Consolidation of a Performance Sensitive Application: Virtualizing Electronic Sports League's Gaming Infrastructure

ManualsBrandsIntel ManualsOtherIntel 10 Gigabit XF SR Server Adapter

Consolidation of a Performance-

Sensitive Application: Virtualizing

Electronic Sports League’s

Gaming Infrastructure

White Paper

Intel® Xeon® Processor

7400-based Server

Abstract

An end-user case study with Electronic Sports League (ESL) using VMware ESX 3.5U1*

with NetQueue feature running on Intel® Xeon® processor 7400 series-based servers with

Intel® NICs supporting VMDq feature. Game servers are mission-critical, single-threaded,

processor-intensive, and network-latency sensitive. They were previously believed to be

“non-virtualizeable” due to the overhead of virtualization. In this case, we will show that

game QoS can be preserved on virtualized game servers when using the latest technologies

noted above. We’ll also show that large server consolidation ratios and cost savings can be

achieved at the same time.

Authors: Bob Albers, I/O Usage Architect, and Sreeram Sammeta,

Sr. Systems Engineer, Intel Digital Enterprise Group

Summary of content (16 pages)

PAGE 1
White Paper Intel® Xeon® Processor 7400-based Server Consolidation of a PerformanceSensitive Application: Virtualizing Electronic Sports League’s Gaming Infrastructure Abstract An end-user case study with Electronic Sports League (ESL) using VMware ESX 3.5U1* with NetQueue feature running on Intel® Xeon® processor 7400 series-based servers with Intel® NICs supporting VMDq feature. Game servers are mission-critical, single-threaded, processor-intensive, and network-latency sensitive.
PAGE 2
White Paper Consolidation of a Performance-Sensitive Application Table of Contents Introduction – Can We Virtualize Everything?. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 “Non-virtualizable” applications. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
PAGE 3
White Paper Introduction — Can We Virtualize Everything? Virtualization of enterprise data center applications using hypervisors or VMMs is taking a predictable path. It started with the Consolidation of a Performance-Sensitive Application ESL Electronic Sports League (ESL) is the largest online gaming community in Europe, with more than 844,000 active users as of August 12, 2008.
PAGE 4
White Paper Consolidation of a Performance-Sensitive Application Problem statement ESL’s perception, along with that of many of its peers in the gaming industry, was that gaming servers “can’t be virtualized” 4 sockets with 6 cores Intel Xeon 4x 1066 Intel Xeon 7400 Series Intel Xeon Intel Xeon due to their mission-critical nature, high CPU utilization, and the critical focus of their user base on the in-game transaction latency.
PAGE 5
White Paper VM1 VM2 VMn Virtual Virtual Virtual Consolidation of a Performance-Sensitive Application 10.0 Virtualization Hypervisor 8.0 Throughput NIC Unused I/O capacity 6.0 4.0 4.0 2.0 without VMDq LAN Result: NIC performance can be up to ~60% underutilized Figure 3. Network data flow for virtualization without the use of VMDq and NetQueue technologies. Figure 4. Impact of virtualization on a 10 GB Ethernet NIC without the use of VMDq and NetQueue.
PAGE 6
White Paper Consolidation of a Performance-Sensitive Application VM1 VM2 VMn Virtual Virtual Virtual VMware with NetQueue 10.0 9.2 8.0 Throughput (GB) NIC with VMDq 9.5 6.0 4.0 Without VMDq With VMDq 4.0 With VMDq Jumbo Frames 2.0 LAN • 2x throughput • Near Native 10 GbE Figure 5. Network data flow for virtualization with VMDq and NetQueue. Figure 6. Tests measure wire speed Receive (Rx) side performance with VMDq on Intel® 82598 10 Gigabit Ethernet Controllers.
PAGE 7
White Paper Consolidation of a Performance-Sensitive Application CounterStrike 1.6* CounterStrike 1.6* Windows Server 2003* 32 bit Windows Server 2003* 32 bit VM1 VMn Average latency (rms) 25 20 15 10 5 VMware ESX 3.5 U1* Intel® Xeon® Processor 7400 Series-based Server Best LAN Best Internet Typical Internet Figure 7. Software stack. Figure 8. In-game latency. Software stack Question the Assumptions Figure 7 shows the software stack used in the PoC. We started with VMware ESX 3.
PAGE 8
White Paper Consolidation of a Performance-Sensitive Application Client 1 1 GbE 40–200 bytes UDP Game Player Force 10 S50* Game Server 10 GbE 2,000 bytes UDP Client 8 Native Server 1 GbE Figure 9. Game network protocol overview. Figure 10. Native lab test setup. Gaming protocol Round-trip network latency tests How does this gaming protocol work? A player sends a UDP In order to run the round-trip network latency test, we used the packet of 40–200 bytes in size to the game server.
PAGE 9
White Paper Consolidation of a Performance-Sensitive Application Client 1 ESX* Server 1 GbE VM1 Force 10 S50* Client 8 Average latency (rms) 0.30 0.25 0.20 0.15 0.10 0.05 10 GbE vSW1 1 GbE 64 VM8 256 Packet size (bytes) Native VMDq 1024 No VMDq Figure 11. Virtualized lab test setup. Figure 12. Netperf 2.4.4 UDP latency test with eight parallel streams. Scenarios 2 and 3: Virtualized with and without VMDq/NetQueue The results from these tests are summarized in Figure 12.
PAGE 10
White Paper Consolidation of a Performance-Sensitive Application VM1 VMn vSwitch 1 Svc Cons vSwitch 0 10 GbE VMDq 1 GbE Force 10 S50N ESL Network Firewall Internet CounterStrike 1.6* CounterStrike 1.6* Windows Server 2003* 32 bit Windows Server 2003* 32 bit VM1 VMn VMware ESX 3.5 U1* Intel® Xeon® Processor 7400 Series-based Server Figure 13. Live Internet test setup. Figure 14. Test software stack.
PAGE 11
Consolidation of a Performance-Sensitive Application 100 740 80 720 700 60 680 40 Power (Watts) Average CPU Utilization (percentage) White Paper 660 20 640 24 32 36 40 Number of VMs Figure 15. ESL virtual game servers on Intel® Xeon® processor 7400 Series-based platforms. ROI Delivered latency or gaming experience. But in ESL’s current native envi- Server consolidation 60–80 percent.
PAGE 12
White Paper Consolidation of a Performance-Sensitive Application Table 1. ESL standard native hardware deployment metrics (“Before”) vs. virtualized deployment Proof-of-Concept metrics as described in this paper (“After”). Before After Server 1P Intel® Core™2 Duo processor 4P Intel® Xeon® processor 7400 Cores 2 24 Game server processes 3 per CPU core 3 per VM; 4.
PAGE 13
White Paper Consolidation of a Performance-Sensitive Application Conclusions We have shown that ESL’s mission-critical, processor-intensive, network latency-sensitive game servers can be very effectively virtualized using the latest technologies. We achieved 18:1 consolidation ratios and saved almost 90 percent on power “ The new Six-Core Intel Xeon 7400 processor series (“Dunnington”) was completely overwhelming in all terms.
PAGE 14
White Paper Consolidation of a Performance-Sensitive Application Resources Additional sources of information on this topic Intel Virtualization Developer Community: www.microsoft.com/windowsserver/compare/compare-windows-to-unix.mspx Intel Virtualization Developer Community: softwarecommunity.intel.com/articles/eng/1424.htm Intel Virtualization Technology: www.intel.com/technology/virtualization/index.htm Intel Xeon processor 7400 series: www.intel.com/performance/server/xeon_mp/virtualization.
PAGE 15
PAGE 16
We want to hear from you Join the conversation on Intel’s Open Port: The Server Room • Bring your questions to the “Ask an Expert” discussion forum • Share best practices with IT influencers • Engage with Intel on your Datacenter challenges Performance tests and ratings are measured using specific computer systems and/or components and reflect the approximate performance of Intel products as measured by those tests.