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.
I would split the load to both heads. What's the point in spending all the money for two heads if you are only going to use one...? 4 shelves of DS4243 will perform extremely well. I saw it on my own 3170, 6070, and 6080 pairs.
I would likely create one large aggr on each head (pending the environment...not knowing yours) and then create the volumes on that. I would also split the DS14mk4 shelves between the heads as well....personally, I would turn off the disk-auto-assign and assign them myself since the systems use software ownership, I can give a half shelf to one head and the other half to the other head.
--tmac Tim McCarthy Principal Consultant
RedHat Certified Engineer 804006984323821 (RHEL4) 805007643429572 (RHEL5)
On Thu, Nov 3, 2011 at 4:11 PM, Fletcher Cocquyt fcocquyt@stanford.eduwrote:
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.
Toasters mailing list Toasters@teaparty.net http://www.teaparty.net/mailman/listinfo/toasters
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
Fletcher,
There are other things to consider than just CPU utilization. One thing that comes to mind is network loading. My assumption is that you currently use both filers for active data. Can the networking on one handle the load for both? You will also want to monitor your networking whenever your backup/DR processes like SnapVault run. This spike may be too much for a single server.
It does seem a shame to have one box basically doing nothing. You should consider future workload. It may be fine now, but what about next year? How much trouble will that migration of data be to your users and your backup environment?
Another thing you should consider for your disk layout is how you will work the 32 bit to 64 bit conversion of your existing aggregates. My understanding is the DataMotion won't work between mixed size aggregate data. Make sure you understand the conversion process since you may have to use some of your new disks to facilitate the change.
I've been running 40 and 70 TB aggregates on my NearStores for a couple of years now. I haven't had any issues due to the sizes. We haven't bothered performance testing them much since the are NearStores running ATA/BSAS drives.
Jeff
On Thu, Nov 3, 2011 at 2:11 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.
Toasters mailing list Toasters@teaparty.net http://www.teaparty.net/mailman/listinfo/toasters
-
Eugene Vilensky -
Fletcher Cocquyt -
Jeff Cleverley -
tmac