UNSOLVED
Why would SRDF/A impact performance?
All:
When I run a symstat command, I can see that my available cache is only 1-2% full. But when I run an IOMeter on a device that's in an SRDF/A pair, I get about 1/2 the IOPS versus the same type and size device without SRDF. If SRDF/A is only capturing a 15 second write cycle, and then sending that over the wire, why would it slow me down that much?
There is nothing else in the device or RDF group. Just the one SRDF/A pair.
Thanks!
Brandon
AshokaReddy
1 Rookie
•
11 Posts
538
0
Posted April 23rd, 2012 08:00
Hi Brendon,
Please Ignore My previous reply. That was intended for SRDF/S.
Regards
Ashoka Reddysauravrohilla
859 Posts
538
0
Posted April 24th, 2012 00:00
Hi Brendon,
Are those devices concatenation thin meta ? Have you enabled device pacing? Open a ticket with support to check the devices from backend.
Regards,
Saurabh
Saurabh:
They're striped. No write pacing was enabled.
I had my local guys on it, along with some performance guy they opened a ticket with.
Their response was that I was only testing with a single LUN and a single host. They said this wasn't indicative of what the box would be doing when it had more hosts, and more devices. They asked me to create 32 devices on multiple hosts, and retry the test.
I'm not sure what that would prove. Regardless, I don't have the time to do that, so I will just be running my SRDF/A workloads at 50% the speed of non-replicated ones.
Thanks for trying.
AshokaReddy
1 Rookie
•
11 Posts
538
0
Posted April 23rd, 2012 07:00
Hello Brandon, What is the latency for the device pair?. In SRDF/A the host writes will wait for the acknowledgement from the R2 symm before accepting new writes. The latency will increase depends on the distance between the arrays. Regards Ashoka Reddy