Recently I was helping a customer architect a solution and I went back to the Best Practices for Vpshere and the EVA you can find it here: http://h20195.www2.hp.com/V2/GetPDF.aspx/4AA1-2185ENW.pdf
I thought I would make a quick summation and comment as some of it seems neglected in terms of being updated.
- When creating hosts in commandview create them as Type: VMware
- Use a single disk group with the same size disks and the largest possible disk group
- Single Diskgroup sparing is sufficent unless MTTR is > 7 days
- Don’t use VRaid 6 – Vraid 5 provides adequate redundancy
- When using disks of varying performance characteristics still use a single diskgroup
- Balance your VMFS Luns across both controllers for ownership with alternating failback and failover policies read more here (http://www.ivobeerens.nl/?p=465)
- Configure Round Robin advanced parameters to IOPS=1….I dont think HP is right here this was fixed fixed in Vpshere 4 update 1; This shouldnt be necessary
- When using DR groups, ensure that DR groups managing controllers are spread across controllers
- Use less than 16 VMDKs per VMFS datastore
- Use the same lunid per host per lun
- Create VMFS datastores through Vcenter for proper alignment
- For heavy i/o load vms use the paravirtualized virtualized virtual adapters for the VM data Luns
- Verify data drives for windows 2003 machines within os for proper alignment
- Set Round Robin multipathing on your hosts and then reboot; dont forget to set your Microsoft clusters to MRU here is the quick and easy command:
esxcli nmp satp setdefaultpsp --satp VMW_SATP_ALUA --psp VMW_PSP_RR
on 
on 
Twitter
RSS
LinkedIn