Tech Book: WAN Optimization Controller Technologies

WAN Optimization Controller Technologies

Version 2.0

• Network and Deployment Topologies

• Storage and Replication

• FCIP Configuration

• WAN Optimization Controller Appliances

Vinay JonnakutiEric Pun

WAN Optimization Controller Technologies TechBook2

Copyright © 2012- 2013 EMC Corporation. All rights reserved.

EMC believes the information in this publication is accurate as of its publication date. The information issubject to change without notice.

THE INFORMATION IN THIS PUBLICATION IS PROVIDED “AS IS.” EMC CORPORATION MAKES NOREPRESENTATIONS OR WARRANTIES OF ANY KIND WITH RESPECT TO THE INFORMATION IN THISPUBLICATION, AND SPECIFICALLY DISCLAIMS IMPLIED WARRANTIES OF MERCHANTABILITY ORFITNESS FOR A PARTICULAR PURPOSE.

Use, copying, and distribution of any EMC software described in this publication requires an applicablesoftware license.

EMC2, EMC, and the EMC logo are registered trademarks or trademarks of EMC Corporation in the UnitedState and other countries. All other trademarks used herein are the property of their respective owners.

For the most up-to-date regulator document for your product line, go to EMC Online Support(https://support.emc.com).

Part number H8076.3

Contents

Preface.............................................................................................................................. 5

Chapter 1 Network and Deployment Topologies andImplementationsOverview............................................................................................ 12Network topologies and implementations ................................... 13Deployment topologies.................................................................... 15Storage and replication application................................................ 17

Configuration settings............................................................... 17Network topologies and implementations ............................ 17Notes............................................................................................ 17References ................................................................................... 18

Chapter 2 FCIP ConfigurationsBrocade FCIP ..................................................................................... 20

Configuration settings............................................................... 20Brocade FCIP Tunnel settings.................................................. 20Rules and restrictions................................................................ 21References ................................................................................... 21

Cisco FCIP.......................................................................................... 22Configuration settings............................................................... 22Notes............................................................................................ 22Basic guidelines.......................................................................... 23Rules and restrictions................................................................ 24References ................................................................................... 24

WAN Optimization Controller Technologies TechBook 3

Contents

Chapter 3 WAN Optimization ControllersSilver Peak appliances...................................................................... 26

Overview .................................................................................... 26Terminology ............................................................................... 27Features ....................................................................................... 29Deployment topologies............................................................. 30Failure modes supported ......................................................... 30FCIP environment ..................................................................... 30GigE environment ..................................................................... 31References ................................................................................... 32

Riverbed appliances ......................................................................... 33Overview .................................................................................... 33Terminology ............................................................................... 34Notes............................................................................................ 38Features ....................................................................................... 39Deployment topologies............................................................. 39Failure modes supported ......................................................... 39FCIP environment ..................................................................... 40GigE environment ..................................................................... 42References ................................................................................... 44

WAN Optimization Controller Technologies TechBook4

Preface

This EMC Engineering TechBook provides a high-level overview of theWAN Optimization Controller (WOC) appliance, including network anddeployment topologies, storage and replication application, FCIPconfigurations, and WAN Optimization Controller appliances.

E-Lab would like to thank all the contributors to this document, includingEMC engineers, EMC field personnel, and partners. Your contributions areinvaluable.

As part of an effort to improve and enhance the performance and capabilitiesof its product lines, EMC periodically releases revisions of its hardware andsoftware. Therefore, some functions described in this document may not besupported by all versions of the software or hardware currently in use. Forthe most up-to-date information on product features, refer to your productrelease notes. If a product does not function properly or does not function asdescribed in this document, please contact your EMC representative.

Audience This TechBook is intended for EMC field personnel, includingtechnology consultants, and for the storage architect, administrator,and operator involved in acquiring, managing, operating, ordesigning a networked storage environment that contains EMC andhost devices.

EMC Support Matrixand E-Lab

InteroperabilityNavigator

For the most up-to-date information, always consult the EMC SupportMatrix (ESM), available through E-Lab Interoperability Navigator(ELN) at http://elabnavigator.EMC.com, under the PDFs andGuides tab.

Under the PDFs and Guides tab resides a collection of printableresources for reference or download. All of the matrices, includingthe ESM (which does not include most software), are subsets of the


http://elabnavigator.EMC.com


https://elabnavigator.emc.com

6

Preface

E-Lab Interoperability Navigator database. Included under this tabare:

◆ The EMC Support Matrix, a complete guide to interoperable, andsupportable, configurations.

◆ Subset matrices for specific storage families, server families,operating systems or software products.

◆ Host connectivity guides for complete, authoritative informationon how to configure hosts effectively for various storageenvironments.

Under the PDFs and Guides tab, consult the Internet Protocol pdfunder the "Miscellaneous" heading for EMC's policies andrequirements for the EMC Support Matrix.

Relateddocumentation

The following documents, including this one, are available throughthe E-Lab Interoperability Navigator, Topology Resource Center tab,at http://elabnavigator.EMC.com.

These documents are also available at the following location:

http://www.emc.com/products/interoperability/topology-resource-center.htm

• Backup and Recovery in a SAN TechBook

• Building Secure SANs TechBook

• Extended Distance Technologies TechBook

• Fibre Channel over Ethernet (FCoE) Data Center Bridging (DCB)Concepts and Protocols TechBook

• Fibre Channel over Ethernet (FCoE) Data Center Bridging (DCB)Case Studies TechBook

• Fibre Channel SAN Topologies TechBook

• iSCSI SAN Topologies TechBook

• Networked Storage Concepts and Protocols TechBook

• Networking for Storage Virtualization and RecoverPoint TechBook

• EMC Connectrix SAN Products Data Reference Manual

• Legacy SAN Technologies Reference Manual

• Non-EMC SAN Products Data Reference Manual

◆ EMC Support Matrix, available through E-Lab InteroperabilityNavigator at http://elabnavigator.EMC.com >PDFs and Guides

◆ RSA security solutions documentation, which can be found athttp://RSA.com > Content Library

WAN Optimization Controller Technologies TechBook






http://RSA.com

Preface

EMC documentation and release notes can be found at EMC OnlineSupport (https://support.emc.com).

For vendor documentation, refer to the vendor’s website.

Authors of thisTechBook

This TechBook was authored by Vinay Jonnakuti and Eric Pun, alongwith other EMC engineers, EMC field personnel, and partners.

Vinay Jonnakuti is a Sr. Corporate Systems Engineer in the UnifiedStorage division of EMC focusing on VNX and VNXe products,working on pre-sales deliverables including collateral, customerpresentations, customer beta testing and proof of concepts. Vinay hasbeen with EMC's for over 5 years. Prior to his current position, Vinayworked in EMC E-Lab leading the qualification and architecting ofsolutions with WAN-Optimization appliances from various partnerswith various replication technologies, including SRDF (GigE/FCIP),SAN-Copy, MirrorView, VPLEX, and RecoverPoint. Vinay alsoworked on Fibre Channel and iSCSI qualification on the VMAXStorage arrays.

Eric Pun is a Senior Systems Integration Engineer and has been withEMC for over 12 years. For the past several years, Eric has worked inE-lab qualifying interoperability between Fibre Channel switchedhardware and distance extension products. The distance extensiontechnology includes DWDM, CWDM, OTN, FC-SONET, FC-GbE,FC-SCTP, and WAN Optimization products. Eric has been acontributor to various E-Lab documentation, including the SRDFConnectivity Guide.

Conventions used inthis document

EMC uses the following conventions for special notices:

Note: A note presents information that is important, but not hazard-related.

Typographical conventionsEMC uses the following type style conventions in this document.

Bold Use for names of interface elements, such as names of windows, dialog boxes, buttons, fields, tab names, key names, and menu paths (what the user specifically selects or clicks)

Italic Use for full titles of publications referenced in text


8

Preface

Where to get help EMC support, product, and licensing information can be obtained asfollows:

Note: To open a service request through the EMC Online Support site, youmust have a valid support agreement. Contact your EMC sales representativefor details about obtaining a valid support agreement or to answer anyquestions about your account.

Product informationFor documentation, release notes, software updates, or forinformation about EMC products, licensing, and service, go to theEMC Online Support site (registration required) at:

https://support.EMC.com

Technical supportEMC offers a variety of support options.

Support by Product — EMC offers consolidated, product-specificinformation on the Web at:

https://support.EMC.com/products

The Support by Product web pages offer quick links toDocumentation, White Papers, Advisories (such as frequently usedKnowledgebase articles), and Downloads, as well as more dynamiccontent, such as presentations, discussion, relevant CustomerSupport Forum entries, and a link to EMC Live Chat.

Monospace Use for:• System output, such as an error message or script• System code• Pathnames, filenames, prompts, and syntax• Commands and options

Monospace italic Use for variables.

Monospace bold Use for user input.

[ ] Square brackets enclose optional values

| Vertical bar indicates alternate selections — the bar means “or”

{ } Braces enclose content that the user must specify, such as x or y or z

... Ellipses indicate nonessential information omitted from the example


https://support.EMC.com

http://support.EMC.com

https://support.EMC.com/products

Preface

EMC Live Chat — Open a Chat or instant message session with anEMC Support Engineer.

eLicensing supportTo activate your entitlements and obtain your Symmetrix license files,visit the Service Center on https://support.EMC.com, as directed onyour License Authorization Code (LAC) letter e-mailed to you.

For help with missing or incorrect entitlements after activation (thatis, expected functionality remains unavailable because it is notlicensed), contact your EMC Account Representative or AuthorizedReseller.

For help with any errors applying license files through SolutionsEnabler, contact the EMC Customer Support Center.

If you are missing a LAC letter, or require further instructions onactivating your licenses through the Online Support site, contactEMC's worldwide Licensing team at [email protected] or call:

◆ North America, Latin America, APJK, Australia, New Zealand:SVC4EMC (800-782-4362) and follow the voice prompts.

◆ EMEA: +353 (0) 21 4879862 and follow the voice prompts.

We'd like to hear from you!Your suggestions will help us continue to improve the accuracy,organization, and overall quality of the user publications. Send youropinions of this document to:

[email protected]

Your feedback on our TechBooks is important to us! We want ourbooks to be as helpful and relevant as possible. Send us yourcomments, opinions, and thoughts on this or any other TechBook to:

[email protected]




10

Preface


1

This chapter provides the following information for the WANOptimization Controller (WOC) appliance:

◆ Overview............................................................................................. 12◆ Network topologies and implementations..................................... 13◆ Deployment topologies ..................................................................... 15◆ Storage and replication application................................................. 17

Network andDeployment

Topologies andImplementations

Network and Deployment Topologies and Implementations 11

12

Network and Deployment Topologies and Implementations

OverviewA WAN Optimization Controller (WOC) is an appliance that can beplaced In-line or Out-of-Path to reduce and optimize the data that isto be transmitted over the LAN/MAN/WAN. These devices aredesigned to help mitigate the effects of packet loss, networkcongestion, and latency while reducing the overall amount of data tobe transmitted over the network.

In general, the technologies utilized in accomplishing this areTransmission Control Protocol (TCP) acceleration,data-deduplication, and compression. Additionally, features such asQoS, Forward Error Correction (FEC), and Encryption may also beavailable.

Network links and WAN circuits can have high latency and/orpacket loss as well as limited capacity. WAN OptimizationControllers can be used to maximize the amount of data that can betransmitted over a link. In some cases, these appliances may be anecessity, depending on performance requirements.

WAN and data optimization can occur at varying layers of the OSIstack, whether it be at the network and transport layer, the session,presentation, and application layers, or just to the data (payload)itself.



Network topologies and implementationsTCP was developed as a local area network (LAN) protocol.However, with the advancement of the Internet it was expanded to beused over the WAN. Over time TCP has been enhanced, but evenwith these enhancements TCP is still not well-suited for WAN use formany applications.

The primary factors that directly impact TCP's ability to be optimizedover the WAN are latency, packet loss, and the amount of bandwidthto be utilized. It is these factors on which the layer 3/4 optimizationproducts focus. Many of these optimization products willre-encapsulate the packets into UDP or their proprietary protocol,while others may still use TCP, but optimize the connections betweena set of WAN Optimization Controllers at each end of the WAN.While some products create tunnels to perform their peer-to-peerconnection between appliances for the optimized data, others mayjust modify, or tag other aspects within the packet to ensure that thefar-end WOC captures the optimized traffic.

Optimization of the payload (data) within the packet focuses on thereduction of actual payload as it passes over the network through theuse of data compression and/or data de-duplication engines (DDEs).Compression is performed through the use of data compressionalgorithms, while DDE uses large data pattern tables and associatedpointers (fingerprints). Large amounts of memory and/or hard-drivestorage can be used to store these pattern tables and pointers.Identical tables are built in the optimization appliances on both sidesof the WAN, and as new traffic passes through the WOC patterns arematched, and only the associated pointers are sent over the network(versus resending data.) While typical LZ compression ratio is about2:1, DDE ratios can range greatly, depending on many factors. Ingeneral the combination of both of these technologies, DDE andcompression, will achieve around a 5:1 (and sometimes much higherratios) reduction level.

Layer 4/7 optimization is what is called the "application" layer ofoptimization. This area of optimization can take many approachesthat can vary widely, but are generally done through the use ofapplication-aware optimization engines. The actions taken by theseengines can result in benefits, including reductions in the number oftransactions that occur over the network or more efficient use ofbandwidth. It is also at this layer the TCP optimization occurs.

Network topologies and implementations 13

14


Overall, WAN optimizers can be aligned with customer networkingbest practices, and it should be made clear to the customer thatapplications using these devices can, and should, be prioritized basedon their WAN bandwidth/throughput requirements.



Deployment topologiesThere are two basic topologies for deployment:

◆ In-path/in-line/bridge

◆ Out-of-path/routed

An in-path/in-line/bridge deployment, as shown in Figure 1, meansthat the WOC is directly in the path between the source anddestination end points where all inbound and outbound flows willpass through the WAN Optimization Controllers. The placement ofthe WOC devices at each site is typically placed as close as possible tothe WAN circuit.

Figure 1 In-path/in-line/bridge topology

An out-of-path/routed deployment, as shown in Figure 2, means thatthe WOC is not in the direct path between the source and destinationend points. The traffic must be routed/redirected to the WOC devicesusing routing features such as WCCP, PBR, VRRP, etc.

Figure 2 Out-of-path/routed topology

Deployment topologies 15

16


◆ WCCPv2 (Web Cache Communication Protocol) is a contentrouting protocol that provides a mechanism to redirect traffic inreal-time. WCCP also has built-in mechanisms to support loadbalancing, fault tolerance, and scalability.

◆ PBR (Policy Based Routing) is a technique used to make routingdecisions based on policies or a combination of policies such aspacket size, protocol of the payload, source, destination, or othernetwork characteristics.

◆ VRRP (Virtual Router Redundancy Protocol) is a redundancyprotocol designed to increase the availability of a default gateway.

In the event of a power failure or WOC hardware or software failure,it is necessary for the WOC to provide some level of action. The WOCcan either continue to allow data to pass through, unoptimized, or itcan block all traffic from flowing through it. The failure modestypically offered by WAN optimizers are commonly referred to as:

◆ Fails-to-Wire

The appliance will behave as a crossover cable connecting theEthernet LAN switch directly to the WAN router and traffic willcontinue to flow uninterrupted and unoptimized.

◆ Fails-Open / Fails-to-Block

The appliance will behave as an open port to the WAN router.The WAN router will recognize that the link is down and willbegin forwarding traffic according to its routing tables.

Depending upon your deployment topology, you may determine thatone method may be better suited for your environment than theother.



Storage and replication applicationThis section provides storage and replication application details forEMC® products:

◆ Symmetrix®/VMAX™ SRDF®

◆ RecoverPoint

◆ SAN Copy™

◆ Celerra Replicator™

◆ MirrorView™

Configuration settingsConfigurations settings are as follows:

◆ Compression on GigE (RE) port = Disabled

Note: For Riverbed Steelhead RiOS v6.1.1a or later, the compressionsetting could be Enabled on the Symmetrix system. The Steelheadautomatically detects and disables compression on the Symmetrixsystem.

◆ SRDF Flow Control = Enabled

Network topologies and implementationsIn general, it has been observed that optimization ratios are higherwith SRDF/A than SRDF Adaptive Copy. There are many factors thatimpact how much optimization will occur, therefore results will vary.

NotesNote the following:

For Symmetrix configuration settings

Compression Compression should be disabled on the GigE ports on the MPCD andthe GigE director when a WAN optimization device employing datadeduplication is used. If compression is enabled on the GigE ports onthe MPCD and the GigE director, data deduplication benefits will beseverely impacted, resulting in increased WAN bandwidth needs.

Storage and replication application 17

18


SRDF Flow Control SRDF Flow Control should be enabled for increased stability of theSRDF links. Further tuning of SRDF flow control can be made toimprove performance. For more information, please contact yourEMC Customer Service representative.

For SRDF modes and data reductionIn general, it has been observed that optimization ratios are higherwith GigE ports on the MPCD and the GigE director as opposed toFCIP. There are many factors that impact how much optimization willoccur, therefore results will vary.

References◆ For further information, refer to the EMC Symmetrix Remote Data

Facility (SRDF) Connectivity Guide, located on the E-LabInteroperability Navigator at http://elabnavigator.EMC.com>PDFs and Guides.



2

This chapter provides FCIP configuration information for:

◆ Brocade FCIP ...................................................................................... 20◆ Cisco FCIP ........................................................................................... 22

FCIP Configurations

FCIP Configurations 19

20

FCIP Configurations

Brocade FCIPThis section provides configuration information for Brocade FCIP.

Configuration settingsConfiguration settings are as follows:

◆ FCIP Fastwrite = Enabled

◆ Compression = Disabled

◆ TCP Byte Streaming = Enabled

◆ Commit Rate = in Kbps (Environment dependent)

◆ Tape Pipelining = Disabled

◆ SACK = Enabled

◆ Min Retransmit Time = 100

◆ Keep-Alive Timeout = 10

◆ Max Re-Transmissions = 8

Brocade FCIP Tunnel settingsConsider the following:

◆ FCIP Fastwrite

This setting accelerates SCSI Write I/Os over the FCIP tunnel.This can not be combined with FC Fastwrites.

◆ Compression

This simply compresses the data that flows over the FCIP tunnel.This should be disabled when using with WOC devices, thusallowing the WOC device to perform the compression and datade-duplication.

◆ Commit Rate

This setting is environment dependent. This should be set inaccordance with the WAN Optimization vendor. Considerationssuch as Data-to-be-Optimized, Available WAN circuit size andData-Reduction ratio need to be taken into account.

◆ TCP Byte Streaming


FCIP Configurations

This is a Brocade feature which allows a Brocade FCIP switch tocommunicate with a 3rd party WAN Optimization Controller.This feature supports a FCIP frame which has been split into amaximum of 8 separate TCP segments. If the frame is split intomore than eight segments, it results in prematurely sending aframe to the FCIP layer with an incorrect size and the FCIP tunnelbounces.

Rules and restrictionsConsider the following rules and restrictions when using TCP bytestreaming:

◆ Only one FCIP tunnel is allowed to be configured for a GigE portthat has TCP Byte Streaming configured.

◆ FCIP tunnel cannot have compression enabled.

◆ FCIP tunnel cannot have FC Fastwrite enabled.

◆ FCIP tunnel must have a committed rate set.

◆ Both sides of the FCIP tunnel must be identically configured.

◆ TCP byte streaming is not compatible with older FOS revisions,which do not have the option available.

ReferencesFor further information, refer to https://support.emc.com andhttp://www.brocade.com.

◆ EMC Connectrix B Series Fabric OS Administrator's Guide

◆ Brocade Fabric OS Administrator’s Guide

Brocade FCIP 21

http://www.brocade.com/

http://www.brocade.com/

22

FCIP Configurations

Cisco FCIPThis section provides configuration information for Cisco FCIP.

Configuration settingsConfiguration settings are as follows:

◆ Max-Bandwidth = Environment dependent (Default = 1000 Kb)

◆ Min-Available-Bandwidth = Recommended setting: 50-80% ofMax-Bandwidth

◆ Estimated roundtrip time = Set to measured latency (round-triptime - RTT) between MDS switches

◆ IP Compression = Disabled

◆ FCIP Write Acceleration = Enabled

◆ Tape Accelerator = Disabled

◆ Encryption = Disabled

◆ Min Re-Transmit Timer = 200 ms

◆ Max Re-Transmissions = 8

◆ Keep-Alive = 60

◆ SACK = Enabled

◆ Timestamp = Disabled

◆ PMTU = Enabled

◆ CWM = Enabled

◆ CWM Burst Size = 50 KB

NotesConsider the following information for Cisco FCIP tunnel settings:

◆ Max-Bandwidth

The max-bandwidth-mbps parameter and the measured RTTtogether determine the maximum window size. This should beconfigured to match the worst-case bandwidth available on thephysical link.

◆ Min-Available-Bandwidth


FCIP Configurations

The min-available-bandwidth parameter and the measured RTTtogether determine the threshold below which TCP aggressivelymaintains a window size sufficient to transmit at minimumavailable bandwidth. It is recommend that you adjust this to50-80% of the Max-Bandwidth.

◆ Estimated Roundtrip-Time

This is the measured latency between the 2 MDS GigE interfaces.Ping can be used to determine the roundtrip-time.

◆ FCIP Write Acceleration

Write Acceleration is used to help alleviate the effects of networklatency. It can work with Port-Channels only when thePort-Channel is managed by Port-Channel protocol (PCP). FCIPwrite acceleration can be enabled for multiple FCIP tunnels if thetunnels are part of a dynamic Port-Channel configured withchannel mode active. FCIP write acceleration does not work ifmultiple non-Port -Channel ISLs exist with equal weight betweenthe initiator and the target port.

◆ Min Re-Transmit Timer

This is the amount of time that TCP waits before retransmitting.In environments where there may be high packet loss /congestion, this number may need to be adjusted to 4x themeasured roundtrip-time. Ping may be used to measure theround trip latency between the 2 MDS switches.

◆ Max Re-Transmissions

The maximum number of times that a packet is retransmittedbefore the TCP connection is closed.

Basic guidelinesConsider the following guidelines when creating/utilizing multipleFCIP interfaces /profiles:

◆ Gigabit Ethernet Interfaces support a single IP address.

◆ Every FCIP profile must be uniquely addressable by an IPaddress and TCP port pair. Where FCIP profiles share a GigabitEthernet interface, the FCIP profiles must use different TCP portnumbers.

Cisco FCIP 23

24

FCIP Configurations

◆ A FCIP interface is linked to a single FCIP profile. Up to threeFCIP interfaces can link to an FCIP profile, but only three FCIPinterfaces can be active on any Gigabit Ethernet interface.

◆ A dedicated FCIP profile per FCIP link is recommended.

Rules and restrictionsConsider the following rules and restrictions when enabling FCIPWrite Acceleration:

◆ It can work with Port-Channels only when the Port-Channel ismanaged by Port-Channel Protocol (PCP).

◆ FCIP write acceleration can be enabled for multiple FCIP tunnelsif the tunnels are part of a dynamic Port-Channel configured withchannel mode active.

◆ FCIP write acceleration does not work if multiplenon-Port-Channel ISLs exist with equal weight between theinitiator and the target port.

◆ Do not enable time stamp control on an FCIP interface with writeacceleration configured.

◆ Write acceleration can not be used across FSPF equal cost paths inFCIP deployments. Also, FCIP write acceleration can be used inPort-Channels configured with channel mode active orconstructed with Port-Channel Protocol (PCP).

ReferencesFor further information, refer to the following documentation onCisco's website at http://www.cisco.com.

◆ Wide Area Application Services Configuration Guide

◆ Replication Acceleration Deployment Guide

◆ Q&A for WAAS Replication Accelerator Mode

◆ MDS 9000 Family CLI Configuration Guide


http://www.cisco.com/

3

This chapter provides information on the following WANOptimization Controller (WOC) appliances, along with RiverbedGranite, which is used in conjunction with Steelhead:

◆ Silver Peak appliances ....................................................................... 26◆ Riverbed appliances........................................................................... 33

WAN OptimizationControllers

WAN Optimization Controllers 25

26

WAN Optimization Controllers

Silver Peak appliancesThis section provides information on the Silver Peak appliancesoptimization controller. The following topics are discussed:

◆ “Overview” on page 26

◆ “Terminology” on page 27

◆ “Features” on page 29

◆ “Deployment topologies” on page 30

◆ “Failure modes supported” on page 30

◆ “FCIP environment” on page 30

◆ “GigE environment” on page 31

◆ “References” on page 32

OverviewSilver Peak appliances are interconnected by tunnels, which transportoptimized traffic flows. Policies control how the appliance filtersLAN side packets into flows and whether:

◆ an individual flow is directed to a tunnel, shaped, and optimized;

◆ processed as shaped, pass-through (unoptimized) traffic;

◆ processed as unshaped, pass-through (unoptimized) traffic;

◆ continued to the next applicable Route Policy entry if a tunnelgoes down; or

◆ dropped.

The appliance manager has separate policies for routing,optimization, and QoS functions. These policies prescribe how theappliance handles the LAN packets it receives.

The optimization policy uses optimization techniques to improve theperformance of applications across the WAN. Optimization policyactions include network memory, payload compression, and TCPacceleration.

Silver Peak ensures network integrity by using QoS management,Forward Error Correction, and Packet Order Correction. WhenAdaptive Forward Error Correction (FEC) is enabled, the applianceintroduces a parity packet, which helps detect and correct



single-packet loss within a stream of packets, reducing the need forretransmissions. Silver Peak can dynamically adjust how often thisparity packet is introduced in response to changing link conditions.This can help maximize error correction while minimizing overhead.

To avoid retransmissions that occur when packets arrive out of order,Silver Peak appliances use Packet Order Correction (POC) toresequence packets on the far end of a WAN link, as needed.

TerminologyConsider the following terminology when using Silver Peakconfiguration settings:

◆ Coalescing ON — Enables/disables packet coalescing. Packetcoalescing transmits smaller packets in groups of larger packets,thereby increasing performance and helping to overcome theeffects of latency.

◆ Coalesce Wait — Timer (in milliseconds) used to determine theamount of time to wait before transmitting coalesced packets.

◆ Compression — Reduces the bandwidth consumed by traffictraversing the WAN. Payload compression is used in conjunctionwith network memory to provide compression on "first pass"data.

◆ Congestion Control — Techniques used by Silver Peak to managecongestion scenarios across a WAN. Configuration options arestandard, optimized, and auto. Standard uses standard TCPcongestion control. Optimized congestion control is the mostaggressive mode of congestion control and should only be used inenvironments with point-to-point connections for a dedicated tosingle application. Auto congestion control aims to improvethroughput over standard congestion control, but may not besuitable for all environments.

◆ FEC / FEC Ratio — Technique used by Silver Peak to recoverfrom packet loss without the need for packet retransmissions.Hence, loss is corrected on the Silver Peak appliance resulting inhigher throughout during the data transmission.

◆ IP Header Compression — Enables/disables compression of theIP header in order to reduce the packet size. Header compressioncan provide additional bandwidth gains by reducing packetheader information using specialized compression algorithms.

Silver Peak appliances 27

28


◆ Mode — Refers to the Silver Peak tunnel configuration. Thedefault setting is GRE. Alternative option is UDP.

◆ MTU (Maximum Transmission Unit) — The size, in bytes, of thelargest PDU that a given layer of a communications protocol canpass onwards.

◆ Network Memory — Silver Peak's implementation of real-timedata reduction of network traffic. This de-duplication technologyis used to inspect all inbound and outbound WAN traffic, storinga local instance of data on each appliance. The NX Seriesappliance compares real-time traffic streams with to patternsstored using Network Memory. If a match exists, a short referencepointer is sent to the remote Silver Peak appliance, instructing itto deliver the traffic pattern from its local instance. Repetitivedata is never sent across the WAN. If the content is modified, theSilver Peak appliance detects the change at the byte level andupdates the network's memory. Only the modifications are sentacross the WAN. These are combined with original content by NXSeries appliances at the destination location.

Currently, it is recommended to enable network memory and setthe network memory mode to 1. Mode 1 is referred to as "lowlatency mode" and enables network memory to better balancedata reduction versus high throughput. While network memorycan be enabled from the GUI, configuring it for mode 1 must beperformed through the CLI.

◆ Payload Compression — Uses algorithms to identify relativelyshort byte sequences that are repeated frequently over time.These sequences are then replaced with shorter segments of codeto reduce the size of transmitted data. Simple algorithms can findrepeated bytes within a single packet; more sophisticatedalgorithms can find duplication across packets and even acrossflows.

◆ Reorder Wait — Time (in milliseconds) that the Silver Peakappliances will wait to reorder packets. This is a dynamic valuethat will change based on line conditions. Recommendation is toleave this as the default for SRDF traffic.

◆ RTP Header Compression — Used to compress the size of theRTP protocol packet header used in Voice over IPcommunications. Header compression can provide additionalbandwidth gains by reducing packet header information usingspecialized compression algorithms.



◆ TCP Acceleration — References several techniques used by SilverPeak to accelerate the TCP protocol. TCP acceleration usestechniques such as selective acknowledgement, window scaling,and transaction size adjustment to compensate for poorperformance on high latency links.

◆ Tunnel Auto Max BW — Allows the Silver Peak to automaticallydetermine the maximum bandwidth available. Recommendationis to disable this in SRDF environments.

◆ Tunnel Max BW — For manually configuring the maximumbandwidth accessible to the Silver Peak. This is recommended inSRDF environments where bandwidth values are known. This isa static configuration.

◆ Tunnel Min BW — For manually configuring the maximumbandwidth accessible to the Silver Peak. This does not need to beset for proper operation. This is a static configuration. A value of32kbps is recommended, which is the default.

◆ WAN Bandwidth — Applies to the WAN side of the applianceand should be set to the amount of bandwidth to be madeavailable to the appliance on the WAN side. Inputting a valuealso configures the tunnel max bandwidth configuration variable.

◆ Windows Scaling — Used to overcome the effects of latency onsingle-flow throughput in a TCP network. The window-scalefactor multiplies the standard TCP window of 64 KB by 2 to thepower of the window-scale. Default window-scale is 6.

FeaturesFeatures include:

◆ Compression (payload and header)

◆ Network memory (data-deduplication)

◆ TCP acceleration

◆ QoS (Quality of Service)

◆ FEC (Forward Error Correction)

◆ POC (Packet Order Correction)

◆ Encryption - IPsec


30


Deployment topologiesDeployment topologies include:

◆ In-line (bridge mode)

• In-line

◆ Out-of-path (router)

• Out-of-path with Policy-Based-Routing (PBR) redirection

• Out-of-path with Web Cache Coordination Protocol(WCCPv2)

• Out-of-path with VRRP peering to WAN router

• Out-of-path with Policy-Based-Routing (PBR) and VRRPredundant Silver Peak appliances

• Out-of-path with Web Cache Coordination Protocol (WCCP)redundant Silver Peak appliances

◆ The Silver Peak appliances can only be deployed in out-of-path(Router) mode when using 10 Gb Ethernet Fibre data ports asoptical interfaces to do not fail to wire

◆ The Silver Peak NX-8700, NX-9700, and NX-10000 appliancessupport 10 Gb Ethernet Fibre data ports

◆ The SilverPeak VX (virtual appliances) and the Silver Peak VRX(virtual appliances) are supported when deployed on theVMWARE ESX or ESXi servers. The virtual appliances can onlybe deployed in out-of-path configurations.

Failure modes supportedThe following failure modes are supported:

• Fail-to-wire

• Fail-open

FCIP environmentThe following Silver Peak configuration settings are recommended inan FCIP environment:

◆ WAN Bandwidth = (Environment dependent)

◆ Tunnel Auto Max BW = Disabled (Unchecked)



◆ Tunnel Max BW = in Kb/s (Environment dependent)

◆ Tunnel Min BW = 32 Kb/s

◆ Reorder Wait = 100 ms

◆ MTU = 1500 (For 3.1 code and higher, maximum MTU = 2500)

◆ Mode = GRE

◆ Network Memory = Enabled

◆ Compression = Enabled

◆ TCP Acceleration = Enabled

◆ CIFS Acceleration = Disabled

◆ FEC = Enabled

◆ FEC Ratio = 1:5 (Recommended)

◆ Windows Scale Factor = 8

◆ Congestion Control = Optimized

◆ IP Header Compression = Enabled

◆ RTP Header Compression = Enabled

◆ Coalescing On = Yes

◆ Coalesce Wait = 0 ms

◆ From the CLI run: "system network-memory mode 1"

GigE environmentThe following Silver Peak configuration settings are recommended ina GigE environment:

◆ WAN Bandwidth = (Environment dependent)

◆ Tunnel Auto Max BW = Disabled (Unchecked)

◆ Tunnel Max BW = in Kbps (Environment dependent)

◆ Tunnel Min BW = 32 Kb/s

◆ Reorder Wait = 100 ms

◆ MTU = 1500

◆ Mode = GRE

◆ Network Memory = Enabled

◆ Compression = Enabled


32


◆ TCP Acceleration = Enabled

◆ CIFS Acceleration = Disabled

◆ FEC = Enabled

◆ FEC Ratio = 1:5 (Recommended)

◆ Windows Scale Factor = 8

◆ Congestion Control = Optimized

◆ IP Header Compression = Enabled

◆ RTP Header Compression = Enabled

◆ Coalescing On = Yes

◆ Coalesce Wait = 0 ms

◆ From the CLI run: "system network-memory mode 1"

ReferencesFor more information, refer to Silver Peak's website athttp://www.silver-peak.com.

◆ NX Series Appliance Operator Guide

◆ NX Series Appliance Network Deployment Guide

◆ Quick Start Guide, VX Virtual Appliance, VMware vSphere / vSphereHypervisor for configuring the VX virtual appliance

◆ Quick Start Guide, VRX-8 Virtual Appliance, VMware vSphere /vSphere Hypervisor, for configuring the VRX-8 virtual appliance

◆ VX Host System Requirements

◆ VRX-8 Host System Requirements


http://www.silver-peak.com/


Riverbed appliancesThis section provides information on the Riverbed Steelhead WANOptimization Controller and the Riverbed Granite system. Thefollowing topics are discussed:

◆ “Overview” on page 33

◆ “Terminology” on page 34

◆ “Notes” on page 38

◆ “Features” on page 39

◆ “Deployment topologies” on page 39

◆ “Failure modes supported” on page 39

◆ “FCIP environment” on page 40

◆ “GigE environment” on page 42

◆ “References” on page 44

OverviewRiOS is the software that powers the Riverbed's Steelhead WANOptimization Controller. The optimization techniques RiOS utilizesare:

◆ Data Streamlining◆ Transport Streamlining◆ Application Streamlining, and◆ Management Streamlining

RiOS uses a Riverbed proprietary algorithm called Scalable DataReferencing (SDR) along with data compression when optimizingdata across the WAN. SDR breaks up TCP data streams into uniquedata chunks that are stored in the hard disk (data store) of the devicerunning RiOS. Each data chunk is assigned a unique integer label(reference) before it is sent to a peer RiOS device across the WAN.When the same byte sequence is seen again in future transmissionsfrom clients or servers, the reference is sent across the WAN insteadof the raw data chunk. The peer RiOS device uses this reference tofind the original data chunk on its data store, and reconstruct theoriginal TCP data stream.

After a data pattern is stored on the disk of a Steelhead appliance, itcan be leveraged for transfers to any other Steelhead appliance across

Riverbed appliances 33

34


all applications being accelerated by Data Streamlining. DataStreamlining also includes optional QoS enforcement. QoSenforcement can be applied to both optimized and unoptimizedtraffic, both TCP and UDP.

Steelhead appliances also use a generic latency optimizationtechnique called Transport Streamlining. Transport Streamlining usesa set of standards and proprietary techniques to optimize TCP trafficbetween Steelhead appliances. These techniques ensure efficientretransmission methods, such as TCP selective acknowledgements,are used, optimal TCP window sizes are used to minimize the impactof latency on throughput to maximize throughput across WAN links.

Transport Streamlining ensures that there is always a one-to-one ratiofor active TCP connections between Steelhead appliances, and theTCP connections to clients and servers. That is, Steelhead appliancesdo not tunnel or perform multiplexing and de-multiplexing of dataacross connections. This is true regardless of the WAN visibility modein use.

TerminologyConsider the following terminology when using Riverbedconfiguration settings:

◆ Adaptive Compression — Detects LZ data compressionperformance for a connection dynamically and turns it off (setsthe compression level to 0) momentarily if it is not achievingoptimal results. Improves end-to-end throughput over the LANby maximizing the WAN throughput. By default, this setting isdisabled.

◆ Adaptive Data Streamlining Mode SDR-M — RiOS uses aRiverbed proprietary algorithm called Scalable Data Referencing(SDR). SDR breaks up TCP data streams into unique data chunksthat are stored in the hard disk (data store) of the device runningRiOS. Each data chunk is assigned a unique integer label(reference) before it is sent to a peer RiOS device across the WAN.When the same byte sequence is seen again in futuretransmissions from clients or servers, the reference is sent acrossthe WAN instead of the raw data chunk. The peer RiOS deviceuses this reference to find the original data chunk on its datastore, and reconstruct the original TCP data stream. SDR-Mperforms data reduction entirely in memory, which prevents theSteelhead appliance from reading and writing to and from the



disk. Enabling this option can yield high LAN-side throughputbecause it eliminates all disk latency. SDR-M is most efficientwhen used between two identical high-end Steelhead appliancemodels; for example, 6050 - 6050. When used between twodifferent Steelhead appliance models, the smaller model limitsthe performance.

IMPORTANT!You cannot use peer data store synchronization with SDR-M. Incode stream 5.0.x, this must be set from the CLI by running:"datastore anchor-select 1033" and then "restart clean."

◆ Compression Level — Specifies the relative trade-off of datacompression for LAN throughput speed. Generally, a lowernumber provides faster throughput and slightly less datareduction. Select a data store compression value of 1 (minimumcompression, uses less CPU) through 9 (maximum compression,uses more CPU) from the drop-down list. The default value is 1.Riverbed recommends setting the compression level to 1 inhigh-throughput environments such as data center to data centerreplication.

◆ Correct Addressing — Turns WAN visibility off. Correctaddressing uses Steelhead appliance IP addresses and portnumbers in the TCP/IP packet header fields for optimized trafficin both directions across the WAN. This is the default setting.Also see "WAN Visibility Mode" on page 38.

◆ Data Store Segment Replacement Policy — Specifies areplacement algorithm that replaces the least recently used datain the data store, which improves hit rates when the data in thedata store are not equally used. The default and recommendedsetting is Riverbed LRU.

◆ Guaranteed Bandwidth % — Specify the minimum amount ofbandwidth (as a percentage) to guarantee to a traffic class whenthere is bandwidth contention. All of the classes combined cannotexceed 100%. During contention for bandwidth the class isguaranteed the amount of bandwidth specified. The class receivesmore bandwidth if there is unused bandwidth remaining.

◆ In-Path Rule Type/Auto-Discover — Uses the auto-discoveryprocess to determine if a remote Steelhead appliance is able tooptimize the connection attempting to be created by this SYN


36


packet. By default, auto-discover is applied to all IP addressesand ports that are not secure, interactive, or default Riverbedports. Defining in-path rules modifies this default setting.

◆ Multi-Core Balancing — Enables multi-core balancing whichensures better distribution of workload across all CPUs, therebymaximizing throughput by keeping all CPUs busy. Corebalancing is useful when handling a small number ofhigh-throughput connections (approximately 25 or less). Bydefault, this setting is disabled. In the 5.0.x code stream, thisneeds to be performed from the CLI by running: "datastoretraffic-load rule scraddr all scrport 0 dstaddr all dstport "1748"encode "med".

◆ Neural Framing Mode — Neural framing enables the system toselect the optimal packet framing boundaries for SDR. Neuralframing creates a set of heuristics to intelligently determine theoptimal moment to flush TCP buffers. The system continuouslyevaluates these heuristics and uses the optimal heuristic tomaximize the amount of buffered data transmitted in each flush,while minimizing the amount of idle time that the data sits in thebuffer.

For different types of traffic, one algorithm might be better thanothers. The considerations include: latency added to theconnection, compression, and SDR performance.

You can specify the following neural framing settings:

• Never — Never use the Nagle algorithm. All the data isimmediately encoded without waiting for timers to fire orapplication buffers to fill past a specified threshold. Neuralheuristics are computed in this mode but are not used.

• Always — Always use the Nagle algorithm. All data is passedto the codec which attempts to coalesce consume calls (ifneeded) to achieve better fingerprinting. A timer (6 ms) backsup the codec and causes leftover data to be consumed. Neuralheuristics are computed in this mode but are not used.

• TCP Hints — This is the default setting which is based on theTCP hints. If data is received from a partial frame packet or apacket with the TCP PUSH flag set, the encoder encodes thedata instead of immediately coalescing it. Neural heuristicsare computed in this mode but are not used.



• Dynamic — Dynamically adjust the Nagle parameters. In thisoption, the system discerns the optimum algorithm for aparticular type of traffic and switches to the best algorithmbased on traffic characteristic changes.

◆ Optimization Policy — When configuring In-path Rules you havethe option of configuring the optimization policy. There aremultiple options that can be selected and it is recommended to setthis option to "Normal" for EMC replication protocols, such asSRDF/A. The configurable options are as follows:

• Normal — Perform LZ compression and SDR

• SDR-Only — Perform SDR; do not perform LZ compression

• Compression-Only — Perform LZ compression; do notperform SDR

• None — Do not perform SDR or LZ compression

◆ Queue - MXTCP — When creating QoS Classes you will need tospecify a queuing method. MXTCP has very different use casesthan the other queue parameters.

MXTCP also has secondary effects that you need to understandbefore configuring, including:

• When optimized traffic is mapped into a QoS class with theMXTCP queuing parameter, the TCP congestion controlmechanism for that traffic is altered on the Steelheadappliance. The normal TCP behavior of reducing theoutbound sending rate when detecting congestion or packetloss is disabled, and the outbound rate is made to match theminimum guaranteed bandwidth configured on the QoS class.

• You can use MXTCP to achieve high-throughput rates evenwhen the physical medium carrying the traffic has high lossrates. For example, MXTCP is commonly used for ensuringhigh throughput on satellite connections where alower-layer-loss recovery technique is not in use.

• Another usage of MXTCP is to achieve high throughput overhigh bandwidth, high-latency links, especially whenintermediate routers do not have properly tuned interfacebuffers. Improperly tuned router buffers cause TCP toperceive congestion in the network, resulting in unnecessarilydropped packets, even when the network can support highthroughput rates.


38


IMPORTANT!Use caution when specifying MXTCP. The outbound rate forthe optimized traffic in the configured QoS class immediatelyincreases to the specified bandwidth, and does not decrease inthe presence of network congestion. The Steelhead appliancealways tries to transmit traffic at the specified rate.

If no QoS mechanism (either parent classes on the Steelheadappliance, or another QoS mechanism in the WAN or WANinfrastructure) is in use to protect other traffic, that other trafficmight be impacted by MXTCP not backing off to fairly sharebandwidth. When MXTCP is configured as the queueparameter for a QoS class, the following parameters for thatclass are also affected:

Link share weight — The link share weight parameter has noeffect on a QoS class configured with MXTCP.

Upper limit —The upper limit parameter has no effect on aQoS class configured with MXTCP.

◆ Reset Existing Client Connections on Start-Up — Enables kickoff.If you enable kickoff, connections that exist when the Steelheadservice is started and restarted are disconnected. When theconnections are retried they are optimized. If kickoff is enabled,all connections that existed before the Steelhead appliance startedare reset.

◆ WAN Visibility Mode/CA — Enables WAN visibility, whichpertains to how packets traversing the WAN are addressed. RiOSv5.0 or later offers three types of WAN visibility modes: correctaddressing, port transparency, and full address transparency. Youconfigure WAN visibility on the client-side Steelhead appliance(where the connection is initiated). The server-side Steelheadappliance must also support WAN visibility (RiOS v5.0 or later).ALso see "Correct Addressing" on page 35.

NotesConsider the following when using Riverbed configuration settings:

◆ LAN Send and Receive Buffer Size should be configured to 2 MB



◆ WAN Send and Receive Buffer Size is environment dependentand should be configured with the result utilizing the followingformula:

WAN BW * RTT * 2 / 8 = xxxxxxx bytes

FeaturesFeatures include:

◆ SDR (Scalable Data Referencing)

◆ Compression

◆ QoS (Quality of Service)

◆ Data / Transport / Application / Management Streamlining

◆ Encryption - IPsec

Deployment topologiesDeployment topologies include:

◆ In-Path

• Physical In-Path

◆ Virtual In-Path

• WCCPv2 (Web Cache Coordination Protocol)

• PBR (Policy-Based-Routing)

◆ Out-of-Path

• Proxy

◆ Steelheads 7050 and 701 support 10 Gb Fibre data ports

◆ The virtual steelheads are supported when deployed onVMWARE ESX or ESXi servers. The virtual appliances can onlybe deployed in out-of-path configurations.

Failure modes supportedThe following failure modes are supported:

◆ Fail-to-wire

◆ Fail-to-block


40


FCIP environmentThe following Riverbed configuration settings are recommended in aFCIP environment:

◆ Configure > Networking > QoS Classification:

• QoS Classification and Enforcement = Enabled

• QoS Mode = Flat

• QoS Network Interface with WAN throughput = Enabled forappropriate WAN interface and set available WAN Bandwidth

• QoS Class Latency Priority = Real Time

• QoS Class Guaranteed Bandwidth % = Environmentdependent

• QoS Class Link Share Weight = Environment dependent

• QoS Class Upper Bandwidth % = Environment dependent

• Queue = MXTCP

• QoS Rule Protocol = All

• QoS Rule Traffic Type = Optimized

• DSCP = All

• VLAN = All

◆ Configure > Optimization > General Service Settings:

• In-Path Support = Enabled

• Reset Existing Client Connections on Start-Up = Enabled

• Enable In-Path Optimizations on Interface In-Path_X_X forappropriate In-Path interface

• In RiOS v5.5.3 CLI or later: “datastore codec multi-codecencoder max-ackqlen 30"

• In RiOS v6.0.1a or later: "datastore codec multi-codec encoderglobal-txn-max 128"

• In RiOS v6.0.1a or later: "datastore sdr-policy sdr-m"

• In RiOS v6.0.1a or later: " datastore codec multi-core-bal"

• In RiOS v6.0.1a or later: "datastore codec compression level 1"

◆ Configure > Optimization > In-Path Rules:

• Type = Auto Discovery

• Preoptimization Policy = None



• Optimization Policy = Normal

• Latency Optimization Policy = Normal

• Neural Framing Mode = Never

• WAN Visibility = Correct Addressing

• In RiOS v5.5.3 CLI or later for FCIP: “in-path always-probeenable”

• In RiOS v5.5.3 CLI or later for FCIP: “in-path always-probeport 3225”

• In RiOS v6.0.1a or later: "in-path always-probe port 0"

• In RiOS v6.0.1a or later: "tcp adv-win-scale -1"

• In RiOS v6.0.1a or later: "in-path kickoff-resume"

• In RiOS v6.0.1a or later: "protocol FCIP enable" for FCIP

• In RiOS v6.0.1a or later: "protocol srdf enable " for SymmetrixDMX and VMAX

Or, in RiOS v 6.1.1.a or later, you can use the GUI as follows:

– Configure > Optimization > FCIP

- FCIP Settings- Enable FCIP- FCIP Ports: 3225, 3226, 3227, 3228

• In RiOS v6.0.1a or later: "protocol fcip rule scr-ip 0.0.0.0 dst-ip0.0.0.0 dif enable" for EMC Symmetrix VMAX™

Or, in RiOS v 6.1.1.a or later, you can use the GUI as follows:

– Rules > Add a New Rule

- Enable DIF if R1 and R2 are VMAX and hosts are OpenSystems or IBM iSeries (AS/400)

- DIF Data Block Size: 512 bytes (Open Systems) and 520Bytes (IBM iSeries, AS/400)

- No DIF setting is required if mainframe hosts are in use

• In RiOS v6.0.1i or later: "sport splice-policy outer-rst-port port3226" for Brocade FCIP only

◆ Configure > Optimization > Performance:

• High Speed TCP = Enabled

• LAN Send Buffer Size = 2097152

• LAN Receive Buffer Size = 2097152


42


• WAN Default Send Buffer Size = 2*BDP (BW * RTT * 2 / 8 =xxxxxxx bytes)

Note: BDP = Bandwidth delay product.

• WAN Default Rcv Buffer Size = 2*BDP (BW * RTT * 2 / 8 =xxxxxxx bytes)

• Data Store Segment Replacement Policy = Riverbed LRU

• Adaptive Data Streamlining Modes = SDR-M

Note: Adaptive Data Streamlining Modes = SDR-Default for the7050/701 appliances.

• Compression Level = 1

• Adaptive Compression = Disabled

• Multi-Core Balancing = Enabled

Note: Multi-Core Balancing should be disabled if the number ofconnections through the steelheads is greater than the number ofcores on the Steelhead appliance.

GigE environmentThe following are Riverbed configuration settings recommended in aGigE environment:

In RiOS v6.1.1a or later, Steelheads will be able to automaticallydetect and disable the Symmetrix VMAX and DMX compression bydefault. Use show log from the Steelhead to verify that compressionon the VMAX/DMX has been disabled. The "Native Symmetrix REport compression detected: auto-disabling" message will display onlyon the Steellhead present on the Symmetrix local or remote sidewhich initiates the connection.

With Riverbed firmware v6.1.3a and above, the SRDF SelectiveOptimization feature is supported for SRDF group level optimizationfor end-to-end GigE environments with VMAX which have EMCEnginuity v5875 and later. Refer to the Riverbed Steelheaddeployment and CLI guide for further instructions.

◆ Configure > Networking > Outbound QoS (Advanced):

• QoS Classification and Enforcement = Enabled



• QoS Mode = Flat• QoS Network Interface with WAN throughput = Enabled for

appropriate WAN interfaces and set to available WANBandwidth

• QoS Class Latency Priority = Real Time• QoS Class Guaranteed Bandwidth % = Environment

dependent• QoS Class Link Share Weight = Environment dependent• QoS Class Upper Bandwidth % = Environment dependent• Queue = MXTCP• QoS Rule Protocol = All• QoS Rule Traffic Type = Optimized• DSCP = Reflect

◆ Configure > Optimization > General Service Settings:

• In-Path Support = Enabled• Reset Existing Client Connections on Start-Up = Enabled• Enable In-Path Optimizations on Interface In-Path_X_X• In RiOS v5.5.3 CLI and later: “datastore codec multi-codec

encoder max-ackqlen 30• In RiOS v6.0.1a CLI or later: "datastore codec multi-codec

encoder global-txn-max 128"

◆ Configure > Optimization > In-Path Rules:

• Type = Auto Discovery• Preoptimization Policy = None• Optimization Policy = Normal• Latency Optimization Policy = Normal• Cloud Acceleration = Auto• Neural Framing Mode = Never• WAN Visibility =Correct Addressing• In RiOS v5.5.3 CLI or later for GigE: “in-path always-probe

enable”• In RiOS v5.5.3 CLI or later for GigE: “in-path always-probe

port 1748”• In RiOS v5.0.5-DR CLI or later for GigE: “in-path asyn-srdf

always-probe enable”• In RiOS v6.0.1a or later: "in-path always-probe port 0"• In RiOS v6.0.1a or later: "tcp adv-win-scale -1"• In RiOS v6.0.1a or later: "in-path kickoff-resume"• In RiOS v6.0.1a or later: "protocol srdf enable " for Symmetrix

DMX and VMAXOr, in RiOS v 6.1.1.a or later, you can use the GUI as follows:– Configure > Optimization > SRDF


44


– SRDF Settings– Enable SRDF– SRDF Ports: 1748

• In RiOS v6.0.1a or later: "protocol srdf rule src-ip 0.0.0.0 dst-ip0.0.0.0 dif enable” for Symmetrix VMAXOr, in RiOS v6.1.1.a or later, you can use the GUI as follows:– Rules > Add a New Rule

– Enable DIF if R1 and R2 are VMAX and hosts are OpenSystems or IBM iSeries (AS/400)

– DIF Data Block Size: 512 bytes (Open Systems) and 520Bytes (IBM iSeries, AS/400)

◆ Configure > Optimization > Transport Settings:

• High Speed TCP = Enabled• LAN Send Buffer Size = 2097152• LAN Receive Buffer Size = 2097152• WAN Default Send Buffer Size = 2*BDP (BW * RTT * 2 / 8 =

xxxxxxx bytes)◆ Configure > Optimization > Performance

• WAN Default Rcv Buffer Size = 2*BDP (BW * RTT * 2 / 8 =xxxxxxx bytes)

• Data Store Segment Replacement Policy = Riverbed LRU• Adaptive Data Streamlining Modes = SDR-M

Note: Adaptive Data Streamlining Modes = SDR-Default for the7050/701 appliances.

• Compression Level = 1• Adaptive Compression = Disabled• Multi-Core Balancing = Enabled

Note: Multi-Core Balancing should be disabled if the number ofconnections through the steelheads is greater than the number ofcores on the Steelhead appliance.

ReferencesFor more information, refer to Silver Peak's website athttp://www.silver-peak.com.

◆ NX Series Appliance Operator Guide

◆ NX Series Appliance Network Deployment Guide


http://www.silver-peak.com/


◆ Quick Start Guide, VX Virtual Appliance, VMware vSphere / vSphereHypervisor for configuring the VX virtual appliance

◆ Quick Start Guide, VRX-8 Virtual Appliance, VMware vSphere /vSphere Hypervisor, for configuring the VRX-8 virtual appliance

◆ VX Host System Requirements

◆ VRX-8 Host System Requirements


46



Tech Book: WAN Optimization Controller Technologies

Technology

unique integer

forward error

valid support

failure modes

configuration

receive buffer

tcp byte streaming

emc field