upgrading memory on RX8640 server

[Donald Fisher]

I have an RX8640 system configured into two npars. npar0 has 1 cell.
npar1 has 3 cells and is subdivided into 2 vpars. npar1,vpar0 has been given 8 cpus and npar1,vpar1 has been given 4 cpus.
I am currently using 2GB memory sticks on the system.
I would like to add more memory to the system and the only way to do so would be to increase the size of the sticks. What is the best way to do this. for example say I want to add 16 8GB sticks to the system. would I load them to cell 1 slots 0A/0B, cell 2 slots 0A/0B,cell 3 0A/0B etc..... or would I just fill all 16 slots on the lowest numbered cell of that npar?

[Sameer_Nirmal]

As per the performance tunning guidelines for nPar available at http://h18000.www1.hp.com/products/quickspecs/12471_div/12471_div.HTML which says :

* For best performance, a cell should be configured with a multiple of 8 DIMMs or four pairs (although the server will execute properly with an odd number of pairs). It takes 8 DIMMs to populate both memory buses. Populating only one of the two memory buses on a cell board will deliver only half the peak memory bandwidth.

* Load memory equally across the available cell boards.

Now within this nPar and depending upon the need of the vPars, you can move around memory.

[tkc]

i think firstly, you will need to determine whether your npar0 or npar1 will need the additional memory resources. after that, you can apply the guide given by sameer.

[Donald Fisher]

So,
It sounds as if I should upgrade by adding 8 DIMMS to the first cell board and then 8 DIMMS to the next cell board etc... to fully utilize the memory buses on each cell.
Do all of the CPUs configured in an npar share all of the memory on all of the cells or do they more efficiently (or only) use the memory from a particular cell?

Thanks

Don

[Sameer_Nirmal]

It's all depends on the use of ILM and CLM. These parameters configuration is dependent on the Application behaviour,load and environment (npar and vPar).

Usually, in an nPar with ILM configured, all the CPUs can share memory on all of the cells. Within a vPar enviornment, CLM maybe beneficial depending on the application.