SD2 memory configuration

[SYSTEM]

On the following SD2 memory configuration, can I set all memory as ILM ?
All cell blades are belong to one nPar.

　　CellBlade#1 : 64GB (8GB x4DIMM x2)
　　CellBlade#2 : 64GB (8GB x4DIMM x2)
　　CellBlade#3 : 64GB (8GB x4DIMM x2)
　　CellBlade#4 : 64GB (8GB x4DIMM x2)
　　CellBlade#5 : 32GB (4GB x4DIMM x2)
　　CellBlade#6 : 32GB (4GB x4DIMM x2)
　　CellBlade#7 : 32GB (4GB x4DIMM x2)
　　CellBlade#8 : 32GB (4GB x4DIMM x2)

There is a SD2 with CB#1/#2/#3/#4 above and is going to add CB#5/#6/#7/#8 (borrowed from other system) temporally due to deal with performance problem.

And the memory will be configured to 100% ILM because of the requirement.

There's some worries about
- interleave between different DIMM size.
- whether the excess memory on CB#1-#4 works as ILM as expected.

[HEUNG]

> On the following SD2 memory configuration, can I set all memory as ILM ?

Are all these populations legal and supported for an SD2 Blade ? Yes, these are all legal and supported populations, so what would make anyone think this nPartition configuration would not be supported ? Granted that an arbitrary choice to configure this memory as 100% ILM would be considered beyond stupid, but it is certainly well within SD2 rules. On another note, no one should be using the term CellBlade anymore to describe an SD2 Blade.

> There is a SD2 with CB#1/#2/#3/#4 above and is going to add CB#5/#6/#7/#8 (borrowed from other system) temporally due to deal with performance problem.

> And the memory will be configured to 100% ILM because of the requirement.

> There's some worries about
> - interleave between different DIMM size.
> - whether the excess memory on CB#1-#4 works as ILM as expected.

If this customer is concerned about performance, then they probably should not be configuring their partitions as 100% ILM. Unless their workload is very unusual and has an enormous amount of data which must exist in shared memory, this configuration pretty much guarantees that SD2 performance will be about as bad as it could get. Excessive amounts of ILM should not be considered a "friend" to performance on SD2 or with 11i v3. The disparity of memory capacity (64GB on some, 32GB on others) has a much bigger negative impact to global interleaving than the size of the DIMM, and this will result in CPU cores on Blades 5-8 making remote memory accesses 100% of the time for a full 1/3rd of the total memory capacity, and 93.75% of the time in the other 2/3rds of the nPartition's memory. In the off chance this isn't obvious - that's really bad in memory latency terms.