1

As I understand it Windows Server 2008 R2 has the ability to send IO requests to the storage subsystem of up to 1MB in size. Back in the Server 2003 era (and before) this used to be limited to 64KB.

I seem to remember reading that it was possible to manipulate the maximum IO request size away from the default value. I think that a registry key under HKLM\SYSTEM\CurrentControlSet\Control\Session Manager is required but Google isn't finding any answers today.

I want to change the IO requests size to asses the impact on a VMWare FT virtual machine when it is hosting a specialist video recording application that is doing a lot of IO to virtual disks. The root problem we are trying to address is one of VMWare FT Primary/Secondary replication convergence when experiencing a mix of large read and write requests. The hope is that by varying the IO request size we may be able to increase the efficiency of this.

1 Answer 1

1

So i assume your running this server in ESXI? Its not clear completely from your question.

To your question. You wont get any benefit from doing this. In fact you might decrease performance. (if i remember correct smss.exe does not do what your describing). To get better I/O performance with Windows Server 2008 r2 you need to have MPIO (Multipath I/O) directly to storage volume (would keep the OS on VMDK and have data on iscsi drive).

This assumes you have a storage solution that support MPIO.(you will have to setup ISCSI so you have direct access to the volume on your SAN without going though a VMDK).

This isnt a 5 minute setup, and requires some home-work to get right. Since i dont know anything about your environment i cant advice anything else.

I know some people in ESXI 4.x have had to lower there I/O request on Vmware hyper visor to get better performance. (I don't recommend this).

edit: Did you consider using Vmware FT as a On-Demand task for this case? It would be the quickest fix. By running that host as Vmware HA temporarily or scheduled based. Its supported by vmware.

2
  • I'm looking at a VMWare FT setup which mandates virtual disks rather than direct SAN connections. The issue we are seeing is related to the mix of IO requests (reads vs. writes) and it affecting VMWare FTs replication technology getting swamped. The hope is that by changing the IO request size we can vary the IO mix.
    – davefiddes
    Oct 23, 2012 at 14:06
  • Did you consider using Vmware FT as a On-Demand task for this case? It would be the quickest fix. By running that host as Vmware HA temporarily or scheduled based. Its supported by vmware.
    – Sarge
    Oct 23, 2012 at 14:17

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .