We have received a sugestion from Digital to try a patch to vmunix, setting
prmetaonly=1 to increase write throughput to NFS mounted ADVfs partitions.
This is supposed to have PrestoServe cache metadata only.
Has anyone else tried it?
We will try it as soon as we can pull one of our PrestoServe enable servers
from production.
doug
>> I've not seen it mentioned on this list...
>>
>> We are seeing 3 times performance increase by changing
>> _from_ an ADVFS _to_ a UFS on our prestoserve enhanced NFS server.
>> The network is _not_ loaded. This seems to be client independent.
>>
>> We have logged it with Digital support, they have other reports but no
>> patch as of yet.
>>
>> ouch!
>> doug
>
> Pardon my lack of clarity... the hit we were taking was on the write side
> of NFS. First a thank you to those who replied.
>
> John Stoffel <john_at_WPI.EDU>
> "Craig I. Hagan" <hagan_at_ttgi.com>
> Rick Muse <m6u_at_oaunx1.ctd.ornl.gov>
> Ron Barrett <Ron_Barrett_at_corp.Cubic.COM>
> aidan_at_cse.unsw.edu.au (Aidan Williams)
> Benoit Maillard <maillard_at_fgt.dec.com>
>
> I believe the biggest problem is that PrestoServe doesn't cache metadata on
> ADVFS. There is also a penalty to pay for the logging in ADVFS on local
> writes.
>
> We have an escalated call open with Digital on this, but have not heard
> from engineering yet. We may soon have to revert our 120+GB disk farm back
> to UFS unless there is some help soon.
>
> double ouch!
> doug
Received on Wed Jan 17 1996 - 23:08:47 NZDT