HPE Storage Users Group

Currently our 22 esx hosts cluster is mixed with other hosts on the 3par T400 4 nodes host ports. No issues have surfaced for IO latency so far. As our cluster grows and more P2Vs are under way, is it worth the effort to dedicate host ports just for esx cluster use?


Thanks
-J

With 4 nodes, you should have 16 ports total? I started with just using 8 and ran into concerns, and ended up plugging in all 16. We share AIX, Windows, and ESX (also a large cluster) all on the same 16 ports. Each physical box has 2 HBA ports and is zoned to 2 different nodes, thus each physical host sees all 4 nodes. We intentionally spread our ESX hosts evenly across all 16 ports.

I divided my 16 ports into 4 "sets". So a host will connect to "set 1" which means 2 ports on the A-side of the SAN and 2 ports on the B-Side. The aliases for the ports in the SAN show their set name so I can easily provision a host to a set.

The 3PAR Gui shows you your # hosts per port on the hosts section, by clicking on the T400... not drilling down under it. Thats how I tell which set to use next for the next provisioning.

It is best practice to isolate your Netapp V-series devices to their own ports, since they do not balance their traffic well and are very chatty.

Windows 2008 and ESX MPIO settings posed a challenge to us, meaning we couldn't ensure our windows/esx admins would reliably set the policy to round robin. We had 2 hosts having perfo issues (out of 100s) due to those 2 servers set to active/passive failover and both being active on the same port. We ended up deploying SAN Screen simply for its ability to alert/report on IO port balance per host based on metrics gathered from the switches.

I hope this helps.

Thanks Richard!

Our setup is similar to yours. Different OS and Apps mixed throughout host ports.
Any concern about Oracle/SQL boxes being on the same port as ESX clusters?

No special concerns about sharing anymore (other than the Netapp mentioned above). On an older version of Inserv (2.24) we had to set the personas per port instead of per host... which lead to 8 ports for AIX and 8 for Windows/ESX... but since 2.3x and the ability to set personas on a per host basis... we've taken the same approach to ports as the 3par takes to drives... distribute them widely.

I have alerts setup in System Reporter for when any of my host ports exceed either an IO/s threshold or a MB/s threshold. I also have a weekly scheduled report setup in system reporter to capture a weeks worth of host port performance. Since we manually balance our hosts (with zoning) among the 16 total ports, we keep an eye on the results... and adjust if needed. So far we have only used the data to decide where to put the next new host, and have not had a need to re-arrange existing ones.

Good to hear. Thanks for sharing!

Getting a new V400 sometimes this year and hopefully with some SSDs. Very excited!