VDI Architecture Chris House, VCP Senior Network Analyst DataCenter Team
Dec 23, 2015
VDI Architecture
Chris House, VCP
Senior Network Analyst
DataCenter Team
Quick facts
• 1,500 Windows XP SP2 “sessions”• 1 or more statically assigned to users• 2 datacenters, 750 sessions per datacenter• 2 VirtualCenter servers (1 per datacenter)• 38 ESX servers (19 per datacenter/VC)• 4 ESX clusters (2 per datacenter/VC)• 15 datastores per cluster/server• 25 sessions per datastore• 2 HP StorageWorks XP1024 Fibre-Channel SAN
Arrays (1 per datacenter)
Concept• Concurrent VDI and blade PC deployment• Complete desktop replacement for office and clinical staff• Reduce PC maintenance costs/time• Provide fastest possible access for roaming users to desktop and
apps• All sessions run 24x7• Users encouraged to leave apps running• Sessions statically assigned (no pooling)• Support multiple sessions per user (software testing)• Wyse dual-monitor & HP thin terminals, 19-inch LCD monitors,
biometric keyboards, Windows XP Embedded• Manually failover sessions to remote datacenter if disaster
Sessions
• Windows XP SP2• 10GB C: drive• 512MB RAM (pagefile 768-
1536MB), 1 vCPU, 1 NIC• Sessions never
suspended/shut down– Will reboot every 2 weeks
once idle for 8 hours
• Reimaged as needed using Altiris (PXE boot)
• “My Documents” redirected to network drive
• Sessions not backed up
Session Deployment
• Sessions deployed once using ESX service console shell script (bash)
• VM name & hostname same as custom MAC address– 00:50:56:0A for VDC sessions– 00:50:56:0D for BDC sessions
• Deployment process– Create directory, customize new VMX– vmkfstools clone sysprep-ready VMDK– register with server and power on for sysprep
Location-based computing
• Sessions aware of user location within the enterprise
• Printing & applications require user location to function properly
• Specific printers dynamically installed at RDP login depending on user’s location– “Roaming Printer Agent”
• Custom solution developed in-house
– Microsoft Terminal Services API used to obtain thin terminal hostname, correlate to a location
Application deployment
• Initially considered SVS/Appstream
• Currently use Novell ZENWorks, Altiris
• Application objects assigned to users in eDirectory
• Icons for ZEN applications presented on desktop
• Application installed at first launch
• Moving to ZEN 10 for Active Directory
• Custom broker developed in-house
• Brokers RDP connections to VDI and other endpoints:– VDI– CCI (HP blade PCs)– Desktop PCs
• Assignments made in Active Directory
• Based on “Remote Operator” property of computer object in AD
• Supports multiple endpoints per user
Broker:
Clusters & Servers• 38 x ESX 3.0.1 Enterprise• 2 x VirtualCenter 2.0.2• 40 sessions per server (DRS
balances load)• 4 clusters, 9-10 servers per cluster• HA & DRS enabled• 375 sessions per cluster
• HP BL480c G1 c-Class blades• Dual Intel Xeon Quad-Core
processors @ 2.666GHz• 24 GB memory• 4 x 1000MB NICs• Dual-port 4GB FC HBA
Network
• 4 x 1000MB NICs• c-Class enclosures with two Cisco switches• ESX vSwitch vmnics uplink to ports on both
enclosure switches (two nics per vSwitch)• Cisco switches have 2 x 1000MB Etherchannel
uplinks, link state tracking enabled• vSwitch0
– Service Console (VLAN for servers)– VMKernel (VLAN for VMotion)
• vSwtich1– Multiple Virtual Machine portgroups (separate VLANs)
Storage
• Dual-port 4GB Qlogic FC HBA• c-Class enclosures with two 4/12 Brocade 4GB
switches, single uplinks to core switches• HBA connects to each switch• Metro has dual-fabric FC ‘star’ layout (core &
edge switches, 32 total)• 15 datastores per cluster• 25 sessions per datastore• 263GB datastore size• 9-10 hosts per datastore
Storage
• 2 x HP StorageWorks XP1024 arrays• 7.89TB dedicated to local datacenter VM disks• 7.89TB dedicated on remote array for synchronous
continuous replication of local VM disks• 15.78TB/array, 31.5TB total• Dedicated array ports for VDI ESX (4 per fabric, 8 total)• Datastores balanced across all paths (8)• 18 RAID5(7D+1P) disk groups per array
– Local VDI LDEVs– Replicated VDI LDEVs from remote array
• 1 LUN per datastore (no extents)• Datastore LUNs are XP LUSEs of 17 LDEVs (from
different disk groups - 136 spindles per LUN)
Achievements• Staff appreciate lightning-fast access to applications and improved mobility• Performance on par with and sometimes faster than PCs/BladePCs (server-
class hardware)• Decreased time to access information – improves patient safety• Very convenient• Information secure in Datacenter• Reduced PC support calls• Reduced electricity usage through fewer PCs/CRTs• Streamlined infrastructure• Safe from ESX server failure with HA• Safe from runaway CPU/memory hogging apps in sessions affecting
servers with DRS load balancing• Service desk has more tools to manage sessions (VM reset and View
console through VC WebAccess)• Able to perform server maintenance during the day by utilizing VMotion to
shuffle sessions around
Issues
• Multimedia limits of RDP– streaming video/Flash– high-res images/animation– Currently using Wyse multimedia redirector, supports
only few codecs– USB support
• VDI performance suffers when all sessions perform high disk I/O at same time– Windows Updates – disabled, failing retries– Antivirus – randomized across week– Large software deployment/updates – Epic– Re-imaging
VMware vClient / VDM3
• Metro invited to private beta at Palo Alto HQ• Several features of VDM could be beneficial for
our environment• Herculean effort to move existing environment to
VDM– rip & replace would be quickest (versus migration)– What to do with 1,500 displaced users?
• Conflicts exist between current session management processes/configuration and VDM’s methods = retraining
Future
• Software vendors working on multimedia improvements for RDP
• Considering switching to RGS or another protocol
• Introducing session pooling• Eliminating array replication or switching to
asynchronous• Finding cheaper storage that performs• Continuing to evaluate VDM
Questions?
http://communities.vmware.com/people/chouse