Hello All,

Currently stuck in a little situation and i am leaning towards trying to use GFS to get my way out of it. Heres the situation

I have a load balanced environment balancing mod_jk between 4 systems 2 apache instances on the outside that talk to 4 other servers each running tomcat and 2 load balancers above my apache instances running wrr distributing connections... the only problem is is that if one of my tomcat sessions die then the data being uploaded to that server will be lost when the session is passed to another server because the data is being stored locally to that certain tomcat instance.. and the one it fails over to has no idea what the session is talking about resulting in a failed upload and having to restart the session... i am pretty sure that if i have a GFS running on the background with all these tomcat instances it will allow all of them to store the data to a central location and when one tomcat instance fails and gets pushed to another the other will be able to continue receiving the data thats being uploaded. if i am wrong please somebody correct me...

If this is possible i was looking at CentOS's release of GFS 6.1 and am thinking this is the route to take... i was also looking at iscsi(using backend switch) but the only problem with that is is when one of the tomcat instances is writing to that certain folder i am pretty sure it locks it? or if it doesnt then the other machines wont be able to see the data till its remounted? with this again i am not totally sure but i had a feeling its not the ideal solution also because i get a transfer cap at i think 50mb and if i have over 100 people uploading data(comes in 64k chunks) then i can run into a problem...

Anybodies opinions or knowledge on this would be great... thanks

From the filesystem point of view this will work as described.

Don't mix up iSCSI and GFS. iSCSI is a block based storage protocol underneath any filesystem also GFS. When going the shared storage block based way (this includes usage of GFS or the like) you first need to think about providing a shared storage infrastructure for all the four nodes. This means either iSCSI, Fibre Channel, parallel SCSI (will be a little hard with more then two nodes) or the cheaper ways like SAS.

When this is settled bring in the means to share filesystem data between the nodes. This implies a Cluster Filesystem like GFS.

On scalability with GFS (implies a little bit of locking):
Normally a symmetric cluster filesystem like GFS should scale linear even with concurrent access (Locking is done distributed). But when having many files in one directory you might not see a linear scalability line.

so truly the GFS route would be an ideal solution for each tomcat instance being able to contact the same files?

Basically yes. From filesystem point of view yes. But the application itself might see this different. For example if the application concurrently writes the same file the application must keep track on file data consistency. The filesystem only keeps the file itself in the filesystem consistent. But this holds for any filesystem local or distributed or shared.

But if read from multiple sources this is no problem or if written to different files still no problem.

yeah all the files that are being written are all different constantly different.