When you initially set up a Delphix Engine, there is a one-time opportunity (and requirement) to run your storage performance tests on unconfigured storage. Because of the test method we use, we cannot run it again in the same way after the engine is set up. When run from a Windows target host, DiskSpd can help establish a performance baseline that combines the network and storage performance over iSCSI. This can be a significant help in determining whether your current or future performance is within normal expectations. Similarly, you can use fio on the *nix side from a target system to establish a performance baseline that combines the network and storage performance over NFS.
Troubleshooting and information gathering questions
These questions are presented roughly in order of priority. The answers to these questions should help narrow down possible causes of poor performance.
- All storage traffic for virtualization Target hosts goes over the network. Have you reviewed Network Performance Expectations and Troubleshooting?
- Were all of the storage best practices been applied?
- In particular, check the IO Operation limit and round-robin settings.
- Was a baseline ever created for this host using DiskSpd or fio? (How does our current performance compare?)
Gather storage specs from VM/storage administrator:
- Vendor, Model (EMC VMAX 40k, HP 3PAR StoreServ 7000)
- IO latency SLO (e.g. 5ms 99.999%)
- IOPS/GB SLO (e.g. 0.68 IOPS/GB for EMC Gold-1)
- Cache type and size (e.g. FAST cache 768GB)
- Tier, #Pools; if auto – tiering; relocation schedule (e.g. Gold/Silver/Auto/3 pools/etc)
- Pool detail: (#) drives, RPM, Type (e.g. Pool1: (20) EFD, (30) 15k SAS, Pool 2: (40) 10k SATA)
- Connection (XXGb Fibre Channel)
Dedicated or Shared pool (how many applications/servers)
If you need further help, please contact Delphix Support or Professional Services to assist in getting the best performance possible from your environment.