Re: Slow-running archive. Sometimes. Sometimes not.
"Jonathan Leffler" <jleffler@earthlink.net> wrote in message
news:9TMpc.2539$SZ4.82@newsread2.news.pas.earthlink.net...
> Neil Truby wrote:
>
> > IDS 9.21 FC4 on HP-UX 11.11
> >
> > For no reason that I can see, the back-up of a 350GByte database, which
> > normally goes to the 3 tape devices via onbar and takes about 5 hours,
runs
> > extermely slowly. Like now. The archive's been running for about 4
hours
> > and so far has written out just 58Gbytes.
> >
> > According to top the server is lightly loaded. Disk response times are
fine
> > according to sar -d. There are no major processes are running on the
> > Informix server (apart from the archive). There's just very little
going on
> > at all, on a quiet day. But I can see that at this rate the archive
will
> > take 24 hours to complete instead of the normal 5 (and I'll probably can
it
> > anyway as it has a detrimental effect on batch processing which starts
> > around midnight).
> >
> > Any ideas?
>
> Sounds like the very old archive page bug that afflicts very old
> (should be archived) versions of IDS. You should be upgrading to 9.40.
I bounced the server and re-started the archive. Now it's flying: 60-odd
GBytes in about an hour.
I did notice that, when it was running slowly, there was a very large amount
of write activity (none of it cached), as reported by onstat -p. Now the
amount of writes is neglibible. This obviously has something to do with it.
But is the "old page" bug? I've read, for example, Keith Simmons' account
of it on the cdi archive, and it doesn;t sound as simple as just bouncing
the instance ....
It's all gone crap again now :-((
Lots and lots of uncached writes.
"Neil Truby" <neil.truby@ardenta.com> wrote in message
news:2gpojrF4l92oU1@uni-berlin.de...
> "Jonathan Leffler" <jleffler@earthlink.net> wrote in message
> news:9TMpc.2539$SZ4.82@newsread2.news.pas.earthlink.net...
> > Neil Truby wrote:
> >
> > > IDS 9.21 FC4 on HP-UX 11.11
> > >
> > > For no reason that I can see, the back-up of a 350GByte database,
which
> > > normally goes to the 3 tape devices via onbar and takes about 5 hours,
> runs
> > > extermely slowly. Like now. The archive's been running for about 4
> hours
> > > and so far has written out just 58Gbytes.
> > >
> > > According to top the server is lightly loaded. Disk response times
are
> fine
> > > according to sar -d. There are no major processes are running on the
> > > Informix server (apart from the archive). There's just very little
> going on
> > > at all, on a quiet day. But I can see that at this rate the archive
> will
> > > take 24 hours to complete instead of the normal 5 (and I'll probably
can
> it
> > > anyway as it has a detrimental effect on batch processing which starts
> > > around midnight).
> > >
> > > Any ideas?
> >
> > Sounds like the very old archive page bug that afflicts very old
> > (should be archived) versions of IDS. You should be upgrading to 9.40.
>
> I bounced the server and re-started the archive. Now it's flying: 60-odd
> GBytes in about an hour.
>
> I did notice that, when it was running slowly, there was a very large
amount
> of write activity (none of it cached), as reported by onstat -p. Now the
> amount of writes is neglibible. This obviously has something to do with
it.
> But is the "old page" bug? I've read, for example, Keith Simmons' account
> of it on the cdi archive, and it doesn;t sound as simple as just bouncing
> the instance ....
>
>
On Sun, 16 May 2004 16:32:21 -0400, Neil Truby wrote: > It's all gone crap again now :-(( > Lots and lots of uncached writes. > Right, because it's gotten past the pages whose timestamps have already been updated and now the archive is hitting very old pages again. Art S. Kagel
Yes, Keith pointed this out to me as well. I think we'll just have to grin and bear it. We have been trying to convert this customer to IDS 9.40 for almost a year: they are however in the hugely competitive convenience foods sector - very high-volume/low-margin - and are thus enormously risk-adverse. And since 9.40 has not yet released a version in which the regression to a previous release works, we've held off. Now we have a patched version - 9.40FC3W2 - in which the regression does work, so we're now rolling it out. Hopefully over the long weekend we have here at the end of May. regards Neil