EMC Clarion 4500 performance

[Michael Ball]

I am using a 4500 with Brocade switches to three HP L3000 boxes. This is for PeopleSoft implementation. We currently have 5 Oracle 8.1.7 databases for peoplesoft with an average of 40 GB. My performance sucks! I have 40 luns on 3 drive groups with the majority on one Raid 5 Group (30 luns). I am having 50-70 % iowait on the cpu and 80-100 io/sec with avg service times of 150ms. Has anyone experienced these types of problems before with the EMC 4500? EMC wants to add more disks to spread the luns, but with 36GB drives I already have adequate space. Any other suggestions.

[Bill McNAMARA_1]

They want more spindles not space.
Adding more disk to raid5 is not so cool but creating R1/0 would be a first step.

You might want to look at your lvm configuration and try to access the emc via a bus/loop of it's own.
And try to spread lun access across controller inputs to the array.. although your problem really is that old clariion lic technology.

Later,
Bill

[Vincent Fleming]

There can really be only a couple of sources of your performance problem... the Clariion controllers are the bottleneck or the drives themselves.

To help determine which it is, take the number of drives in the array, and multiply by 75 IOPS - this is about the most IOPS you will ever see with that number of drives. Then look at "sar -d", and see how many IOPS you're doing (r+w/s). If the sar output is far below the calculated number, then the controller is probably the culprit, and you need a faster (newer) disk unit. If not, (ie: the sar output is close (within 10%) of the calculated number) you can probably add more drives to increase performance.

As the previous poster noted, RAID 0/1 will also usually help performance - but only when the controller's the bottleneck.

Good luck

[Rosslyn Bradley]

Michael, You mention that you're using Brocade Switches. Are you connected using QuickLoop or Fabric? I have seen instances where QuickLoop ports can revert to half-duplex mode; this will adversely affect your performance. To check the status of your switch ports:
1. Run the following command for the suspect port to detect half-duplex:
portregshow

Look towards the end of the listing for the "loop_config" entry. If the value for loop_config has the format "0xNNNN," the port is in half-duplex.
Example: ?0x80032c04: ?loop_config 0x1001
?????
If the value for loop_config has the format "0xNNNNNN," the port is in full-duplex.
Example: 0x80032c04: ?loop_config 0x103001 ?????

2. If this is the port to which the CLARiiON SP is attached, use the following commands to reset the port, but do this ONLY if there is no I/O going to the array:
FC_Switch:admin> portdisable
FC_Switch:admin> portenable ?

3. Repeat the portregshow command to verify that the port is now in full-duplex.

NOTE: ?Running the portdisable command will effectively halt I/O on that port.If the SP switch port is disabled, I/O cannot be performed on that path to the array.?Do NOT run this command if the array is active. Instead,?quiesce I/O before performing this procedure!