A new VM enviroment I am starting of with IBM 4700 16 drive 146 10k FC drives. Is 16 spindles going to be enough for 30 decently hit servers.
Please give some feedback I need to know if I need to purchase more!
Depends on what you consider to be a decent hit server.
Are you using 810 cabinets with that?
In a raid 5 setup, perhaps a couple of luns, I would think you easily support 30 moderate IO guests...but there's a lot of gray area in what you are asking. We don't know the IO level you require, if you're using more than one cabinet, how you are connected, dual controller with lots of cache?....etc. Too many if's in this one.
I have a 2 node IBM SVC cluster. The 4700 has a 810 SATA drive cabinet connected. There is 2 controllers in the 4700 and only one FC cable from each controller A and B is plugged into a FC switch. I have 7 other various IBM 4000 series units all behind the SVC. I recently adopted the storage side of things,I am mostly networking. I am really liking the SAN storage technology but really wish I knew more. I do use the IBM storage manager client and there is the option for "monitor performance" but unfortunately I am not that well in reading the ouput. I really want a better graps on the storage. I will get there it just takes time.
You seem knowledgable on the storage side. Any tips to get there faster? Books, websites? Just would like to speak with an expert.
Get back to basics. 10k drives offer about 130 IOPS, so 16 drives is about 1300 IOPS max if served directly from disk. Sure, your cache will help, but VMs will produce random, non-sequential reads and writes - cache isn't great with this. Even if you assume a 40% hit rate, thats a best case scenario of 1820 IOPS.
Using averages, that means each VM will only have 60 or so IOPS - not a lot. Assume again at least 1 or 2 of those 30 VMs do more than average and you have a problem.
You really should measure it to be sure, unless you have a strategy to mitigate the risk another way (roll back, small SLAs etc).
Also consider the other risks, i.e. 16 disks in a single RAID-5 array exposes you to multiple drive failures. Splitting them into 2 8 disk arrays will reduce this risk, but reduce your capacity, but more importantly isolate your performance into 2 silos.
Read the San Guide from Vmware, has great info for you.
Dave
One other thing to keep in mind here is that you have several layers of cache. An SVC controller typically has 8GB cache per controller and the DS4700 also has 1GB cache per controller. Depending on the workload and request rate this might help a bit. For this to work as optimally as possible it's important that the workload is distributen evenly among the SAN and SVC controllers. For best possible performance you should also check the CacheHitPercent and IO/s stats from your SAN controllers and compare them to see that the workload is evenly distrubuted.
One thing that you'll have to be carefull about is your SATA disks. SATA disks have typically 1/3 of the performance of a SCSI disk. This might be enough for light workloads, but my I have seen SATA drives hit the roof much more sudden than SAS/SCSI disks. And once you hit the roof, everything on that array will become syrup.
What Dave suggested above are also very good points.
Lars
SATA disks will sustain about 80 IOPS, so you could factor that into the above equation.
That so, because SATA usually has more capacity, you will fill it - therefore less performance per GB needs to be considered.
Dave
