> So, how do you folks feel about VMRM? Does it work as advertised? Sort of. IF your workload changes in the ways that VMRM understands, then it is actually pretty effective. It's not WLM in that it doesn't adjust task priorities the same way, and the formulas used to calculate when it does things are a little hard to dope out. Also, if you primarily have Linux workload, the formulas VMRM uses don't match up very well with the way that Linux uses CPU and I/O, so it's a lot less effective for all-guest environments.
I don't remember if the 5.2 code now has the I/O prioritization code active. It's a no-op in earlier releases (you could code it, but it really didn't do anything). Chris Casey has done presentations at the last few zExpos on VMRM and how it functions. It's worth digging one of those up because the presentation explains a lot more about how VMRM functions. I've been meaning to see if one could do more with PerfKit macros -- I think one could produce something a lot more effective for all-guest environments via that route, since PerfKit can actually get data from inside the guests too. -- db
