[Veritas-vx] Best dmp partitionsize for the XP1024 array

2007-06-07 Thread Johnny Tran

Hi

I have been lurking for a short time and this is my first question for all
you gurus :)

We have a XP1024 array (rebadged HDS 9980V I believe) connected via 4 2GB
HBA's to a HPUX 11.23 PA-RISC running Veritas Storage Foundation 4.1 for
Oracle.

I am looking the recommended partitionsize to use when using the default
'iopolicy' of balanced. The default partitionsize is 1024 (kernel parameter
dmp_pathswitch_blks_shift=10).

I've heard the when the path switches frequently, the XP1024 has
considerable overhead to manage internally to keep track of IO's in the
cache etc...

http://docs.hp.com/en/5187-1370/ch01s04.html - this document shows the
optimal setting for a HDS 9960V is 14 to 16 (16384 to 65536). It has been
speculated that the optimal value for the XP1024 is 17 (2^17 = 131072).

Changing the kernel parameter parameter dmp_pathswitch_blks_shift requires a
reboot, but I understand that this can be changed on-the-fly with 'vxdmpadm
setattr enclosure XP10240 iopolicy=balanced partitionsize=131072'

We are also considering the difference in iopolicy, such as 'adaptive' and
whether or not adaptive will use the partitionsize parameter or is it only
relevant to the 'balanced' iopolicy?

Thanks
Johnny
___
Veritas-vx maillist  -  Veritas-vx@mailman.eng.auburn.edu
http://mailman.eng.auburn.edu/mailman/listinfo/veritas-vx


Re: [Veritas-vx] Best dmp partitionsize for the XP1024 array

2007-06-08 Thread Engle, Victor
Hi Johnny,
 
In my previous job we were in the same situation and were told by HP to
avoid vx dmp for the reasons you mentioned; i.e. the caching algorithm
did not handle the situation where I/O was switching from cluster to
cluster. I/O being serviced from CL1A and CL1B was ok but I/O being
served from CL1A and CL2A caused performance issues within the array.
This particular site was overly sensitive to performance metrics.
 
Before veritas we had used LVM and pvlinks to manually balance I/O
across the HBA's and only switched to veritas because we encountered
some limitations with LVM. The solution for us was to manually balance
luns with veritas. I worked for the storage team but worked closely with
the unix admin team and I know they scripted the manual balancing and it
worked beautifully.
 
Regards,
Vic
 



From: Johnny Tran [mailto:[EMAIL PROTECTED] 
Sent: Thursday, June 07, 2007 10:15 PM
To: veritas-vx@mailman.eng.auburn.edu
Subject: [Veritas-vx] Best dmp partitionsize for the XP1024 array


Hi

I have been lurking for a short time and this is my first question for
all you gurus :)

We have a XP1024 array (rebadged HDS 9980V I believe) connected via 4
2GB HBA's to a HPUX 11.23 PA-RISC running Veritas Storage Foundation 4.1
for Oracle.

I am looking the recommended partitionsize to use when using the default
'iopolicy' of balanced. The default partitionsize is 1024 (kernel
parameter dmp_pathswitch_blks_shift=10).

I've heard the when the path switches frequently, the XP1024 has
considerable overhead to manage internally to keep track of IO's in the
cache etc... 

http://docs.hp.com/en/5187-1370/ch01s04.html - this document shows the
optimal setting for a HDS 9960V is 14 to 16 (16384 to 65536). It has
been speculated that the optimal value for the XP1024 is 17 (2^17 =
131072). 

Changing the kernel parameter parameter dmp_pathswitch_blks_shift
requires a reboot, but I understand that this can be changed on-the-fly
with 'vxdmpadm setattr enclosure XP10240 iopolicy=balanced
partitionsize=131072' 

We are also considering the difference in iopolicy, such as 'adaptive'
and whether or not adaptive will use the partitionsize parameter or is
it only relevant to the 'balanced' iopolicy?

Thanks 
Johnny

___
Veritas-vx maillist  -  Veritas-vx@mailman.eng.auburn.edu
http://mailman.eng.auburn.edu/mailman/listinfo/veritas-vx