Page 1 of 1

Refresh in Director very slow

Posted: Mon Dec 20, 2004 11:21 pm
by chinek
Hi,
We have just installed v7.5 in a new F15K SUN server (Solaris 2.8)
and 4 CPUs and 8 GB memory.
However in DS Director in a project that has 392 jobs , each time the screen refreshes it takes about 30 seconds whereas in the old server running v5.1 it takes just 3 seconds.

So what gives ? Is the way Director in v7.5 refreshes the screen different than v5.1 ? Is there a way to speed this up because this is very annoying.
I have also extended the refresh rate to every 65 secs but would rather increase that to more.

Can this be extended beyond 65 secs ? Also ideally we would want the refresh to refresh quicker.

Can someone help ?
Thanks.

Posted: Mon Dec 20, 2004 11:50 pm
by davidnemirovsky
Are your jobs in that project sorted into categories? The more jobs in a category, the longer the refresh takes. Moving jobs into a category does not alter the jobs in any way.

Not sure about the refresh rate being able to be higher than 65 secs.

Re: Refresh in Director very slow

Posted: Tue Dec 21, 2004 12:04 am
by chulett
chinek wrote:Can this be extended beyond 65 secs ?
No, valid values are 2 - 65 seconds only. Not sure why, that's just the way it is.

Can't say in what way the Director has changed, but I'm sure it has under the covers since the 5.x days. As noted, the best way to speed things up is to minimize the number of items being refreshed at any given time. Simplest way to do that is through the use of Categories. There are also Filters you can apply, which can help at times, but in general the liberal use of categories is best.

Your network speed plays a role as well, but I'm assuming that hasn't changed. FWIW, my largest category currently has 125 jobs in it and it take ~13 seconds to refresh (initially) - at home over VPN. It's quicker when on the network at work and much quicker on subsequent refreshes.

Posted: Tue Dec 21, 2004 12:44 am
by ray.wurlod
Which view in Director?

Status view needs to refer to the RT_STATUSnnn table for each job in the currently selected category. (Hint: don't have too many jobs per category.)

Log view needs to refer only to the RT_LOGnnn table for the currently selected job, but this may involve a large number of events, and they have to be sorted into timestamp order. (Hint: keep logs as small as possible.)

Schedule view needs to refer to the operating system scheduler (crontab or AT) as well as the RT_STATUSnnn table for each job in the currently selected category.

Monitor view needs to look at each record in the RT_STATUSnnn table for the currently selected job, current wave (and, possibly, current invocation ID), and perhaps RT_CONFIGnnn as well. The same is true if you have "show performance statistics" selected in Designer.

Each refresh runs as an interactive process on the server, saying "give me an update". Since interactive processes have higher priority than background processes, this can negatively impact performance of running jobs.

Prefer not to have Director open when jobs are running. Next best is to keep Director minimized as much as possible.