Comments (3)
  • Add a Comment
  • Edit
  • More Actions v
  • Quarantine this Entry

1 specialsystems commented Permalink

...if you are having issues now, are these complexity problems going to "go-away" with Haswell, or POWER8 technology...or should we wait to move to 128-bit servers and 256-bit main operating systems...we don't want <br /> our Dolly Parton issues becoming Anorie Collins issues http://anoriecollins.com WW

2 AshishNainwal commented Permalink

I have a scenario which I am unable to understand or as you say "how hypervisor is thinking" <br /> LPAR with following lssrad output <br /> REF1 SRAD MEM CPU <br /> 0 <br /> 0 63156.44 0-31 <br /> 1 0.00 32-63 <div>&nbsp;</div> LPAR has 64GB RAM with 16 dedicated CPUs (SMT4). Why is no memory being allocated in SRAD 2 when the logical CPUs are being spread across 2 CPUs. It is a P780. Wouldn't it be beneficial for the SRAD1 set of logical CPUs to have a separate set of memory? Please help me understand.

3 Karl-HeinzBraun commented Permalink

Hi Nigel,
I have a similar case here :

 
$ lssrad -av
REF1 SRAD MEM CPU
0
0 36422.44 0-11 16-19 28-31 48-51
2 14691.00 20-23 36-39
1
1 18426.00 12-15 24-27 44-47
3 9213.00 32-35
2
4 5460.44 40-43
 
This is on one of the new 750 Express featuring 4 dual chip modules with 2x4 cores (32 cores in total, 128 LCPUs).This is why we have local, near and far here.
However, the first SRAD seems to have 6 cores, which is impossible.
But in this configuration, we have EC=4 and VP=13. Because the EC/VP ratio <= 0.5, the hypervisor can place more than one VP on the same core. With this sort of overbooking, only 11 cores can be used at the most. I don't know if this is a bug or a feature ;-)
 
Best regards,
 
KH