The VNX2/MCx Array release by EMC in 2013 highlighted the whole “multi-core everything” (MCx) software upgrade to make best use of the available CPU cores in the Sandy Bridge Xeon E5 Processors. This allowed for greater parallelism of tasks and big performance improvements across Data services and functions.
I’m not going to detail it further here, a) because it’s old news and b) because it’s already been done by other bloggers in great detail.
One item I still see people struggling to come to terms with is FAST Cache efficiency reporting. “Why isn’t my FAST Cache being used” “FAST Cache is not working” followed by much angst and gnashing of teeth.
The confusion largely centers around the fact that one of the old, faithful goto methods for measuring FAST Cache performance “Percent Dirty Pages” is pretty well deprecated.
This is the metric shown on a series one VNX;
This is the output shown on a Series 2 VNX MCx;
naviseccli -h arraysp cache -fast -info -detail
Both are operating fine, but appear very different, at least on the surface.
Why ? Figuratively, VNX2 FAST Cache works a bit more like traditional Write-through cache, sort of 😉
At a page level, when a page is written to Fast Cache, it is marked as “clean”. If an inbound write changes this page, it is then marked “dirty”. Dirty pages are copied to the backend disk very quickly and the Page is marked “clean” again, even though it remains in the Cache. Clear ? Groovy. Stale pages are overwritten by fresh ‘hot’ pages that get promoted.
There is also a ‘Proactive Cleaner‘ process that copies dirty pages to the backend disks during periods of low cache activity on a LRU basis. Here’s a pretty picture showing the process;
The net result is that under normal activity, you will not see the “Percent Dirty Pages” value increase. If you watched it in real time, you would see it occasionally burst up then drain away/disappear very quickly. This is by design.
With that in mind, don’t be alarmed with a Percent Dirty Pages value of 0. The best way to check your FAST Cache’s effectiveness is to use the Hit and Miss Ratio metrics in Analyzer, Unisphere Central or M & R. These are the sources of truth. Or perhaps give miTrend a go if you haven’t yet.
Here’s the same VNX2/MCx array, showing fairly good Cache efficiency
And of course, FAST Cache is only reported at a Pool level, not LUN level, except when referring to RAID Groups.
Get on board.