We currently have a MD3000i with an iSCSI LUN shared out to our apache web server. We are going to add another apache web server into the mix using LVS to load balance, however, I am curious how well iSCSI handles file locking and data integrity. I have the iSCSI partition formatted as ext3.
Is my setup totally flawed and will ext3 not allow for data integrity with multiple apache hosts reading/writing to the shared storage? Am I required to also setup GFS across my apache hosts or does iSCSI somehow manage this?
Thanks for helping clear this up, James
jchase wrote:
We currently have a MD3000i with an iSCSI LUN shared out to our apache web server. We are going to add another apache web server into the mix using LVS to load balance, however, I am curious how well iSCSI handles file locking and data integrity. ...
it doesn't, not even close. iscsi is purely a block device protocol, it doesn't know what files are.
Is my setup totally flawed and will ext3 not allow for data integrity with multiple apache hosts reading/writing to the shared storage? Am I required to also setup GFS across my apache hosts or does iSCSI somehow manage this?
NFS would be more suitable for this application.
On 4/22/2010 6:51 PM, John R Pierce wrote:
jchase wrote:
We currently have a MD3000i with an iSCSI LUN shared out to our apache web server. We are going to add another apache web server into the mix using LVS to load balance, however, I am curious how well iSCSI handles file locking and data integrity. ...
it doesn't, not even close. iscsi is purely a block device protocol, it doesn't know what files are.
That was kind of a stupid assumption on my part and major misinterpretation of iSCSI information
Is my setup totally flawed and will ext3 not allow for data integrity with multiple apache hosts reading/writing to the shared storage? Am I required to also setup GFS across my apache hosts or does iSCSI somehow manage this?
NFS would be more suitable for this application.
Is there any reason why you suggest NFS? We would then need to run a NFS server and deal with the complexity of having a failover solution for that, whereas it seemed like GFS is a little more decentralized (at least in terms of not relying on a specific file server) after you configure the cluster environment.
That was my impression from reading through the docs anyways. I've never set it up.
CentOS mailing list CentOS@centos.org http://lists.centos.org/mailman/listinfo/centos
James wrote:
That was my impression from reading through the docs anyways. I've never set it up.
my impression is GFS requires shared storage, I believe there are ways around it, but take a look at this for setting up GFS for use with NFS
http://sources.redhat.com/cluster/doc/nfscookbook.pdf
I think it'd be much easier if you just replicate the data between the servers with rsync or something. GFS sounds like way overkill for a couple of web servers.
nate
my impression is GFS requires shared storage, I believe there are ways around it, but take a look at this for setting up GFS for use with NFS
iSCSI provides the basic foundation needed by GFS for shared storage, so the OP is good for that. GFS, however, is not exactly a simple technology to deploy. NFS is better for more standardized infrastructures where resources and time may be limited. GFS and other shared/clustering filesystems are great, but don't underestimate the resources needed to feed and care for them.
OCSF2 over iSCSI is a good option to look at, too. There is also gluster. But NFS is going to be the mainstream approach with the best support and administration options unless the OP is running into some technical limitations.
--------------------------------- Geoff Galitz Blankenheim NRW, Germany http://www.galitz.org/ http://german-way.com/blog/
On Fri, Apr 23, 2010 at 8:10 AM, Geoff Galitz geoff@galitz.org wrote:
OCSF2 over iSCSI is a good option to look at, too. There is also gluster. But NFS is going to be the mainstream approach with the best support and administration options unless the OP is running into some technical limitations.
I have a number of OCFS2 over plain old SCSI with MD3000 setups. The older versions have a couple of stupid bugs so it's always better to move to the most upto date version of the module from Oracle. I haven't tried it with iSCSI. I haven't tried Gluster yet but I find OCFS2 quite maintenance-free until you hit the damn bug! :)
http://sources.redhat.com/cluster/doc/nfscookbook.pdf
I think it'd be much easier if you just replicate the data between the servers with rsync or something. GFS sounds like way overkill for a couple of web servers.
Maybe you're right that GFS would be overkill -- I know you have to setup the whole clustering environment before it will work. Even though NFS would be easy to setup, it seems like it would just add more servers into the mix. We really want to achieve automatic failover at all levels and setting up NFS to replicate in real time and run the extra servers for that seems like it would require more resources. I'd rather put the time into understanding the complexities of the clustering environment setup and management and save some server sprawl
I'll look into OCFS2 and gluster to see if those are good options. Thanks for those suggestions.
nate
CentOS mailing list CentOS@centos.org http://lists.centos.org/mailman/listinfo/centos
On Apr 23, 2010, at 1:00 PM, James jchase@mandaladesigns.com wrote:
http://sources.redhat.com/cluster/doc/nfscookbook.pdf
I think it'd be much easier if you just replicate the data between the servers with rsync or something. GFS sounds like way overkill for a couple of web servers.
Maybe you're right that GFS would be overkill -- I know you have to setup the whole clustering environment before it will work. Even though NFS would be easy to setup, it seems like it would just add more servers into the mix. We really want to achieve automatic failover at all levels and setting up NFS to replicate in real time and run the extra servers for that seems like it would require more resources. I'd rather put the time into understanding the complexities of the clustering environment setup and management and save some server sprawl
I'll look into OCFS2 and gluster to see if those are good options. Thanks for those suggestions.
Web servers are mostly read-only, so unless your web servers are going to do a lot of writing to shared storage I would simply use rsync to a local disk in each server, or use NFS, even NFS and heartbeat for redundancy will be 100 times simpler to setup and maintain.
GFS/OCFS2/Gluster/Lustre are really for multi-writers to shared storage such as a large NFS server cluster (4-8-16 nodes) serving thousands of clients for general file services, not thousands of clients accessing shared storage directly.
-Ross
Web servers are mostly read-only, so unless your web servers are going to do a lot of writing to shared storage I would simply use rsync to a local disk in each server, or use NFS, even NFS and heartbeat for redundancy will be 100 times simpler to setup and maintain.
GFS/OCFS2/Gluster/Lustre are really for multi-writers to shared storage such as a large NFS server cluster (4-8-16 nodes) serving thousands of clients for general file services, not thousands of clients accessing shared storage directly.
It's a good point. Thanks for making it. OCFS2 seems like a pain to maintain with new rpm's for each new kernel update, and GFS a pain to setup at the least.
We just want scalability for future hosting options; it seems like GFS is the most powerful option. But you're right that there is not a lot of writing going on and NFS is probably the best option.
-Ross
CentOS mailing list CentOS@centos.org http://lists.centos.org/mailman/listinfo/centos
Hi James,
On Thu, Apr 22, 2010 at 11:38 PM, jchase jchase@mandaladesigns.com wrote:
We currently have a MD3000i with an iSCSI LUN shared out to our apache web server. We are going to add another apache web server into the mix using LVS to load balance, however, I am curious how well iSCSI handles file locking and data integrity. I have the iSCSI partition formatted as ext3.
iSCSI is just SCSI protocol over the net and doesn't provide any locking.
Is my setup totally flawed and will ext3 not allow for data integrity with multiple apache hosts reading/writing to the shared storage? Am I required to also setup GFS across my apache hosts or does iSCSI somehow manage this?
GFS or OCFS2 etc. would be a good idea. Alternatively NFS/CIFS would do.
-
Geoff Galitz -
Hakan Koseoglu -
James -
jchase
-
John R Pierce -
nate -
Ross Walker