performance different of a windows server (2003) and a linux server

[ozyozy]

on the same ESX box, we have a number of VMs running, which has both the Linux Redhead and Windows OS servers.

we are not abusing the Windows OS servers by any mean, nor we are not giving it enough memory and CPU. The Windows and the Lunix VMs are sitting on the ESX's local disks.

the issue: a lot of the Windows VMs are acting very slow while the Linux are running fine. Each time we RDP to a Windows VM, it was so slow to just even move the mouse around or to open up to browse some folders. And, trying to execute any program or scripts would just take minutes.

we have set the hardware acceleration to the max on those Windows VMs.

had anyone ever came across situation like this?

what are some reasons that make the Windows servers operate slower compare to Linux servers?

what makes the Linux to outperform the Windows?

[Irithori]

do the windows boxes run in paravirtualization or full virtualization?

[ozyozy]

do the windows boxes run in paravirtualization or full virtualization?

i am sorry, i am not sure what is full virtualization.

however, there is only one box, and that is the ESX box running VMware ESXi 4.0, which host all those VMs.
this ESX box supports VMI and VMI is enabled. All the Guest Linux are also have VMI enabled too.

the Windows OS does not have the VMI.

does that sort of answer your question?

[Irithori]

no.

Full virtualization means, the VM host has to provide virtual hardware, that can be accessed with the regular drivers of the guest

In paravirtualization, there are special harddisc, video and network drivers, that take "shortcuts" (yeah, I know this a simplification)

That is why there are vmwaretools.
After installing these in the guests, performance will be much better.

[Irithori]

I will have to verify, but I *think* in linux you do not need special paravirtualization aware kernel modules (drivers) for IO and networkcards, as they are aware by default.
Only video and usb need the vmwaretools in linux.

[ozyozy]

no.

Full virtualization means, the VM host has to provide virtual hardware, that can be accessed with the regular drivers of the guest

In paravirtualization, there are special harddisc, video and network drivers, that take "shortcuts" (yeah, I know this a simplification)

That is why there are vmwaretools.
After installing these in the guests, performance will be much better.

I just read this article, it talks about this a bit which I found kinda interesting vmware.com/files/pdf/VMware_paravirtualization.pdf

For the vmwaretools, yes we do have the latest tools installed on the ESX VMware box.

[Irithori]

ok, if vmware tools are installed, then the next step might be to get some numbers.

do this in linux:
dd if=/dev/zero of=/opt/deleteme.dd bs=1M
and watch
vmstat 1
This gives you the sequential write speed.

For windows, there are also some free IO tests.

There should be similar results in sequential write speed.

Then use netio for network throughput.
There is a windows version of it.

[Irithori]

ok, netio for linux is hard to find these days.

Let´s ducttape one:
host 1: netcat -l 1234 > /dev/null
(probably the command is called nc)

host2: cat /dev/zero | netcat <IP_host 1> 1234

host(1|2): iptraf, detailed view

[ozyozy]

Hi Irithori,

I have used Perfmon to monitor a few Windows VMs and I found the following:

1. The physical disks response (both read and write) are high (or slow) on those VMs as they almost always pass the 25 ms.

- but, having said that, if the physical disks are slow, this also means that it is slow too on the Linux VMs. But, that doesn't cause the Linux performance to decrease. So, this isn't likely be the problem.

2. The logical disk response (both read and write) are showing no problem.

3. The network bottleneck doesn't seems to be a problem. We have two GB nic, and are both active.

4. Processor and memory status are OK.

PS.

On your previous post,
Prior to implementing those commands, I would like to know what those command do first. Honestly, I don't have access to the Linux environment, as I would have to ask the Linux support engineers to complete those command for me. So, I would like to know what they do and would the results suggest anything, so that I can sort of let them know.

Thanks.

ok, netio for linux is hard to find these days.

Let´s ducttape one:
host 1: netcat -l 1234 > /dev/null
(probably the command is called nc)

host2: cat /dev/zero | netcat <IP_host 1> 1234

host(1|2): iptraf, detailed view

ok, if vmware tools are installed, then the next step might be to get some numbers.

do this in linux:
dd if=/dev/zero of=/opt/deleteme.dd bs=1M
and watch
vmstat 1
This gives you the sequential write speed.

For windows, there are also some free IO tests.

There should be similar results in sequential write speed.

Then use netio for network throughput.
There is a windows version of it.