On Thu, Nov 3, 2011 at 3:11 PM, Fletcher Cocquyt fcocquyt@stanford.edu wrote:
Fellow toasters, we are trading up some old trays for new DS4243 trays and we are trying to decide if we allocate all 8 of our DS4243 trays to one ~40Tb aggregate (we plan to run ONTAP 8.1 for data motion) assigned to one of the cluster's heads will the advantage of all those spindles in one aggr be killed by the head being overloaded? In which case would we be better served by 2 x 4 DS4243 AGGRs - one assigned to each 3270 head? We also have 7 DS14-MK4 shelves to help balance out the workloads All other things equal - I'm inclined to go for the one big 40Tb aggr vs 2 x 20Tb aggrs to reap the biggest IO capacity and performance Currently our CPU weekly average workloads on the two 3270 heads are only 20% and 16% My impression is a single 3270s head should be have enough headroom to push the large aggr to its full IO capacity. But I'd be very interested to hear the advice others who have tested both options and any unforeseen issues with an aggr this large (our previous largest is 16Tb under 7.3.5.1P2) thanks Fletcher.
You would have double the primary size by splitting the disks among the heads, and double the flash cache if you have that. Depending on your IO patterns of course; our working set is easily cacheable so I am always comfortable with more cache/less disk paradigms.
Cheers, Eugene