Monday, May 20, 2013

IBM Power AIX and HDS USP-V/HP XP 24000 - Which FEDs and ports for my Oracle hdisks?

The HDS USP-V is the same underlying storage hardware as the HP XP24000.  I've done quite a bit of work on systems using this storage.  I've never been the captain, always the boatswain. :)

A well-designed system using one of these arrays can provide a high level of predictable performance.  Front end director design makes these arrays more susceptible to front end congestion than many other types of storage, so high performance/high traffic databases require thoughtful planning of which LUNs will share front end directors and ports on the directors.  Its not just a matter of data throughput, IOPs count, or port queue depth.  As the front end director microprocessors become increasingly CPU busy, they introduce increasing latency and front end congestion.  This is often overlooked in performance investigations, in large part because its not easy to correlate microprocessor CPU busy to the added latency for any given workload.

If you are observing a system and see high read AND write latency from the server host, and the array is reporting a low level of write cache utilization and good response time from front end director through to disk media, bottleneck on the front end port queue waiting to get onto the microprocessor is a reasonable suspect.  (Its important to note that write cache utilization level is important to evaluate here - there are various cache thresholds in these arrays that govern behavior with respect to read and write priority as well as a switch that determines whether write cache utilization level has cache partition or global effect on the array.  But, in general, cache utilization performance effects are easier to evaluate on these arrays than the effects of front end director microprocessor CPU busy.)

So, when I evaluate an Oracle on AIX system with HDS USP-V/HP XP24000 storage that I am seeing for the very first time... if I see high latency in iostat results, and ESPECIALLY if I see QFULL conditions, I want to retrieve information as quickly as possible about which LUNs from this server are co-resident on FEDs, and on FED ports.

I ran across this over the weekend -HP documentation suggests that "lscfg -vl hdiskX" will indicate the port and controller for the hdisk device.

In fact, I think that "lscfg -l hdiskX" will indicate the port and controller. 
[not-yeti@sasquatch] $ lscfg -l hdisk33
  hdisk33           U756D.001.DQD56YK-P1-C2-T2-W50060E801530A650-L21000000000000  XP MPIO Disk XP24000 (Fibre)

I think this means that hdisk33 is connected to port1 of controller 2.  I assume each controller is an FED?

As I mentioned, I'm never the captain - always the boatswain.  I'll have to search for confirmation.

Something tells me that Mark Duszyk or a member of his crew will be able to confirm or deny.  I had thought I saw a blog post on this very topic at his excellent AIX and Linux blog  
But when I looked... I couldn't find it.

****Updated 5/20/2013 by sql-sasquatch****

The Ldev and port number are specified in Z1 of "lscfg -vl hdiskX".  Ldev 2006, port 6A in the example below.

[not-yeti@sasquatch] $ lscfg -vl hdisk33
  hdisk33          U756D.001.DQD56YK-P1-C2-T1-W50060E80
1530A650-L21000000000000  XP MPIO Disk XP24000 (Fibre)

        Machine Type and Model......OPEN-V
        Part Number.................
        ROS Level and ID............36303038
        Serial Number...............50 130A6
        EC Level....................
        FRU Number..................
        Device Specific.(Z0)........00000332CF000002
        Device Specific.(Z1)........2006 6A ....
        Device Specific.(Z2).........
        Device Specific.(Z3).........
        Device Specific.(Z4)..........A.
        Device Specific.(Z5)........
        Device Specific.(Z6)........

Also should be able to decode it from the WWN, but it seems like sometimes the WWN and Z1 fields don't quite line up.  Port CL6-A on the system example occurs with WWN W50060E801530A650 (I'd expect those final two characters based on the translation tables at the locations below).  CL6-A also shows up with WWN  W50060E801530A640 on the same system, which should decode to CL5-A.

The other odd thing is that I'd expect the 01530A6 to be 05130A6 to translate to 05:130A6, 05 for XP24000 and 130A6 as the serial number.


****End update****

1 comment:

  1. yes, the drawer the device (disk) is installed, its slot and so forth can be delivered from the output of the "lscfg" command. Next, execute the same command against each controller or just do "lsdev" and find the matching SCSI/FC controller / adapter

    by the way, thanks for the kind world
    the crew of "one" wmdszyk,com (markd:-))