Storage issues

[Andres_13]

Hi all,

I have a L1000 Box with a 12H attached, recently i try to connect an aditional 12H to the same box and the server's performance go down.

Both 12H have an X and Y controller and two SCSI card each. I wonder this configuration is valid and if this kind of box can support it and where i can find docs about this issue.

More detailed info can be post this way if any of you need it.

Thanks & regards!

[Leif Halvarsson_2 · Accepted solution]

Hi,

What kind of performance go down ? Sounds strange that just connecting a disk system should affect performance. Is both systems configured and in use ?

[A. Clay Stephenson]

There are two common errors made when configuring 12H's and either can lead to very poor performance:

1) Having all the I/O go through 1 controller (usually X) and controller Y is there only for failover.

2) Using too much space as LUN's so that the AutoRAID is almost purely RAID5.

The fixes are:
For every VG, create 2 equally sized LUN's. LUN A should use Controller X as primary path (alternate y) and LUN B should use Controller Y as primary path (alternate X).
Stripe each LVOL in the VG across both LUN's
usually in 64KB chunks. There is no performance gain from increasing the number of LUN's per VG to above 2 although host-based performance monitoring tools will make it appear that you have an i/o bottleneck.

Allocate no more than about 50% of the total array space as LUN's. This will leave the array in RAID 1/0 at all times.

[Alessandro Pilati]

Andres,
let's start posting the sar.report file after launched these two commands:

sar -o sarfile 3 30
sar -Af sarfile > sar.report

Then post the output of
swapinfo -tam
vmstat 5 10

Rgds,
Alex

[Torsten.]

Just a question:
Did you connect the arrays using 4 SCSI controllers in your server? Did you configure their ID properly? A. Clay Stephensons reply gives you the best instruction to setup for optimal performance.

[Andres_13]

Thanks all of you,

May be i wasn't say right way, this is what i did:

* One server with one 12H (12H0), shutdown the server, connect another 12H (12H1), startup the server, ioscan -fnC disk and the server "see" all disks, pvcreate to all 12H1 disks, and then when i try to setup a VG i can't add the alternate path to 12H1 disks, server go to slowly after this i can use all disks.

* When i try to resore any thing to these disks who belongs to a vg that resides on a 12H1 disks, server comes very veery slow.

Any way all requested info is in attached file,

Thanks & regards

[Alessandro Pilati]

Andres,
the only not-normal thing I see in your reports is a low average %wcache in you sar.report file: in a time you can,you should increase the buffer cache ( increase the BUFPAGES parameter and I think is necessary the regeneration of the kernel ).

Anyway, maybe it's more effective an analysis ( launch commands above )during the restore of disks in the 2H1. I think anyway that there is a problems with the communication between the two 12H.

Wait for the hints of some more experienced forumer about it.

Rgds,
Alex

[Bob_Vance]

Are you daisy-chaining the 12H's?
Or do you actually have 4 SCSI controllers in the server?

If you daisy chain the Arrays, you must change the controller IDs on the second Array (they default to 0 and 1).

hth
bv

[Andres_13]

...