Managers,
Thanks to the following for your responses:
Alan Rollow - Dr. File System's Home for Wayword Inodes
Peter.Braack_at_degussa.de
Keith MCCABE
1) I decided to go with the default 128 sector stripe width.
2) I followed Peter's recommendation & it worked fine.
The responses & original question are below
============================================================================
=====
Alan Rollow - Dr. File System's Home for Wayword Inodes
--------------------------------------------------------
For the right I/O loads, there might be better choices than the
default, but usually the difference in performance is so minor
that the default is easiest. If you application is easy to
benchmark you may want to try different values with the non-
mirrored stripe (since that is the easiest to change). What
you want to avoid is a size that causes I/O to concentrate on
a single disk. If you find that one disk of the array has a
higher load than the others, then a different chunk size may
be called for.
You also want to try and avoid sizes that will cause a significant
number of I/Os to be split across two disks. Nearly all disks are
fast enough today that the time needed for even a 64 KB transfer
is small compared to the time the hosts spends trying to figure
out how to send that transfer. While having a very large transfer
split among multiple disks can produce very high bandwidth, having
medium sized I/Os get split into evern smaller I/Os is generally
a loss. For small transfer file system loads, the 64 KB size
isn't a problem since it is a multiple of the block size and
the I/Os line up well with the boundaries.
Sequential loads will be more likely to split I/Os at a 64 KB chunk
size, since the largest likely I/O size is also 64 KB. More often
than not an I/O will be split. In this case larger chunk sizes can
increase the number of I/Os that don't split. But, make the chunk
size too large and you start having hot spots.
============================================================================
=====
Peter.Braack_at_degussa.de
-----------------------
Randy,
1) We use the default stripe width of 128 sectors and are satisfied.
2) To create one striped volume containing 3 RZ29s, in my opinion the
easiest way is to use the GUI:
Click on all the 3 disks with MB2 and choose BASIC OPERATION - VOLUME -
CREATE - STRIPED.
Calculate the volume size as follows:
int (A / C) * C * B
A: Space of a single RZ29, click with MB3 on the disk and read the info.
B: Number of drives, i.e. 3.
C: stripe width, i.e. 128
To add the mirror, simply click on all the three drives on the second
BA-box and
the new volume, then choose BASIC OPERATION - MIRROR - CREATE - STRIPE.
Good luck
Peter
============================================================================
=====
Keith MCCABE
------------
Well he LSM literature normally says to make the stripe width equal to
the number of sectors per track. In this case you would select 113
which equals 56k I think.
Cheers
Keith McCabe
Banque Paribas Capital Markets
London W1
============================================================================
=====
>Managers,
>I am preparing to use LSM to create a volume that is striped across 3
>RZ29s. This volume will be mirrored on 3 RZ29s on a second BA350 shelf.
I
>was wondering if anyone out there could answer 2 questions for me?
>1) Should I use the default stripe width of 128 sectors or is there
>something better?
>2) Should I define both plexes & then the volume, or define one plex,
>define the volume & then the 2nd plex.
>tia
[=================================
[ Randy Rodgers
[ rrodgers_at_fwi.com
[ System Administrator
[ Allen County/City of Fort Wayne
[ Information Systems (SCT)
[ (219)449-7660
[=================================
Received on Wed Apr 09 1997 - 16:03:53 NZST