I just ran the same test on a raid5 (4+1) and raid10 (3+3) pool (both with 600 GB SAS disks), and got 3480 IOPS and 109 MB/sec. The limit seems to be the 1 gig link to the VNXe 3100.

Your numbers seems very low compared to mine. Could it be a netcard driver issue or link issue (half duplex?).

/Morten

Sent from my Verizon Wireless 4GLTE smartphone

Hi,

I repeated the test with different network adapters in the vpshere environment and now  I'm getting similar results:

3509 IOPS and 111 MB/s

What I don't understand is if the 111MB/s limit is coming from the VNXe3100 performance or some multipathing problem as we have multipath round robin enabled to the vnxe storage.

Having 2 nics on the esxi host and 2 eth ports for the iscsi server with 2 different ip.

In vsphere client i correctly see al paths and the state to be "active I/O" but never going beyond 111MB/s.

Hi studionc,

Can you expand on what you mean by tried different network adapters?

I have a dell R710 running ESXi 4.1 and tried both the on board nic and a pci intel nic for NFS/vmkernel traffic..

I am getting the same performance results that you were getting originaly:

arround 1500 IOps, 50MBs, 34ms

No matter what I try, I can not get better results that this on the 100% read test..

It seems you were able to solve this, any suggestions you have would be most appreciated.

Has this document been released yet? if so where is it?

I have a vnxe 3100 with 12 600GB sas drives and 2 service processors.

My initial configuration was 10 drives configured in one pool, 2 x 4+1 Raid5, with 2 hot spares.

I wanted to try a configuration with a Raid10 pool and also wanted to set my vnxe back to the base image.

To do this, I actually went to Settings > Service System.  Put each processor in service mode, then chose reinitialize under the Storage System options.  This process to about 2 hours to complete.

I had to go thorugh the setup process again.  This time I created the first storage pool with 5 drives, 1 Raid 4+1 group.

Which left 7 drives available.  I configured a second pool using 6 drives in a Raid10 3+3 group and then a global hot spare.

I was told in one of my sales meetings a while back that the first 5 drives may have to be configured in a Raid5 pool.  You can then add additional 4+1 Raid5 sets to the initial Raid5 pool or configure remaining disks as you want. But for Raid10, you need 3+3 sets and for Raid5 you need sets of 4+1.

I was told in one of my sales meetings a while back that the first 5  drives may have to be configured in a Raid5 pool.  You can then add  additional 4+1 Raid5 sets to the initial Raid5 pool or configure  remaining disks as you want. But for Raid10, you need 3+3 sets and for  Raid5 you need sets of 4+1.

This is not true. You can have the first four drives in the system in whatever pool you want to have it in. The only thing that you will notice is that we use some space (~48GB per drive) from the first four drives in the system for storing log files, etc. You will this space not available for use in all the drives in the pool.

Avi,

Where the first 4 drives have system or log data on them, is there any performance impact from using them in a pool with other disk packs?

Hi Avi, thank you for clearing that up.  I was definitly told that the first 5 drives had to be configured in Raid5 by an EMC representative.

But I did notice that when creating the first pool, I did have the option of creating a pool of 12 drives in Raid10, with 2 3+3 groups.

The remaining portion of the first four drives in the system (or system drives) are usable as regular user storage space, but with a slightly reduced performance when compared to using non-system drives. It is recommended that for very performance sensitive data, you should try to use up storage space from non-system drives in the storage system. However, for a majority of the applications, the system drives may be included and used just like any other non-system drive with only minor amount of performance difference.

Performance testing with some typical random iSCSI 4K read and write mix of IOs typically shows the raid group on the system luns performs within 3-5% of performance of non-system drive Raid Groups.   Performance in other test configurations and scenarios may vary but we expect generally similar amount of impact in most cases.