Thursday, January 10, 2008

Benchmarking Windows 2003 Server in Amazon EC2

Abstract: Some notes on trying to run Windows 2003 Server EE on Amazon EC2 extralarge type of instances (4 way).

I tried to use Qemu's -smp n feature on Amazon Elastic Compute Cloud, but it still uses only one processor. It is possible to set affinity and run each Qemu instance on individual processor, but this is very nontypical scenario. Also this article explains Qemu's "rdtsc" usage on SMP hosts

Moving system disk to RAM decreased disk-related IO by 10-15%, but CPU-bound bottleneck still slowed system down.

Windows Device Manager showed that our installation uses MP kernel according to "ACPI Multiprocessor PC", ACPI APIC MP HAL (Halmacpi.dll).

I'll keep you informed on my other findings.


Anonymous said...


You seem to be some authority on installing windows in a virtual space such as AWS.. can you tell me , in absence of support from AWS, are there any other "cloud player" who do support Windows?

Thanks in advance

Unknown said...

is there any specific application, that you want to run ?

Anonymous said...

Hi, you seem to know your stuff, so I'm reaching out with this quick Q.

I'm running a website that's currently handling about 1.2 million people a month - it's a forum running on .NET, SQL 2005..

The server is under heavy heavy load, i'm optimising constantly, but my current machine a regular 8 core Xeon (2 yrs old) wont be up to the task for much longer..

Is it possible to migrate to EC2, is it a sensible move yet? Is it fastttttt?

I'm hesitant to buy another server, if this is a reality now..

Unknown said...

Hello me,
Amazon supports Windows type of instances now and you can try their m type of instances , if your load is CPU bound. This way you can spread load across 20 cores (virtual, but perhaps still better, than 8, given virtualization takes 15% off). Also it depends on scalability of your forum code, maybe it should be clustered.

Khazret Sapenov

Anonymous said...

Thanks mate,
Much appreciated..

Cloud Computing Google Group