we have experienced a BIG slowdown of CIFS when mounting alf_data to other server. It drooped down from 5000 kb/s to 20-420 kb/s. We have tried to move lucene tables to local disk. but no help.
Hallo Andy, thanx a lot for your answer.. Im only a business user so the short answer from our administrator is: alf_data/ is mounted as 192.168.1.1:/mnt/raid5/alfresco/alf_data /opt/alfresco/alf_data nfs rsize=8192,wsize=8192,rw 0 0. NFS is version 3 and direct performance is about 5-8MB/s
The indexes really need to be local to the machine. Neither we (nor lucene at the version we use) support NFS for indexes. We will in the future. Certianly you must not share indexes in any way over NFS. If you have a cluster each machine must have its own index.
The CIFS protocol tends to be a bit chatty …. and can create and change files in bits. Alfresco could be creating and reading a few files per operation just to support the bits … over NFS and RAID 5 may affect the write speed.
What CIFS client are you using? Is it read, create, or update performance we are talking about?
How does performance for FTP compare doing the same thing?
Now does your NFS performance compare to the direct disk performance you had before?
Have you got any caching on the NFS client side? As we only add files and then treat them as read only (and clean up unused later) caching will work well.
Hallo Andy, sorry for the late answer, but we could not manage to do test sooner.
We are testing on windows XP clients (total commander 6.50 and native explorer)
NFS mount is on RAID5
We tested CIFS in both scenarios - read from alfresco and upload to alfresco. Problem is only with upload when content store is on nfs share. It looks like alfresco does several operations with file after upload in contenstore.
Performance is similar also with ftp. Download from alfresco is at full speed of 100M LAN, upload is about 300-400kB/s. Direct nfs performance is 5500-10000kB/s.
If there is anything else we should test please let me know.