About IBM´s partitioning story ...

While searching for some informations about LPARS i read some interesting articles about the LPARS technology:

  1. According to an article in a pseries forum, you have to calculate 1.25 GB memory overhead for a 32 GB System. To get a 256 MB LPAR up and running you need 756 MB main memory. When the proposed calculation in the article is valid for every memory configuration, the p570 with 756 GB memory, that was used for the TPC-C benchmark, would have an minimal LPAR memory overhead of 12,5 GB for virtualisation alone.
  2. In a IBM Redbook titled with <a href=http://www.redbooks.ibm.com/Redbooks.nsf/RedbookAbstracts/tips0426.html?Open”>Server Consolidation: A Comparison of Workload Management and Partitioning</a>:
The performance penalty of sharing processors depends on factors such as the size of the partition and the number of other partitions running. The causes of the overhead are due to increased context switches and loss of cache. Virtual processors can be dispatched on different physical CPUs, or the cache may be overwritten by another partitions, losing the benefits of affinity.

This matches with a Redbook that was withdrawn quite a while ago. But there is an article at OSnews that gives a summary of the interesting parts:

4-way SMP (4 cores) 1.00 relative performance
4 partitions dedicated (1 core each) ~1.05 relative performance
4 uPARs (2 cores each) ~.96 relative performance
2 uPARs (4 cores each) ~.90 relative performance
4 uPARs (4 cores each) ~.75 relative performance

Holy virtualisation, Batman, how do they sell this kludges …. must be the power of marketing …