Somehow I've missed that it is recommended to make many LUN's for better performance.
I read some blogs, but can anyone give me recommendations what's the best way?
Your storage performances will depend on your underlying storage. CSV is just a mount point to your SAN volumes. To boost your storage, you have to look how to get best storage performances :
- Disks type: sata, ssd...
- controllers type : sata, sas, fibre...
- connectivity to your SAN : FC, iSCSI, SAS...
- your storage configuration : RAID type, cache, voumes number on the same array
Here an example :
- I have a SAN with 8 disks
- i can create one array of 8 disks using RAID5 OR 2 arrays of raid5 using 4 disks
- now i can create a single volume on my first array or 2 volumes i my 2 arrays
- in a csv perspective, i will have one large CSV in the first case, and two CSVs in the second case
- Case 1 : All my VMs will run into a same CSV (same volume in the SAN, same array, concurrent writes/reads), if VM1 access the storage, VM10 will be affected
- Case2 : VMs will be spreaded between CSV1 and CSV2, that means between Array1 and Array2, that eans different physical disks. If VM1 access the storage, VM10 will be not affected ( of course it maybe affected because they use the same links, the same controllers, but we speak about concurrent access to the disk)
So in general, depending on your storage, needs, budget, you will build your storage and CSV design.
As with anything in a cluster, I like to have at least two of everything. So I would configure the cluster with two different LUNs configured as CSV. If one LUN is lost for some reason, I still have the other operational. Then I try to have each CSV owned by a different node in the cluster. By having each CSV owned by a different node in the cluster, the metadata IO is spread to the multiple owning nodes instead of concentrated on a single node. Generally this is low volume, but it does spread the IO. And, it lessens redirected IO if there is a connection problem to the storage. Another rare occurrence.
It is not likely to make any noticeable difference in performance. Generally the VNX is going to be performing most of the IO optimization, and each node is Hyper-V host is doing direct IO to its own VMs. So keeping the VM IO utilization balanced is important.
But, a single LUN/CSV works, as you have found out.
Thanks everyone for your opinions!
"It should give a little performance boost compering to one big LUN"
Not necessarily. Depends on how you are creating those LUNs. If both LUNs are carved out of the same storage pool, it won't make any difference.
Again, only if you trying to set a benchmark record do you need to go to a lot of work. Contact EMC and ask their recommendation. I work with EMC a lot. We often set up very large (>5 TB) LUNs for storing VMs. It's just that the storage pool is spread over many disks. One storage pool created from 20 disks is likely to give you better performance that two LUNs created from two storage pools of 5 disks. But, for the majority of that tasks performed, you are not likely to notice any difference.
I have only one Storage pool on EMC and all of my LUN's are created on it.
I set up Operations Manager monitoring for CSV/LUN usage (reads/writes-latency). I can post the graph here by end of week.
Here are write latency results: