Unsolved
1 Rookie
•
5 Posts
0
163
December 1st, 2024 10:39
NFS performance
Dear community,
We have observed a significant performance degradation after migrating our NFS storage from a normal Linux NFS server to a PowerScale NFS environment (cluster with several nodes). Both systems are configured to use NFSv4.2.
Interestingly, when we switched to NFSv3 on PowerScale, the performance issues seemed to resolve themselves. This suggests a potential incompatibility or configuration issue specific to NFSv4 on PowerScale.
I want to know the exact reason, not just guess
It's worth noting that the PowerScale cluster is not under heavy load; even with minimal usage, we continue to experience performance problems. The NFS data is primarily used by a web application.
Any ideas?
Thanks
DELL-Josh Cr
Moderator
•
9.2K Posts
0
December 2nd, 2024 14:15
Hi,
Thanks for your question.
What kind of performance issues are you seeing? What version of OneFS are you using? Are there any errors in the logs?
Let us know if you have any additional questions.
Gian747
1 Rookie
•
5 Posts
0
December 2nd, 2024 15:58
Hi Josh,
some requests processed by the Apache server (from nfs) take long time: 1-7 seconds...
OneFs 9.5.0.6
No errore on the logs
Thanks
DELL-Josh Cr
Moderator
•
9.2K Posts
0
December 2nd, 2024 16:22
There are some NFS fixes in the release notes, it may be worth updating. https://dell.to/4gc5132
Gian747
1 Rookie
•
5 Posts
0
December 7th, 2024 16:55
I read the release notes but I can't find any fixes that make me think about our problem. We have currently decided to use NFSv3 but will plan to upgrade
Phil.Lam
3 Apprentice
•
608 Posts
0
December 19th, 2024 15:52
@Gian747 ,
https://www.dell.com/support/kbdoc/en-us/000063022/emc14001361-best-practices-for-nfs-client-settings
NFSv4
NFS version 4 is the newest major revision of the NFS protocol, and is increasing in adoption. At this time NFSv4 is generally less performant than v3 against the same workflow due to the greater amount of identity mapping and session tracking work required to reply. Here are some of the key differences between v3 and v4
Gian747
1 Rookie
•
5 Posts
0
December 19th, 2024 16:08
Hi Phil,
I understand that NFSv4 is different from NFSv3, not only because it is newer (introduced in 2000) but also due to the improvements it brings, such as advanced security support and session management. However, I cannot explain why, when using NFSv4 on a server running a Linux distribution, the issue I encounter with OneFS configurations does not occur.
Phil.Lam
3 Apprentice
•
608 Posts
0
December 19th, 2024 16:45
isis the LDAP/NIS authenication provider first in your zone?
isi zone list -v
EXAMPLE
iston-1# isi zone list -v
Name: System
Path: /ifs
Groupnet: groupnet0
Map Untrusted:
Auth Providers: lsa-ldap-provider:LDAP, lsa-local-provider:System, lsa-file-provider:System
NetBIOS Name:
User Mapping Rules: *\* &= * [], *\* += * [group,groups]
Home Directory Umask: 0077
Skeleton Directory: /usr/share/skel
Cache Entry Expiry: 2H
Negative Cache Entry Expiry: 1m
Zone ID: 1
(edited)
cadencep45
3 Apprentice
•
318 Posts
0
January 27th, 2025 13:26
suggest unless already resolved, log a ticket after setting up a environment to recreate the issue
1. With NFS 3 to isilon server
2. With NFS 4 to isilon server
3, With NFS 4 to linux server
once you have that, it should be fairly straightforward to gather logs for 1 & 2
3. Is more for application response time validation
Gian747
1 Rookie
•
5 Posts
0
January 27th, 2025 13:45
we immediately opened the ticket with DELL (before trying here... we are not particularly satisfied with the support)
we had the 3 environments you described and the support collected the various logs
the answer: use NFS3!!
now we are using NFS3 and everything works without problems... I would have liked to understand the reason
cadencep45
3 Apprentice
•
318 Posts
1
January 27th, 2025 14:44
Yep,
Its falls into a grey area, where after break/fix, support can be say, solution is to roll back to pre-existing working solution. Been there ( am there now ), where the solution from support, is the status quo, but its not really helpful in developing an improved solution.
Guess the only way of progressing situations like this is ensuring account team are aware of constraint when trying to get optimal use of kit.