Running IDS9.30 with 1 CPU VP on a virtual HP system

[Malc P]

Morning all
For operational reasons I've been into several times before, we're running a relatively old IDS on HP-UX11i, virtualised on an 8 CPU HP Integrity container. We have at present 2 IDS instances - a development area and a live-copy area for integration testing, but will be running 3rd instance for a live implementation with up to 25 concurrent users in a couple of months. We are running KAIO on all instances.
Due to a known issue with HP containers in this setup, IDS running more than one CPU VP can lead to the system hanging so we've had to restrict it. What we're finding is that under high load (many heavy queries and concurrent OLTP access) that performance tanks (and 'top' shows one CPU at 99.99% load while the other 8 are at 1% or so - as expected).

How can we best improve the performance here?

PS we have CPU noage set, "SINGLE_CPU_VP" set to "0", OPTCOMPIND set to 0 and PDQPRIORITY set to 1. We have 64 AIO VPs running (there are 28 dbspaces and 300 chunks), and the users all connect via NETTYPE ipcshm. All the databases are Buffered logging.
Also:
There are 12 LRUs
MAX and MIN DIRTY are 4 & 2
DD_HASHSIZE 67, DD_HASHMAX 15
DS_HASHSIZE 67, DS_POOLSIZE 127

TIA

Malc

[s***@t-online.de]

Hello Malc,

You state you are using KAIO why have 64 AIOVPS?? either use KAIO and set AIOVPS to say 2 or 3 or switch off KAIO. and use 64 AIOVPS.

I would maximize memory and set RESIDENT to -2 so Buffers, SHM and message portion remains resident. Also in the kernel set dbc_max_pct to a normal value
eq the % of memory you would dedicate to the filesystem cache may be 5 % or so.


if one cpu vp why "SINGLE_CPU_VP" set to "0", not 1??

Also play with multiprocessor!!.
Dependant on how much info you may want to loose whenever you experience a crash i would also crank up log buf. and physbuf.


See you

Superboer.

Post by Malc P
Morning all
For operational reasons I've been into several times before, we're running a relatively old IDS on HP-UX11i, virtualised on an 8 CPU HP Integrity container. We have at present 2 IDS instances - a development area and a live-copy area for integration testing, but will be running 3rd instance for a live implementation with up to 25 concurrent users in a couple of months. We are running KAIO on all instances.
Due to a known issue with HP containers in this setup, IDS running more than one CPU VP can lead to the system hanging so we've had to restrict it. What we're finding is that under high load (many heavy queries and concurrent OLTP access) that performance tanks (and 'top' shows one CPU at 99.99% load while the other 8 are at 1% or so - as expected).
How can we best improve the performance here?
PS we have CPU noage set, "SINGLE_CPU_VP" set to "0", OPTCOMPIND set to 0 and PDQPRIORITY set to 1. We have 64 AIO VPs running (there are 28 dbspaces and 300 chunks), and the users all connect via NETTYPE ipcshm. All the databases are Buffered logging.
There are 12 LRUs
MAX and MIN DIRTY are 4 & 2
DD_HASHSIZE 67, DD_HASHMAX 15
DS_HASHSIZE 67, DS_POOLSIZE 127
TIA
Malc