Unsolved
This post is more than 5 years old
1 Message
0
395
Help with VMAX 10K writes are slower than old Clariion CX4-120
We are running a Data Warehouse on SQL 2005, Windows 2008 Server Enterprise.
Qlogic 4Gb HBA cards. Dedicated Fiber networks to each SAN.
We see slower writes to the new VMax 10K on our SQL server. We have used SQLIO and IOMeter to run additional test and are seeing slower writes. Reads a little faster. Any Idea's? We have tried multiple test scenarios with all very similar results.
| Target | Access Specification | IOps | Read IOps | Write IOps |
|---|---|---|---|---|
| Clariion CX4-120 Drive P | 100% Write | 8057 |
0 | 8057 |
| VMAX 10K Drive Q | 100% Write |
4612 | 0 | 4612 |
| Clariion CX4-120 Drive P | 100% Read | 9130 | 9130 | 0 |
| VMAX 10K Drive Q | 100% Read | 12507 | 12507 | 0 |
| Clariion CX4-120 Drive P | 50% Read / 50% Write | 8943 | 4455 | 4487 |
| VMAX 10K Drive Q | 50% Read / 50% Write | 5628 | 2816 | 2811 |
NOTE: Testing completed with single server connected to both SANs. Drives mapped to each.
Message was edited by: kg0523
KW160
121 Posts
0
September 22nd, 2015 06:00
There are two main causes of write latency on Symmetrix:
1. Front-End Queueing: I/O comes into the FA CPUs faster than it can be processed. The typical solution for this issue is to spread the workload across more FA CPUs and/or use more meta members to open up more queuing space.
2. Write-Pending Events: This is caused by back-end CPUs and/or disks that cannot keep up with the writes that need to be flushed to disk. When the entire cache is filled with write-pending tracks the incoming writes will be slowed to disk speed. The remedy here is to open up more bandwidth on the back-end: either change to faster disk, more disk, or more DA CPUs (by way of adding engines)
You'll need to use a tool like Unisphere for VMAX to determine exactly where this bottle neck is occurring. It may help to engage EMC support to assist.