Hi -
I've finished testing the latest utils package, nfs-utils-1-0-1-pre8,
both with udp and tcp mounts to the server, in both a 2.4.18 standard
and patched (Trond's NFS_ALL patch) environment, and it is ready to
go as far as I am concerned. The difference between pre7 and pre8
was simply a warning message in exports.c, so any testing being
done on MP platforms (by those of you who have access to them)
need not be repeated.
A couple of questions have come up in my testing, which may have
to be addressed elsewhere, but are not related to the package:
- Does locking take a much longer time in ext3 file systems,
or in the patched environment? In the patched kernel,
exporting ext3, the connectathon lock tests seemed to take
up to 10x as long, but that was over many hours, with a big
load. When run just once with the other connectathon tests,
it didn't seem so different.
- The client net statistics in /proc/net/rpc/nfs seems to remain
at 0 regardless of the activity and regardless of the kernel.
I'm going to start browsing through the kernel code to see
if there is something obvious, but maybe someone knows about
this already. I didn't see anything in the mailing list.
Anyway, as soon as we hear results of MP testing, I'd say its
ready to ship.
Tom
--
------------------------------------------------------------
Tom McNeal [email protected] (650)906-0761 (cell)
------------------------------------------------------------
_______________________________________________________________
Don't miss the 2002 Sprint PCS Application Developer's Conference
August 25-28 in Las Vegas - http://devcon.sprintpcs.com/adp/index.cfm?source=osdntextlink
_______________________________________________
NFS maillist - [email protected]
https://lists.sourceforge.net/lists/listinfo/nfs
>>>>> " " == Tom McNeal <[email protected]> writes:
> - Does locking take a much longer time in ext3 file systems,
> or in the patched environment? In the patched kernel,
> exporting ext3, the connectathon lock tests seemed to take
> up to 10x as long, but that was over many hours, with a big
> load. When run just once with the other connectathon tests,
> it didn't seem so different.
Flushing to disk would be slower with ext3. As this is done every time
somebody takes or releases a lock...
> - The client net statistics in /proc/net/rpc/nfs seems to remain
> at 0 regardless of the activity and regardless of the
> kernel. I'm going to start browsing through the kernel code
> to see if there is something obvious, but maybe someone
> knows about this already. I didn't see anything in the
> mailing list.
I've never touched the 'net statistics' field, and I've no idea what
it purports to measure. As far as I'm concerned it can be removed
entirely...
Cheers,
Trond
_______________________________________________________________
Sponsored by:
ThinkGeek at http://www.ThinkGeek.com/
_______________________________________________
NFS maillist - [email protected]
https://lists.sourceforge.net/lists/listinfo/nfs
I have run similar tests against a 4-way server running the 04/04/2002
snapshot of Trond's linux-2.4.18-NFS_ALL.dif patch and
nfs-utils-1-0-1-pre7. The tests consisted of simultaneous runs of
bonnie++, fsx-linux, and different invocations of Connectathon2002. There
were no apparent issues (although I'm not very familiar with bonnie++).
Tom, as for your question regarding client net statistics, IIRC these stats
were never collected.
Regards,
---
Bruce Allan <[email protected]>
Software Engineer, Linux Technology Center
IBM Corporation, Beaverton OR
503-578-4187 IBM Tie-line 775-4187
Tom McNeal
<[email protected]> To: NFS maillist <[email protected]>
Sent by: cc:
[email protected] Subject: [NFS] Utils package testing
ceforge.net
06/11/2002 09:33 AM
Hi -
I've finished testing the latest utils package, nfs-utils-1-0-1-pre8,
both with udp and tcp mounts to the server, in both a 2.4.18 standard
and patched (Trond's NFS_ALL patch) environment, and it is ready to
go as far as I am concerned. The difference between pre7 and pre8
was simply a warning message in exports.c, so any testing being
done on MP platforms (by those of you who have access to them)
need not be repeated.
A couple of questions have come up in my testing, which may have
to be addressed elsewhere, but are not related to the package:
- Does locking take a much longer time in ext3 file systems,
or in the patched environment? In the patched kernel,
exporting ext3, the connectathon lock tests seemed to take
up to 10x as long, but that was over many hours, with a big
load. When run just once with the other connectathon tests,
it didn't seem so different.
- The client net statistics in /proc/net/rpc/nfs seems to remain
at 0 regardless of the activity and regardless of the kernel.
I'm going to start browsing through the kernel code to see
if there is something obvious, but maybe someone knows about
this already. I didn't see anything in the mailing list.
Anyway, as soon as we hear results of MP testing, I'd say its
ready to ship.
Tom
--
------------------------------------------------------------
Tom McNeal [email protected] (650)906-0761 (cell)
------------------------------------------------------------
_______________________________________________________________
Don't miss the 2002 Sprint PCS Application Developer's Conference
August 25-28 in Las Vegas -
http://devcon.sprintpcs.com/adp/index.cfm?source=osdntextlink
_______________________________________________
NFS maillist - [email protected]
https://lists.sourceforge.net/lists/listinfo/nfs
_______________________________________________________________
Sponsored by:
ThinkGeek at http://www.ThinkGeek.com/
_______________________________________________
NFS maillist - [email protected]
https://lists.sourceforge.net/lists/listinfo/nfs