Gluster users - Oct 2010 - Gluster in a web cluster

Hi,


I'm trying to figure out a solution for a web cluster, where part of the 
sources will be fixed (those can be handled by a rsync from the dev 
environment) - and the rest will be user sent datas (pictures and so on).
I wanted to avoid the NAS solution for various reasons (price, nfs 
locking, redundancy) and ended getting to know gluster.
The RAID1 setup seems perfect for my case, but I have some questions :
* Reading a file on the raid "array" makes gluster check if there is a
update on other servers - is there a way to optimise this ? (cache for x 
seconds for example)
This seems the goal of performance/io-cache - am I correct ?
* I don't mind to miss a file for under a second, hence is it possible 
to get gluster to send "new/updated files" messages to his brothers - 
instead of having the children check for the new file each time ?
* Quick-read seems a nice idea if the partition hosts small files (won't 
be used here, as I'll be hosting images)

Will I manage to get nice performances with this setup ? (I plan to have 
a cache in front, I want to know about raw performances)
Will those performance degrade a lot when I add more nodes (for now, 
I'll only have 2) ?
Is putting my whole document root on the raid1 a good idea ? (this will 
depend on the refresh issue I suppose)


Thanks in advance,
Mathieu

Hi,


I'm trying to figure out a solution for a web cluster, where part of the 
sources will be fixed (those can be handled by a rsync from the dev 
environment) - and the rest will be user sent datas (pictures and so on).
I wanted to avoid the NAS solution for various reasons (price, nfs 
locking, redundancy) and ended getting to know gluster.
The RAID1 setup seems perfect for my case, but I have some questions :
* Reading a file on the raid "array" makes gluster check if there is a
update on other servers - is there a way to optimise this ? (cache for x 
seconds for example)
This seems the goal of performance/io-cache - am I correct ?
* I don't mind to miss a file for under a second, hence is it possible 
to get gluster to send "new/updated files" messages to his brothers - 
instead of having the children check for the new file each time ?
* Quick-read seems a nice idea if the partition hosts small files (won't 
be used here, as I'll be hosting images)

Will I manage to get nice performances with this setup ? (I plan to have 
a cache in front, I want to know about raw performances)
Will those performance degrade a lot when I add more nodes (for now, 
I'll only have 2) ?
Is putting my whole document root on the raid1 a good idea ? (this will 
depend on the refresh issue I suppose)


Thanks in advance,
Mathieu