Update on 2-Socket and 4-Socket Servers for Virtualization


In a previous white paper, a comparison between Dell 2-socket and HP 4-socket servers running a series of workloads on ESX server was made. At the time of testing, the two systems used were the most powerful available configurations available for the Dell™ PowerEdge™ 2950 and HP Proliant DL585. This means the comparison was between two 8-core servers: A two socket quad--core server and a four socket dual-core server. The conclusion of that paper was that the 2 socket quad-core server was better for virtualization based on performance, price/performance, and performance per watt.

The PowerEdge R900 is a four-socket quad-core server with 16 cores total. This puts it more on par with the Dell 2950 two-socket server in terms of processing power per socket as both the 2950 and R900 are using quad-core processors. To determine how this current generation four-socket server compares with the Dell two-socket server in the previous test, an R900 with four Intel® Xeon® X7350 processors and 128GB of ram was run through the same tests in June 2008. The test environment from the previous test still existed so the test was reproduced in every way. The same VMs, the same storage, and the same DVD store test application were used. The table below summarizes the configuration of the R900 used in the new test and the 2950 from the previous tests.


PowerEdge 2950 PowerEdge R900
Virtualization software VMware ESX Server 3.0.1 VMware ESX Server 3.5
Processor Two quad-core Intel Xeon X5355 processors at 2.66 GHz with 8MB cache (shared) Four quad-core Intel Xeon X7350 processors at 2.93 GHz with 8MB cache
Frontside bus 1,333 MHz 1,066 MHz
Memory 16 GB (8x 667 MHz fully buffered 2GB DIMMs) 128 GB (32x 667Mhz fully buffered 4GB DIMMS)
Internal disks Two Serial Attached SCSI (SAS) 146 GB, 15,000 rpm drives Two Serial Attached SCSI (SAS) 73 GB, 15,000 rpm drives
Network interface card (NIC) Two 10/100/1,000 Mbps internal NICs Two 10/100/1,000 Mbps internal NICs
Disk controller PERC 5/i PERC 6/i

Please see the previous paper for configuration details on the storage, VMs, and DVD Store test application, which were kept the same for this test.

Testing

Only the SQL Server and SuSE LAMP tests were repeated for this update. As this is only a quick update and not a complete performance study, it was determined that two data points would be enough.

In order to have a common point for comparison, the number of VMs was held constant with respect to the earlier paper. Specifically 32 SQL Server VMs and 44 MySQL VMs were used for testing on the R900 to match the number of VMs that originally drove the 2950 to approximately 85% utilization. The workload on each VM was then increased to drive CPU utilization close to 85% on the R900. This was done by increasing the number of DVD Store user threads for each VM. For the SQL Server VMs and the MySQL VMs, the number of threads was tripled in order to drive the server to approximately 85% CPU utilization.


Results

The performance of the R900 is much better than the performance of the 2950 tested in the earlier paper. This is expected because it’s a much newer server and has twice as many cores and 4 times as much memory. Not surprisingly, given that there are 2x as many processors, power consumption is also much higher. The tradeoff results in a performance per watt number for the SQL Server and MySQL tests that are 10% and 23% better respectively for the R900, although the raw performance of the R900 is well over 2X that of the PowerEdge 2950. The complete results are in the tables below.


SQL Server VMs Test PowerEdge 2950 with Intel Xeon X5355 PowerEdge R900 with Intel Xeon X7350
VMs
32
32
User Threads Per VM
4
12
% CPU Utilization
85
86
Orders Per Minute
29346
72386
Avg Watts
449
1007
Performance Per Watt
65
72
 R900 is 10 % better


SLES LAMP VMs Test PowerEdge 2950 with Intel Xeon X5355 PowerEdge R900 with Intel Xeon X7350
VMs
44
44
User Threads Per VM
1
3
% CPU Utilization
86
88
Orders Per Minute
9852
27582
Avg Watts
447
1019
Performance Per Watt
22
27
 R900 is 23 % better


Conclusions

An increase in performance per watt in the range of 10 to 20 percent shows that the new PowerEdge R900 is not only more powerful than the previous generation Dell PowerEdge 2950, but is also more efficient.


todd_muirhead
todd_muirhead 		Latest page update: made by todd_muirhead , Jul 8 2008, 12:41 PM EDT (about this update About This Update todd_muirhead Edited by todd_muirhead


view changes
- complete history)
Keyword tags: 2-socket 4-socket dl585 esx performance price/performance virtualization
More Info: links to this page

Threads

Anonymous  (Get credit for your thread)

By saving these changes, I acceptthis site's content policies and agree to comply with applicable copyright laws.

Started By Thread Subject Replies Last Post
todd_muirhead Performance Improvement Due to ESX 3.5? 6 Jul 29 2008, 12:20 AM EDT by todd_muirhead
todd_muirhead
Thread started: Jul 14 2008, 2:14 PM EDT  Watch
I recently got an email from asking about the testing results published on this page. The question was how much of the improvement in peformance could be linked to the upgrade from ESX 3.0.1 to ESX 3.5?

The answer is that I didn't do the R900 test with ESX 3.0.1, so it is not possible to say how much of the performance advantage that the R900 has is due to ESX 3.5. I can say that with the upgrade from ESX 2.5 to ESX 3.0.1 I did some testing with the same workload on the same server and found that performance was about the same. I would guess that most of the performance advantage is due to the R900 based on this previous experience.

Todd
Do you find this valuable?    
Keyword tags: 2-socket 4-socket dl585 esx performance price/performance virtualization
Show Last Reply