#cifs tuning

1 messages · Page 1 of 1 (latest)

surreal compass
#

Need to tune flexvol/flexgroup for performance with pacs cache on a220 for siemens syngo dynamics... only able to get 300MB/s and it is connected with 40gbps.... HELP!

hushed dome
#

how much storage is behind it? CIFS access is single thread per volume, so by going to flexgroup, you get multithread

brittle jetty
#

utilizing ODX or host based copy ?

surreal compass
#

we have ODX turned on SMB 3.11 - Flexgroup with 16 constituent volumes

#

also removed the smb throttling

#

Here is the current smbclient configuration:

Speed copying to storage (NetApp before disabling bandwidth):

Speed from storage to Windows server

I set the smbclient configuration false

Speed to storage after smb change

speed from storage to windows server

They seem to be the same

#

it is like 300TB all flash on 7.6TB Flash drives.... The VMDK performance of the same array attached via NFS is much better... something isn't right... we even moved it to a private network... Does continous availability help with performance at all?

slim zenith
#

nfs will most likely always be faster than cifs simply because of the underlying transport methods

surreal compass
#

cifs should not be capping out at 300MBPs over two 40gbps shared links when I can get TrueNAS to >900MBPs same hardware

surreal compass
#

any thoughts?

karmic canopy
#

Assuming all hardware is operating without error (including Netapp, host and switch ports), did you consider enabling jumbo frames? , if it is not already ? I know this is not an easy task. at the very least, a frame can fit in two pages of memory

crystal pasture
#

Are you doing a single file single volume transfer?

#

That's kind of a wierd platform because it has 12 cores, but only 6 used for WAFL. I'm not sure how many vol affinities it has.

#

Do you have a perf case open?

loud summit
#

I wonder what the network layout is , lacp, etc. smb is so chatty. Also depends on the data structure that is being moved. Is AQOS applied or something?

brittle jetty
#

i have seen cases when disabling ODX is beneficial on raw speed on the array but not necessarily ok from network congestion perspective

graceful token
#

network congestion? at 40Gbps there is certainly more than enough headroom on the network side. No 2-node AFF system can saturate that 🙂
My suggestion would also be to open a perf case but don't expect miracles 🤷‍♂️ I doubt you'll be hitting 900mbps on that A220