Storevirtual VSA performance loss

[TS1984]

Hi,

I'm new to the HP storevirtual VSA software. I'm have some experience with the P4300 G2 nodes but now we are trying a new setup with 2 HPE G9 servers filled with SSD drives. We have also bought 2 HP VSA 10TB licenses.

We are now experiencing some performance issues with the setup.

We have installed vmware ESX6 on both hosts. Created 2 datastores (on for the VSA OS and one for the LUN). For the test I have created a vm on the datastore where the LUN will reside. A iometer run gives me 91K IOPS with 100% read only and 70K IOPS with a 65% random write (real life test).

Now we have installed VSA on both servers, created a management group, cluster and volume. When I run a vm on that volume iometer gives me max 30K IOPS with the 100% read test and 20K IOPS with the 65% random test.

We have anticipated some performance loss because of the extra virtualization layer but not this much. Does anybody have an idea how much performance loss we can expect? Currently we are still testing different setups (RDM's vs VMFS disk etc)

[oikjn]

Did you try a single VSA in its own cluster?  Network capacity can become an issue and even more with the virtual network stack.  Also, what does the CMC perfmon show?  How is the CPU usage on the VSA during these tests?  What does the IO latency show in CMC as well?

There is definitely a performance drop over DAS SSDs.  You also have to keep in mind that the VSA storage subsystem stripe is 256k, so if you are testing 4k random access and using a 64k cluster size on the SSD volume against a 256k base under a 64k formatted LUN, you are going to have performance overhead simply because of the filesystem storage.