Just a few tiny corrections ...
Thanks so much for coming to the session yesterday!
Actually, I think I said in the slides that Distributed Power Management, a shipping feature of VMware ESX 3.5 and vSphere 4.0, was able to save 73% power in a real world case that was cited. The memory power mgmt demo showed that we could save ~100W - ~200W (in a system showing 1000W partially idled) by placing half of the DIMMs in the system into software-controlled standyby or offline.
Just to clarify, the patrol scrubber is a background autonomous hardware process that is configued by BIOS. Without affecting normal system operation, the memory controller will
continuously perform read/write operations on memory correcting any soft errors that may exist in memory. The write will re-generate the ECC bits, update them if you will. If in the process of the read, an uncorrectable memory error is encountered (double-bit or higher), then an error (such as MCE) can be flagged to software and in older systems, the system would need to reboot. Our demo used new Intel technology that can remove this need if the error hits non-hypervisor pages. (For a quick description of memory scrubbing, see this link: http://h20000.www2.hp.com/bc/docs/support/SupportManual/c00218059/c00218059.pdf)
- Rich Brunner