CIFS - browsing large directories

[drope]

Hello,

Whenever I try to browse a directory that has 1000+ items using CIFS, Alfresco server takes too long to load all the data (about 10 seconds).

First thing I tried  was to increase Java Heap to 1Gb (set JAVA_OPTS=-Xms512m -Xmx1024m -Xss64k -server). Then I turned the ehcache debug on and did some tweaking on ehcache.xml.  Results were better but still too slow.

Now I'm wondering what else can be done to increase the performance on my server. I read in another topic (http://forums.alfresco.com/viewtopic.php?t=2354) that JLAN Server has 30% increased performance over built-in CIFS code in Alfresco? Is this still the case? Anything else I could try?

I also noticed that CPU usage goes up to 90% at those give times (70% java, 20%mysql). I could upgrade my hardware but something tells me it's not just a hardware limitation.

System info:
Alfresco Community 2.1/Tomcat/MySQL/Windows Server 2003
Intel Core 2duo 2.4Ghz/2Gb RAM/2x160Gb Stata 2 - Raid 0
10 concurrent users, 20k items on repository

Thanks in advance,
Pedro Son

[drope]

I'm wondering if no one else have spaces with a large number of documents.

Any reply woul be appreciated. Even something like "Alfresco CIFS implementation is not suposed to handle that well".

Thanks again,
Pedro Son

[dgenard]

Hi, we've the same issue.

Using Windows Explorer via CIFS, opening a space containing 650 documents takes about 20 secs.
CPU power is used mainly by tomcat.
So, unfortunately, it's seems to be a common behaviour.
Did you finally manage to improve the speed ?

Denis

[tim-erwin]

Same here. On a QuadCore with 4GB (that alfresco may use) and two fast hard drives (one for the repository and one for the db) it takes more than 30secs for a folder with around 3000 files. The only way - I guess - is to split the folders up. Have several smaller ones. I heard a number of 500 being the limit for acceptable speed.

Performance is an aspect of alfresco that I'm a bit disappointed of, although the webclient is pretty fast (using the native API and mechanisms to reduce the amount of data read).

Tim