MSA1000 / MSA1500 : poor performance after failover

[wolfgang_12]

hi!

the same problem with a MSA1000 (last firmware) und a MSA1500 (last but one firmware). everythings works fine. good performance and no problems. until a failover happens. when the second controller takes controll, the performance is half of the primary controller. at the MSA1500-site it even happens that when the left controller was active with good performance and a failover activated the right one, that performance was poor. so it shouldn´t be a hardware problem with one controller.

after a manual failback, the privious "fast" controller is fast again.

connected to the MSA1000 are 4 windows 2003 server SP1 with basic MPIO 1.1 für MSA1000. when the MSA1000 is slow, every server has poor disk performace on MSA-LUNs.

the MSA1500 has windows and ESX connections. windows with MPIO 1.1. the same poor performance on every machine when the MSA1500 is slow.

the performance is the only problem. the failover itself works fine. every first machine hba is connected to the first switch, the second to the second.

every help and ideas are welcome!

thank you!
wolfgang

[Basil Vizgin]

May be second HBA inserted in slow PCI slot?
Or some loss on second switch?

[Joshua Scott]

MPIO is multi-pathing software, so when you failover you are actually getting only half of the throughput that you would when both controllers are active. Therefore I would suggest that this is normal behavior.

Josh

[Uwe Zessin]

On the current MSA1000/1500 firmware, there is only one controller doing I/O anyway!

[John Kufrovich]

Wolfgang,

Sounds like the controller cache is disabled on the failover controller.

When the failover occurs, attach the MSA cli cable to the suspect controller and issue the commands.

>show cacheinfo - Cycle this cmd a few times

>Show taskstats - Cycle this cmd a few times.

Provide the readouts.

[wolfgang_12]

MSA1000/1500 are currently just active/passive, so half the I/O in case of a failover isnt´t for sure the normal behaviour.

according to ACU the cache is active. the poor performance is during read and write operation. in case of slow write operations, non active cache could be the problem. i dont think that no read-cache slows down read operation to half the performance. but i will check, thank you for the tip.

[John Kufrovich]

It isn't normal behaviour.

By using those commands you should be see if the cache is locked or there is activity.
After failover, the new controller should have flushed the cache to the drives.