Performance decrease with SRDF/S activated.

Question

Performance decrease with SRDF/S activated... -Distance between R1 site and R2 site is only 15KM. We have restored a 200GB Database on R1 volumes with SRDF/S active (Synchronized) and it took 5 hours. The exact same restore on the same system with SRDF/S in Split mode took only 2h30. I understand that Synchronous replication will always add some latency on IO/Writes but I don't undersand why it is so huge with a distance of only 15KM. Could it be related to unsuficient cache, incompatible microcode version? Anybody would see an explanation or it is a normal behaviour? R1 Symm Microcode: 5772.100  R2 Symm Microcode: 5773.138 Regards

Booyah2 · Accepted Answer

I would think this relates more to normal behavior for a database restore. There are a few things that you may want to look into though.

1. Rather than splitting the the RDF devices, try putting just the database devices into Adaptive Disk Copy mode during the Restore.

2. There is a new feature introduced at 5772 called Single Round Trip (SiRT). This may be an option to speed things up.

SRDF with SiRT

SiRT (Single RoundTrip) for Fibre Channel SRDF directors (RFs) was

introduced in Enginuity 5772 for SRDF/S mode only. It is

dynamically enabled for SRDF/S links > 12 Km for block sizes up to

32K in Enginuity 5773 code. SiRT is compatible with fast write/write

acceleration switches and extenders, as it will measure link latency

and disable automatically if connected to these devices. As a best

practice, it is recommended that either the EMC SiRT feature or the

third-party fast write feature should be used. Both should not be

enabled simultaneously.

The Fibre Channel SiRT feature for the Fibre Channel director can be

set to

Off or Automatic. When set to Automatic, this feature will only

accelerate write I/Os using criteria based on latency and I/O size.

Note:
EMC recommends contacting your EMC Customer Service
Representative to verify that the setting is enabled if required in your
environment.

3. You may also need to investigate the size / utilization of your pipe between R1 and R2.

4. I do not think this has anything to do with the amount of Cache installed in the array.

5. I do recommend upgrading your microcode to the latest EMC Target levels.

Best Regards,

TBM

nicoroy123_15c2b9 · Answer

Gentlemen, thank you so much for these fast answers. I'll look deeper into Adaptive Copy as a potential solution our scenario requirements. Regards

Quincy561 · Answer

I agree that going into adaptive copy for the restore operation is a good suggestion.

If the latency to write to the R1 cache is say 300 micro seconds, then waiting to also write to the R2 cache will add a minimum of another 300 micro seconds. There is also the distance and some overhead to make the handoff to the RA. Make sense?

JasonBailey · Answer

this is probably due to SRDF/S write serialization

Symmetrix

Was this post helpful?