Unsolved
This post is more than 5 years old
2 Intern
•
301 Posts
0
1537
isilon nfs operations slow ?
Hi,
We are looking to migrate some NFS load from an existing 'vanilla' nfs server to using isilon as nfs. With environment being same, except swapping the NFS backend, we see total time for a run go from ~ 2 mins to ~ 14 mins.
We carried out a wireshark capture to get SRT( Service Response Time ) of captured NFS procedures. This, when multiplied with the number of calls seems to be where the 'lost' time is.
Is there a contact point within the Isilon team to investigate NFS procedure performance issues ?
I am unsure if this issue is under the remit of the solutions architect associated with this project, but do not know how to investigate further, any pointers as who to talk to is appreciated.
Peter_Sero
1.2K Posts
0
July 23rd, 2018 10:00
Hope your solutions architect has been helping you out.
As an aside, let us keep in mind that "vanila", in particular the "plain vanilla" flavours,
respond very fast in certain scenarios -- because they CHEAT on the NFS protocol:
Where NFS demands a guarantee that file creates, chmod/chattr, and certain "stable" data writes
only get acknowledged as successfull to the client when the operation has hit
permanent storage (disk/ssd/nvram), "vanilla" NFS servers often adopt the "async"
operation mode of the underlying local file system. Which means that things are
only in the RAM on the server whereas the client would assume things were safely on disk.
As an alternative, vanilla servers offer full "sync" behaviour,
which is usually way too slow for most uses.
In contrast, dedicated NAS systems make sure they buffer to RAM only where allowed
by the NFS protocol, and do wait for permanent storage where required.
(keywords: "stable" / "unstable" NFS writes)
makes sense?
-- Peter