After multiple tests, I can happily say with out any doubt in my mind that 2008 server blows 2003 server in terms of web performance out of the water… How do I come to this conclusion you ask?
If you have been reading up on my partner in crime’s Blog (Alan Lee) and Craig Baileys blog we have been performing some extensive bench testing on the servers. Alan within his blog describes the method that we have used to perform our testing, which in brief is a Linux command which does a simple WGET mirror of the home page this mirror copies all the contents locally and then removes it after the crawl is complete, within the mirror command it also copies all other associated links on the home page (there is roughly 13 links which means it copies another 13 pages and associated images).
One WGET, from the Elcom home page would receive a rough total of 6.4MB which with today’s bandwidth speeds and abundance of images and multimedia everywhere we look this is not a huge number, its what we would call realistic.
The test, the script was set to run 20 concurrent connections 1,000 times. Again 20 concurrent connections is very realistic, if anything it may not be enough, however we though give the size and the connections this was a good benchmark.
After running the tests through a out of the box 2003 Server running IIS 6.0 and SQL 2005 we posted a time of 3h:37min:59sec
then we ran the Server 2008 test again out of the box IIS7 and SQL 2005 and we posted a time of
0h:13min:55sec
JAW DROPPING isn’t it, yes we ran the exact same tests, we had them configured the same, both out of the box with IIS installed as an additional component and installed .NET 2.0 on both machines nothing different.
This worked out to be an amazing 17 times faster…
The test machine was the same the only difference was the drive, as one was loaded with 08 and the other with 03, however both of the drives were the same spec and same manufacture and even same model…
Machine Specs
Intel Core2 Dual core 2.12 GHz, 2GB RAM, 160GB Seagate SATA drive, Gigabit LAN
Testing (Linux) machine: Intel P4 3.0 GHz, 1GB RAM, 80GB Seagate SATA drive, Gigabit LAN
However during these tests we could see for 2003 Server the bottleneck appeared to be the CPU as it was using everything it could get, so we thought ok let try something bigger.
New Machine Specs
Intel Core Quad core 2.4 GHz, 4GB RAM, 160GB Seagate SATA drive, Gigabit LAN
Testing (Linux) machine: Intel P4 3.0 GHz, 1GB RAM, 80GB Seagate SATA drive, Gigabit LAN
Now that we had something bigger and better to test on, we thought ok lets compare something that is a little bit smaller in size, to speed the process up and also see what happens in terms of speeds.
So we deployed a Vanilla Community Manager site, which reduced the WGET to a small, manageable 70kb size. The results are still astonishing:
Win2k3
Start: 17:24:52
Finish: 18:15:53
Total time: 0:51min:01Sec
Win2k8
Start: 12:09:51
Finish: 12:22:23
Total time: 0:12min:3Sec
2008 is still 4 times faster with something tiny and very manageable. There were no bottlenecks here are all, the CPU for 2003 was not varying between 70 and 80%. The memory was very, very much under utilized, it seemed to only use a lousy 350MB, when it had gigs to use.
Where as Server 2008, made much better use of the memory and left more CPU, See below for the stats yourself.
This shows that the SQLSVR is taking little resources, there is obviously not much for it to do there, are really no transactions of it to think about.
The W3WP exe as you can see is using alot more resources, rightful so, but as you can see its using alot of memory, which isn’t a bad thing, considering it has so much (memory is cheaper then processing power)
Here is how the network card was reacting to the masses of connections.
And here is the overview of all performance, as you can see the disk was getting a work out, but dosnt seem to be a bottle neck either.
To conclude… change all your servers to 2008, I know I cant wait to start implementing a few more 2008 Servers across our network at work.
)
