Philbert Rupkins wrote:
Peter - I did not know flash cache would potentially be an issue. We are using flash cache on our 3240 systems. Will definitely look into this further.
By "issue" was meant (presumably): severe restrictions/limitation around sizes w.r.t. config of the FlashPool. It can get virtually unusable in a small system, even if it's officially supported. Can depend on how the Aggrs are layed out in there, and there are restrictions on that as well on small systems (sizes...) I'd go for FlashCache in smaller FAS machines, any day.
So yes, this FlashPool issue is connected to the very limited amount of memory in these older small FAS boxes. E.g. WAFL really needs its memory buffers, it's crucial. Memory handling inside ONTAP is a bit... shall we say convoluted -- not well coordinated beetween different parts of the system
As for CPU cycles, I'm not sure that 8.1 -> 8.2 -> 8.2.2 would increase this in any way that it should potentially affect performance (protocol latency). I don't think so, not in general, but YMMV (as always :-) ) If anything I think that avg CPU could go up, due to higher parallelism in lopri work inside ONTAP (bg scanners been broken out of the big serialised chunk, the "kernel domain" long ago given the name Kahuna).
Lopri work doesn't contribute to protocol latency (except transients due to pressure -> level-off effects), only Hipri work does. So what 8.2.2 is doing is essentially utilising the HW you bought better, w.r.t. CPU. Much better in fact, than 8.1
/M
Mike - glad to hear it has worked out well for you with the 2240s! That makes me feel a little more optimistic about the possibility that we can upgrade our 2240s.