Quickplace

Overview

From the examples given the biggest OS bottle neck in Quickplace is the disk queue length.
I have seen numbers as hight as a constant 50 when 1.2 is high enought.
The default install for quickplace server does not work for clusters. What often happens is that two individual servers are installed and clustered together when in fact you need specific configurations for each server in a cluster. With a disk queue lenght of 50+ the first thing you need to do is to find out what is hitting your disk and how to recude it. Know the tasks that run on your servers and what you can do about them to reduce the hit on the server.

I also remove a lot of tasks form the notes.ini and put them in program documents as this gives me a lot more control over when they run.

Newsletter

By default the newsletter daily task runs is run from the servertasksat1. If you have two servers in a cluster you don't need to run it on both servers. Also think when you need to run it and if you need to run it. If the server is in the USA then 1 is 7am in central europe so its not a great time to run disk tasks. I set it to run on the second server in the cluster at 19:00 US time and only on Monday through to Friday. I try and leave the weekend for heavy i/o admin tasks. The weekley newsletter I run on the primary server at 12:00 on a Saturday. The admin tells you to set a program document for this so why not also set one for the newsletter daily.

Updall

The updall task does everything an update does plus some. The trouble is that its about 10+ times more I/O intensive and it also is set to run at 04:00. Ohh look 10am central Europe time great time to run it. NOT! I have seen quickplace systems where the updall task takes 5 days to run. But then its loaded every day. What a way to kill your i/o and put your disk queue lenght thorugh the roof. I set this in a program document and run it at 02:00 on a Saturday. You still need to monitor how long it takes to run and this goes for every taks. You need to know all your task and its impact on the system but cutting this down is going to be a massive saving on your i/o.

placecatalog -update

Again you don't need this running on each server in the cluster. The replicamaker and replicator keep the servers in line so you only need to keep the placecatalog -update on one server. I set a program document to run this at 20:00 on MTWTF (Monday through to Friday) on the primary server in the cluster. The QPTool placecatalog -update command can be run on the Place Catalog server to synchronize data between a place's entries on each physical server, and the place's virtual entry. For example, before QPTool placecatalog -update is run, the place's virtual entry contains the membership information, but the place's physical server entries do not. After QPTool placecatalog -update, both entries contain the same field values.

Updatemember -allmember

This will keep user inforation in quickplace in line with your ldap directory. So should a users email address change then this will be reflected in quickplace. The admin manual says put this in the servertasksat3. But I don't think you need this ever day. I set it up to run at 21:30 on Wednesday and Saturday. But it depends on your system and rates of turn over and change. But not ever day. Every think we cut out reduces the diskqueuelenght and we want 1.2 not 50+.

Catalog

I find very little use for the catalog task in quickplace as I tend to use the placecatalog. That said I still run it but I run it once a week 12:00 on a Saturday. Remember ou might need to set CATALOG_POOL_SIXE_MB=5 else you will get memory errors from the catalog task on a quickplace server with lots of places.

Compact

It a balance between getting back disk splace and hitting your i/o. If you have a static number of places then I put this -B -U -S 20 at 04:00 on a Saturday. I get my disk space back once a week from comact tasks.

Design

Again I run the design task once a week at 20:00 on a Saturday. I only run it on once server and let the replicator move any changes around.