Operating System - HP-UX
1748117 Members
3806 Online
108758 Solutions
New Discussion юеВ

Re: Multiple vPars with Only One Core in a business critical environment

 
UdayNaik
Occasional Visitor

Multiple vPars with Only One Core in a business critical environment

One of my customer is having multiple vPars with just 2 Cores of PA8900... Now he is looking to migrate to Itanium Montvale 1.66/18MB ... Looking into the Performace improvements of IA over PA ... can i reduce the core count to Half .. means just one Core per vPar??

How advisable is that from Resilence and Performance point of view....
2 REPLIES 2
Steven E. Protter
Exalted Contributor

Re: Multiple vPars with Only One Core in a business critical environment

Shalom,

vpar requires one physical CPU minimum.

So if a core is treated by vpar as a physical CPU, it is possible to do what your customer desires.

Advisable? Probably not. A high performance system benefits from having more than one CPU to handle intense process or i/o lock or spinlock requests from applications.

I would suggest an advisable configuration would be two cores per vpar.

SEP
Steven E Protter
Owner of ISN Corporation
http://isnamerica.com
http://hpuxconsulting.com
Sponsor: http://hpux.ws
Twitter: http://twitter.com/hpuxlinux
Founder http://newdatacloud.com
Torsten.
Acclaimed Contributor

Re: Multiple vPars with Only One Core in a business critical environment

From the manual:

Processor Terminology Processing resources under vPars, both as input arguments and command
outputs, are described as ├в CPUs.├в For multi-core processors such as the PA-8800 and dual-core
Intel Itanium processors, the term ├в CPU├в is synonymous with ├в core.├в The term ├в processor├в
refers to the hardware component that plugs into a processor socket. Therefore a single processor
can have more than one core, and vPars commands will refer to the separate cores as distinct
├в CPUs,├в each with its own hardware path.

...



With the PA-8800s and other dual-core processors, there are two CPUs per socket. (On a cell
board with four sockets, this allows 8 CPUs per cell board.) The CPUs that share the socket are
called sibling CPUs.
Splitting sibling CPUs across virtual partitions refers to assigning one sibling CPUto one partition
and assigning the other sibling CPU to a different virtual partition. No noticeable performance
degradation has been seen when splitting sibling CPUs. Due to items such as the larger L2 cache
size, there actually can be a small performance boost if the siblings are split such that one of the
virtual partitions has noworkload. If you require consistently predictable performance, configure
the virtual partitions consistently; in other words, decide whether to split siblings or keep them
together, and maintain that policy across all virtual partitions.

Hope this helps!
Regards
Torsten.

__________________________________________________
There are only 10 types of people in the world -
those who understand binary, and those who don't.

__________________________________________________
No support by private messages. Please ask the forum!

If you feel this was helpful please click the KUDOS! thumb below!