Eric> Does anyone have experience with running production workloads on Eric> FC-based LUNs on NetApp? Am curious to know how performance of Eric> hosting virtual machines (including Exchange, database Eric> environments) compares to more traditional block-based SANs Eric> (EMC, 3Par, Hitachi, etc), since what I’ve read is that NetApp’s Eric> LUNs feature still sits on top of WAFL?
I've run Oracle DBs on Netapp LUNs in 7-mode. We ended up with performance issues which were solved by adding in a new dedicated shelf for that particularly crappily setup Oracle DB and application.
We now run a cDOT cluster on four nodes doing alot of VMware ESX DataStores, but using NFS. We do use some iSCSI LUNs for MS SQL stuff that requires it, and also Oracle on NFS.
Eric> We have some native FAS NetApps, along with many N-series Eric> rebranded NetApps, but all are run in 7-Mode and using NFS Eric> connections.
This works well, I like NFS and Netapps run it well. The 32xx series are a but low powered and serverly expansion limited in my experience. The new 80xx series is better, but still limited at times.
Eric> Also, how do you all implement data tiering in your NetApp Eric> environments? We are currently using IBM SAN (Storwize/V7000) Eric> and this has tiering capability. We’d consider moving some SAN Eric> workloads to NetApp if we could get as good SAN performance and Eric> also address the tiering capability.
What do you mean exactly by tiering? Are you moving LUNs between types of disks with different performance characteristics? I find that moving LUNs/Volumes is doable, but takes alot of time and alot of space with a decent impact on performance, since you need to copy all the blocks from one aggregate to another using snapmirror before you can cutover.
I can't say I've moved many LUNs, it's usually been simpler to just spin up a new LUN, mirror it on the host, and then detach the original LUN and destroy it when done.
But it all really depends on your current workloads on the FAS filers and if you can add the needed FC Fibre Cards for doing the LUNS.
But overall, I dislike LUNs because you pay a big overhead penatly and lose lots of disk space and the flexibility to dynamically grow and shrink NFS/CIFS volumes in a much simpler and quicker way.
LUNs are just a pain, unless you plan them out ahead of time from the host side with block management in place to help grow/shrink/move stuff around.
John