I just updated one of our hosts to 6.5U3. I have a mix of EqualLogic and Compellent iSCSI datastores. Exiting maintenance mode resulted in a hang, and the process never finished. There are several errors in the vmkernel/hostd logs that point to the EqualLogic datastores, and specifically the MEM module:
2019-07-25T20:04:00.468Z info hostd[104C2B70] [Originator@6876 sub=Vimsvc.ha-eventmgr] Event 201 : Path redundancy to storage device naa.60fff18a0e80b172c3be25cfdd6eedb8 degraded. Path vmhba64:C0:T17:L0 is down. Affected datastores: eql-ssd-03. 2019-07-25T20:04:00.755Z info hostd[104C2B70] [Originator@6876 sub=Vimsvc.ha-eventmgr] Event 202 : Path redundancy to storage device naa.6090a0c840d7b406555315b4900520de degraded. Path vmhba64:C0:T3:L0 is down. Affected datastores: prod-r10-esxi-02. ---
Hello, If you are running mixed Dell PS Series and SC series SANs on same ESXi nodes, then only native MPIO should be used. How many PS arrays are in that group and how many in a pool? There isn't enough info here to offer much help. I would also look for any messages on the PS array at the time of these messages. Connections get torn down and rebuilt periodically to optimize performance. Especially in multi-member pools. But I would remove MEM and change to Round Robin and change default of 1000 IOs per path to 3. This doc describe how to set that and other best practices https://downloads.dell.com/solutions/storage-solution-resources/BestPracticesWithPSseries-VMware%28TR1091%29.pdf Regards, Don
I have two groups. One is single-member, the other has 5 members with 3 pools. The SC array is setup with two fault domains, but only one subnet, and that /24 subnet is shared with the EqualLogic arrays as well. (this config was required for PS coexistance). I wasn't aware that we should not be using MEM. I did reboot that host and seems like things have started functioning, but if I should not use MEM along with the SC arrays in the picture, I will remove it and use native MPIO similar to compellent. At this point I don't trust the host to let it run like this.
Yes MEM sets a lower IOs per path value. It's based on a much different algorithm than simply switching paths every so often. It uses the paths to send or receive data from the member having that data when you have multimember pools. But even single member pools are optimized automatically with MEM
Do you have to do anything after just installing MEM?
Build wise I usually install ESXi, create the vSwitch and PortGroups and configure jumbo frames, assign the PortGroups to the iSCSI software initiator and then join vCenter, update the host, install MEM and iSM and migrate to a dvSwitch and join a cluster.
I've never done anything more than install MEM and noticed it changes to the DELL_PSP_EQL_ROUTED policy and I call the job complete.
Using VMW_PSP_RR with the recommended configuration the VMs actually feel faster, now I feel like I might have been missing a step.
dwilliam62
6 Operator
•
1516 Posts
•
6374 Points
2234
0
Posted July 25th, 2019 19:00