: Application-Independent Item-Level Recovery

30

SQL SERVER GRANULAR RECOVERY

Backup Exec 2010 R2 provides a Granular Recovery Technology option for fullbackups of VMs with a Remote Agent installed and running Active Directory,Exchange Server, or SQL Server. For our VM running SQL Server 2008 R2, we wereable to restore any full database to our VM.

More importantly, there are no restrictions on IT when it comes to invokingapplication tools to repair and restore item-level objects in a virtual lab. We were able tomasquerade the backed up VM on the production environment with a static productionaddress via the virtual lab appliance. As a result, we were able to use the Remote DesktopConnection to set up a session on the backed up VM to run any native application fromany other VM. Since IT administrators can use the application’s own tools to recoveritem-level data, there is no need to wait for an upgrade to Veeam Backup & Replication,when new versions of a guest VM’s OS or application software are introduced.

Test 7: Item-Level Recovery

31

ITEM-LEVEL RECOVERY WITH SUREBACKUP

Item-level restoration with Veeam Backup & Replication v5 can be invoked with the Restore Wizard, which calls SureBackup or it canbe started directly within SureBackup. Using item-level recovery with our VM running SQL Server 2008 R2, we were presented with aricher set of options than we had with Backup Exec 2010 R2. In particular, we were able to restore database items, such as individualtables, views, and stored procedures, to any VM running SQL Server. Furthermore, we could tailor a unique restore operation with adirect SQL query to any database on the backed up VM.

INCREMENTAL RPO

Enterprise-class data replication solutions attempt to meet lossless RPO requirements,and introduce a level of high availability (HA) for DR support. To accomplish these twingoals, replication solutions for virtual environments maintain multiple copies of VMs.

At the highest level of functionality and cost, VMs are kept in lockstep via synchronousdata transfers to all related VMs at the same time, using atomic data writes that must beconfirmed by all devices before they are committed, zero data loss among VMs can beensured. To lower complexity and trim costs, many replication schemes replacesynchronous with asynchronous updates, which can approach a near-lossless RPOdepending on network and device latencies.

For sites that define RPO requirements in minutes rather than milliseconds, point-in-time replication offers a very easy way to implement a solution that provides all of theHA benefits and meets the RPO granularity that the site requires to maintain businesscontinuity. In particular, Veeam Backup & Replication v5 is able to provide IT with avery cost-effective alternative to zero-data-loss, synchronous, replication solutionsthrough the use of periodic reversed incremental backups.

Veeam’s Replication feature essentiallyprovides an HA alternative to the vPowerInstant Recovery feature. With InstantRecovery, the Veeam Server becomes asingle point of failure, unless VeeamBackup & Replication v5 is installed onmultiple servers and the backup files areshared or duplicated, Also direct access tothe second Veeam Backup server isrequired to run Instant Recovery. On theother hand, Veeam Replication creates anoperationally independent VM replicawithin a distributed vSphereenvironment.

The replica is created as a fully-populated VM with the name of the original VM extended with the suffix _replica. Likea standard backup, the replica assumes the logical network identity of the original VMonce it is started. What’s more, periodic incremental backups are stored with the replica’sVMFS files to keep the replica up to date with the original VM.

Test 8: Point-in-Time VM Replication

32

VM REPLICATION TOPOLOGY

We set up a replica of our VM running Exchange Server on a separatevSphere host. We then used automated incremental backups of the originalVM to keep the replica current. Operationally, the replica has all of the dataneed to start at any time within its own datastore directory.

“For large sites that need to replicate mission-critcalVMs to an off-site location, Veeam provides a built-

in solution that overcomes the need to restore data at theDR site. ”


Typically an IT administrator creates an automated schedule of incremental backupupdates to reflect the site’s RPO goal. Veeam supports a continuous update feature, whichlaunches new backup when an existing backup completes. To help conserve space, whentight RPO constraints require frequent backups, the total number of backups stored canbe set to a fixed number, which creates a FIFO (First In First Out) queue for backup files,which represent restore points.

An important aspect ofa replica is that it isdesigned to remainsynchronized with theoriginal VM. As a result,running in failover modeis an aberration, sincesaving any data changeswould break the ability ofthe original VM tosynchronize data stateswith the replica. As aresult, a replica is similarto a VM running directlyfrom a backup file, in thata special mechanism isneeded to store datachanges.

Veeam Backup &Replication v5 uses a VMsnapshot to capture datachanges to a replica infailover mode. From anI/O viewpoint, this makesa replica similar to InstantRecovery with dataredirection. When we ranJetstress on a replica ofour Exchange Server VM,

email and log transaction activity levels were nearly identical to the levels sustained whenwe ran a backup of the VM with Instant Recovery and data redirection.

For large sites that need to replicate mission-critical VMs to an off-site location,Veeam provides a built-in solution that overcomes the need to restore data at the DR site:Only the original VMs need to be restored to enable resynchronization. On the otherhand, sites can still garner all of the RPO, RTO, and performance advantages ofreplication locally using Veeam’s incremental backups and Instant Recovery

33

VM REPLICATION PERFORMANCE

We ran Jetstress on the Exchange Server replica in failover mode and measured performancesimilar to Instant Recovery with data redirection, which also uses a snapshot to capture datachanges. In particular, we sustained 270 email transactions plus 80 log transactions per second.


REWRITING THE RULES OF BUSINESS CONTINUITY

Driving the next wave of virtualization projects are the growing concerns expressedby Line of Business (LoB) executives over business continuity. In a competitive 24x7economic environment, computer downtime represents more than lost revenue to salesand marketing executives. These executives equate lengthy computer outages withpotential losses in customer confidence and market share. As a result, senior LoBexecutives expect IT to meet an RTO that is measured in hours rather than days and anRPO that is close to lossless. With more mission critical applications moving into virtualenvironments, Veeam Backup & Replication’s advanced features are essential for meetingtight data protection constraints.

With respect to meetingRPO and RTO goals, VeeamBackup & Replication v5 wasdesigned explicitly to leveragethe advanced technologyfound in a vSphereenvironment. As a result,Veeam Backup & Replicationv5 is able to support IT inproviding the VMs in avSphere environment withnear lossless RPO andaggressive RTO goalsmeasured in minutes.

To provide this level ofdata protection support,Veeam Backup andReplication v5 introducesvPower technology, whichallows a VM to be started andrun from a backup file. Thekey is a vPower NFS datastorewhich is created on the Veeambackup server, populated withpointers to the contents of the

Test Summary

Test Summary

“Veeam Backup & Replication v5 is able to support IT in providingthe VMs in a vSphere environment with near lossless RPO and

aggressive RTO goals measured in minutes. ”

34

VEEAM BACKUP & REPLICATION V5 FEATURE BENEFITS1) Agentless Backup for any VM: IT administrators do not install agents on VMs or hosts, which allows all backup and restore operations, including file and data-item recovery, to work with any VM running any guest OS.

2) Restart VMs from Backup Files for Near-Instantaneous RTO: Instant VMRecovery allows a VM to be started and run directly from a backup file with dataredirection until the VM is fully restored online using vMotion, hot-cloning, or .Veeam’s FastSCP VM copy option..

3) Test and Authenticate Every Backup Job for Recoverability: SureBackupRecovery Verification starts VMs from backup files and runs application test scriptson the VMs in an isolated virtual lab environment to validate that applications arerunning correctly.

4) Compress and Deduplicate Data Inline on any Backup Process: VeeamBackup & Replication v5 includes compression and inline data deduplication thatcan be combined with incremental VM backups to provide the equivalent spacesavingings of enterprise-class data deduplication systems while dramaticallyreducing the backup window.

5) Automate incremental Backups For Near-Lossless RPO: With datadeduplication and compression applied to incremental backups, which can beused with any recovery process as a restore point, low-impact incrementalbackups can be automated to run continuously.

6) Restore or Repair Application Items: Run VMs from backup files in a virtual labto recover application-related objects using Veeam provided wizards or VM-resident application tools.

backup file, and exported to the vSphere host server as read-only files. Depending on therecovery scenario, Veeam institutes a number of optimal solutions, such as cache files,redo logs, and VM snapshots, for capturing data updates to the VM’s logical volumes. ITadministrators can complete the restoration task by consolidating new and previous dataat any time using vMotion or hot cloning.

Another key aspect of Veeam Backup and recovery is the use of data deduplicationand compression that focuses solely on the current backup process. This enables Veeamto employ these technologies to any backup including incremental backups. This processis very light weight when compared to deduplication schemes that cache and comparefingerprints of all distinct strings of blocks encountered during backup processes. Whilethis scheme provides dramatic data reduction rates, the combination of the vSphere CBTmechanism, incremental backup, and local data deduplication results in similar sizeddata file in a fraction of the wall clock time required for the backup process.

Moreover, IT can double down on the advantages garnered in incremental backupprocessing to utilize very frequent—even continuous—automated incremental backupsto provide minimally spaced recovery points for mission critical systems. For the savvyCIO, Veeam Backup & Replication is able to provide capabilities that go well beyondwhat can be accomplished within a physical server environment. As a result, thesededicated functions will not be found in any general purpose legacy product.

Customer Value

35

Files White Papers Protected Veeam Be2010 Test Openbench-jack-fegreus

Documents