Changing nfs.sendbuffersize - NetApp Arrays

[mblake4u]

Hi All,

I was wondering if anyone has changed the ESXi nfs.sendbuffersize parameter for NetApp filers (as recommended for the EMC Celerra arrays)?

EMC recommendations are to use a multiple of 32KB with 64KB as the recommended value (http://blog.scottlowe.org/2010/01/31/emc-celerra-optimizations-for-vmware-on-nfs/).

I wonder how VMware got the value of 264KB as their recommended value due to 'internal performance testing'.

I'm looking into this not because I'm having issues with my NetApp filers but because one of the NAS (Iomega IX2) I'm using for storing syslogs is reporting a RPC problem "fragment too large". It has a default NFS block size of 16KB (/proc/fs/nfsd/max_block_size).

I can increase this to 264KB but then it causes memory issues on the Iomega so what I'd like to find out is if I can reduce the nfs.sendbuffersize parameter without affecting my NetApp filer performance (this hosts my VM's). I'd also be interested in knowing if my NetApp performance might be increased by modifying the nfs.sendbuffersize parameter to something like 64KB (as with the EMC Celerras).

If anyone's got any experience with changing nfs.sendbuffersize for their NetApps or knows of any NetApp guidelines I'd really like to hear.

Cheers Michael

[maishsk]

Hi

TR-3749 - NetApp

will be where you can find the information.

Another good post - PowerCLI: NFS Settings for vSphere (NetApp NFS Recommendations)

None of the above say anything though about changing the NFS.SendBufferSize

KB 1007909 says that NFS.SendBufferSize shouldnot be changed

The default is 264

I would speak to NetApp.

Maish

VMTN Moderator | vExpert

Author of VMware vSphere Design

@maishsk | My Blog

[mblake4u]

Hi Maish

Thanks for your comments. I opened a call with NetApp and the person who helped me said he couldn't find any public or internal recommendations regarding VMware parameters, and he couldn't say what effect changing it would have on the filer.

NetApp do have some recommendations for their filer parameters:

https://kb.netapp.com/support/index?page=content&id=1012513

which talks about changing the nfs.tcp.recvwindowsize to 64240 but it's hard to tell if this is related to the VMware parameter. Firstly, it's also hard to know exactly if this NetApp parameter is bits/bytes/KB as it's a hidden parameter and secondly the value isn't a multiple of 1024 - if it were 65536 then it would likely be 64KB.

Another thing about this recommendation from NetApp is that the KB specifically refers to 10GE infrastructure.

Still not sure on whether changing the nfs.sendbuffersize is a good idea or not.. I'd love to know how VMware arrived at the value of 264KB which goes against EMC's recommendation of 64KB..

[joshodgers]

You should consider installing the Netapp VSC (Virtual Storage Console) plugin for vSphere.

The plugin has a feature which sets the recommended values for NFS on the hosts, by right clicking the host and selecting "Set recommended values"

See the graphic below.

Hope that helps.

[mblake4u]

Hi Josh,

Thanks for the post - I thought the VSC 'recommended settings' didn't modify that parameter but after your post I thought I'd double check. Applying the recommended values to one of my hosts doesn't change that paramter.

Regards, Michael