Hi, Although I have used the NetApp discussion forums several times, this is my first post. I have been working with Netapp for 3 years the whole time has been with 3140's. I have started a new position this week with a new company. They have two 3240s exporting datastores to VMware via NFS. There are two Aggregates on each head. One is Aggr0 with vol0 and the other is Aggr1 with several volumes (vol1 4TB,vol2 15TB, vol3 1TB,vol4 150GB,vol5 100GB). Both filers have the same amount the same amount of Aggregates and volumes. Our shelves are DS4243 with 2.5 10k SAS drives. Each filer owns roughly 84 disks. Before I got here they were experiencing disk latency becuase of the volumes getting full and not autogrowing. They dont have a lot of confidence in the way the NetApp vendore they used setup their enviroemnt correclty. At my old job we presented our DBs with iSCSI LUNs using Snapdrive and SnapManager for SQL. The DBs they have here are massive in size (1-2TB). Instead of using snapmirror to DR they are doing SQL mirroring to live SQL sevrers at our DR site ( I plan on changing this to SnapMirror eventually). Our ESXi boxes are running on Cisco UCS B230 blades. All datastores are exported to ESXi via NFS. We are on ESXi 5.0. Our SQL servers in VMware use LSI Logic SCSI adapters. The SQL sevrers have 32 processors and over 200GB of RAM. This is the first time I have worked n a big SQL shop like this one. My questions are : We have PAM II cards that are 512GB in both filers but the Im not sure what the best caching method is. Does anyone know what the best caching method is for SQL intensive datastores? I see there are three different caching methods. This is the first time I have worked with PAM cards also. Will we be fine using NFS for our SQL enviroment? Should we move everything to iSCSI and use snapdrive? If so what is there to gain? I have attatched a systat and a stats show ext_cache_obj for each filer. Toaster 1 is the filer hosing the datastores for our biggest and busiest SQL Servers. Im trying to see if I can tune the Netapp for better perfomance. I am looking for any recomendations or helpful info from other NetApp guys with experience in this setup. I appreciate all the input I can get.
MS SQL on VMware won't really care overly much about how its VMDKs are stored, NFS or iSCSI. Personally I would stick with database mirroring to a DR site over SnapMirror replication to a DR site. SnapMirror is a transfer of a point-in-time application-consistent snapshot (when used with SnapManager/SnapDrive), whereas database mirroring is as close to synchronous replication as you're going to get (WAN/LAN dependent of course, with no requirement to quiesce the data as it is transaction log streaming).
Based on the cache stats (24% and 54% usage, 100% and 99% hit rate respectively, sub-millisecond read latency) I would say don't mess with the caching configuration. Based on the limited sysstat output, I would move more virtual machines over to toaster2 to balance the average load between them.