Showing results for 
Search instead for 
Did you mean: 

Expert advice required on superdome upgradation

Valued Contributor

Expert advice required on superdome upgradation

Hi Experts,

I appreciate your advice on my following queries.

1.Do we suggest upgrading SD32A currently configured with PA8800 processor with additional PA 8800 processors or PA 8900 processor in a separate nPar?

2.Do we suggest upgrading SD16K currently configured with PA8700 processor with additional PA 8700 processors, PA 8800 processor or PA 8900 processor in a separate nPar?

3.For each PA 8600 processor on V2600, how many PA8700/8800/8900 processors are required on Superdomes?

4. When we create a new nPar on Superdomes, do we need to have a separate management network connectivity specifically for the new nPar? Is there a need for adding any management specific components for the new nPar?

5. Do we need a separate OS license for the new nPar?

6.Is configuring new cell boards on Superdome disruptive to existing applications?

7.Is it possible to configure some of the expansion slots on existing IO chassis on Superdomes to newly created nPar with processors configured on new cell boards?

8.My understanding is that there is no way Superdome with PA8700 processors cannot be upgraded with PA8800 or PA8900 processors even when new cell boards are configured for new nPars. Is this correct?

I hope I will receive more replies/advices on the above questions... if any of you get answer for some of the questions please answer me with the question number... so that I can get an idea.

Thanks in advance...

Honored Contributor

Re: Expert advice required on superdome upgradation

1.) PA 8900 is newer and likely to cost more than PA 8800. Is the extra performance worth the extra price to you?

2.) SD16K... you mean a rp8400, right?
I'm afraid you can only install PA8700 CPUs to it. For higher performance CPUs you'll need a rp8420.

3.) Depends on the application: for a worst-case example, we have one application that has a rather inflexible scaling behavior. Replacing CPUs with faster ones will shorten the processing time per transaction, but the number of transactions that can be processed in parallel is tied to the amount of CPU cores available.

(I think this indicates the design and architecture of that application may have some fairly serious flaws; but we'll have to live with it for now.)

Without knowing your application and how it scales (both in terms of more/fewer CPUs and faster/slower CPUs), the question cannot really be answered.

Network connections:
A new nPar is like a new physical server, except that power and console connections are bundled together with the other nPars of the same complex. If you have a policy to have a dedicated network connection for an enterprise monitoring system (like OVO), then it would make sense to have one for each nPar too.

Other components:
A new nPar requires a Core I/O element: in older Superdomes this is a separate PCI card. The Superdome SD16B, SD32B and SD64B models integrate the core I/O functionality to the cell boards.

6.) If configuring the new cell to become a new nPar: no (but see below).

If the new cell is going to be used to expand an existing nPar: yes, you'll need to perform a reconfiguration reboot to the nPar to allow the new cell to join it. Also review the applications: if an application is licensed in a per-CPU basis, you may need to upgrade the application license when adding a new cell with new CPUs.

If the new cell hardware version requires updating the firmware version levels in the entire Superdome complex (i.e. you haven't kept your Superdome firmware levels up to date): at the worst case, you may need to shutdown the entire Superdome complex for the firmware upgrade.

7.) No. As far as I understand, each IO chassis is physically connected to a specific Superdome cell. It cannot be "split" between cells (and thus between nPars).

8.) It depends on the model of the Superdome hardware. For example, the oldest 32-way PA-RISC Superdomes have a model string "SD32000", the newer have "SD32A" and the latest model is "SD32B". The oldest model cannot handle multi-core CPUs (PA8800/PA8900).

The newer ones have a newer chipset: sx1000 for the "A" models, sx2000 for the "B" models. Both these chipsets can handle the latest PA-RISC CPUs, although the sx2000 chipset has more bandwidth -> better performance.