Limitations of only one bound cpu in a vpar?

[Ian Vaughan]

Howdy,
I have a pair of rp7420's and am investigating the feasability of a new configuration. What I want to do is merge the 2 npars on each and create 2 vpars from the resulting single npar.

Each npar is a ServiceGuard node and there is an active and a standby node in each chassis. The hosts will be unchanged other than becoming vpars rather than npars.
Each chassis has 8 active cores and 8 icap. My plan is to have 7 cores in the "active" SG node and 1 core in the "standby" with extra cores that can be brought online with TiCAP.

The main problem this presents is that the OS is 11iv1 and the vpars version is 3.x.
Any TiCAP cores will be unbound and therefore unable to handle I/O interrupts (as I understand it).
As the package that would be failed over into the standby partition is an Oracle 9i data warehouse (few users, very little system activity, multi TB, occasional big queries) is one bound cpu (and say 3 TiCAP) going to be enough?

I have heard various rumours about the I/O handling on these cell based systems including one which said that only one CPU handles the I/O interrupts anyway.

I am now a bit confused. Also there is no chance of upgrading to 11iv2 in the near future.

Thanks in advance for your comments and observations.
Ian

[Steve Lewis]

What about Global ICAP (G-ICAP) and Global work load manager (GWLM)?

[Ian Vaughan]

Hi,
We've ruled out GWLM mainly because of cost and GiCAP needs the complex that it is "borrowing" resource from to be up in order to transfer the entitlement.

We figured that if we lost one of the rp7420 chassis then we couldn't dynamically grab resource from the dead one to give to the survivor.

Thanks for looking

Ian