HP StoreVirtual NFS Support

[lithium_5]

I was in the process of selecting a distributed file system solution and was recommended the HP StoreVirtual range and/or the HP StoreVirtual VSA by our HP account manager.  I am having trouble finding good documentation on the product and therefore am unable to fully evaluate what seems to be a good product.

Just to give some background - We have an Active-Active DC environment, so we are looking at the Mult-site cluster option.  We have a requirement for file storage access - small footprint - could be a 2-3 TB with a couple of thousand new files a day.  We are looking into this product from the HA perspective rather than the massive scalability. 

I have a couple of questions:

1. As I understand it, the only method of connectivity is via iSCSI/FC (although I was told it natively supports NFS and CIFS).  The datasheet mentions that  CIFS, NFS etc. support can be added with the addition of a HP StoreEasy 3830.  Is this just a server that can "see" the LUNs you have presented via the StoreVirtual Nodes and then shares these out as NFS?  And if so what kind of HA options are available and is there any relationship to the StoreVirtual Cluster.

Would I build out 2 of these servers per site and load balance my client servers across them? 

2. Do I need this gateway server?  Could I instead connect my client machines (mainly RHEL 5/6) to the cluster via iSCSI?

Any answers or any advice would be greatly appreciated.

Datasheet link:

http://h18004.www1.hp.com/products/quickspecs/14437_div/14437_div.pdf

[oikjn]

Lets just call that acct manager a typical salesman ;)

The StoreVirtual System is 100% iSCSI.  any NFS solution is another solution layered on top.  The StoreEasy might be that layer, but I cannot speak about that as I don't have any experiance with it.

If your goal is AVAILABILITY and redundancy, you really can't compete with the StoreVirtual solution until you start spending some serious bucks.  Thats exacly why I went with it.  The downside is that it is extreemely inefficient when you get into your usable SAN space compared with raw HDD space.  That said, for 3TB usable storage space, that isn't really much of a problem.

Two nodes per site would be a good plan.  Just keep in mind if you want to have just site failover or if you want to have site+node failover will mean your raw storage requirement will effectively double.  That means if you need 3TB usable, if you only need site redundancy, you can have two 1.5TB nodes at each site(HP calls it Network Raid10), but if you want a site to go down and one node, you will have to have 3TB nodes (HP calls it Network Raid10+1).

The good news is the SAN and the windows DSM is Site aware so the servers will talk directly to the local nodes first, but that said, your speed and latency is still going to be limited by your site-site bandwidth and latency and if you don't have a really really good connection you might be better off using remote snapshots instead of a true multi-site SAN.

Also keep in mind that any multi-site install requires a THIRD site to house the FOM to handle quorum.   That has very little bandwidth or capacity needs, but is required in order to have a seamless failover when either site goes offline.

All in all, StoreVirtual is only iSCSI, but if you manage your physical environment correctly and WAN, you can get MANY 9's of availability.

[lithium_5]

Many thanks for your reply.   I got some more information on the StoreEasy product - It is just a gateway server that mounts the block iSCSi storage presented by the StoreVirtual SAN.  It runs windows storage server and the only way to provide HA on this would be to cluster two of these machines, which is not something I would be willing to support cross-site.

I played around with mounting the iSCSi Lun on a VM that presented the storage as a NFS share and found that although the Block level layer redundancy was really good, I would then have to provide HA for the NFS server, both in-site and across both sites.

StoreVirtual itself is a fantastic product and I had some great results in my test environment, but I guess what I need is a software defined storage product  that presents a distributed file system.

Any suggestions would be greatly appreciated.

Thanks,

Cormac

[oikjn]

can you say what the reason for the NFS requirement is?  that might help direct a better suggestion.

[lithium_5]

Sorry for the delay in getting back to you.  I should have stated that NFS connectivity is not a hard  requirement. 

What I basically need is a file store that is replicated between both of our datacentres and is higly available to client machines (Linux servers) in both sites.  

I have looked into RHSS (Redhats version of GlusterFS) which allows client machines to mount the volumes with the FUSE driver.  This is attractive as the driver is cluster aware, but the solution itself does not have the replication factor that I require.

I am also looking into SwiftStack which is an object based storage solution rather than file based.  This provides an API for "putting" and retrieving files rather than providing a file system. 

Basically, I am very open to any ideas - I just need a solution that will allow client servers in each site read/write files to/from a distributed redundant storage area.

Thanks,

Cormac

[oikjn]

when you say replicated, do you require synchronous replication or is asynchronous sufficient?  synchrounous is typicallty significantly more expensive and more complicated and I'm not just saying it because typically vendors charge more for that feature, you have to also suddenly have to deal with bandwidth and latency RTT and performance at BOTH sites as any one of those items can become the weak link in the chain and cause your storage to slow down.

If you can manage to stay asynchronous your costs will go way down and your options for solutions will go way up.

Do you need a clustered file system?  Does it have to be accessible to both sites at the same time (read/write), or is this for DR failover active/passive only?  Is this for providing space to something like ESX/Hyper-V to provide virtual drives to VMs, or is this more of you trying to replicate something like dropbox or maybe a traditional silmple file share but spanning two sites?

Type/pattern of IO or file access is important to understand as well.  iSCSI may be the best solution if you are talking about needing many servers at different locations that only need access to a raw disk, but iSCSI isn't a solution to anything more complicated than "hey I want to place a HDD on machine X and machine Y and maybe I also want them to share that same HDD."