I'm wondering whether it makes more sense to deploy a traditional FAS architecture with a combination of Flash and traditional disks if I have a combination of workloads that require Flash as well as traditional disks, or should I be positioning AFF + FAS. In some cases when deploying VDI solutions, it makes sense to utilise Flash to optimise the gold images, but some data such as "home drives" don't demonstrate any scales of economy when deployed on an AFF. Am I misinterpreting this or is there a methodology to deploy an AFF + FAS?
You've got a few choices here for bringing flash into your deployment.
1. A single FAS array that has a combination of aggregate types e.g
Hybrid SATA + Flashcache / Flashpool - This works really well, a highly active set of home directories can benefit greatly from a small amount of flash to host what is mostly hot metadata (file permissions, access times, name etc). In ONTAP 8.3 it will be possible to add flash capacity to specific aggregates in small increments. If you do decide to go down this path, it generally makes sense to keep all your SATA drives on a single controller.
Hybrid SATA + SAS 10K Disks - I've seen this combination work really well for most virtualised workloads, though there are some design considerations you might want to be careful of, mostly around Cache sizing. The good news is that ONTAP 8.2.1+ has a feature called automated workload analyser (AWA) that takes the guesswork out of that. If you're interested in how that works let me know and I'll write up a list of good resources for you
All Flash aggregate - All the goodness of ONTAP plus all the goodness of flash.
If you are using Clustered Data ONTAP, it's easy to move your workloads between these aggregate types on the fly, and OnCommand performance advisor gives you plenty of data to help you make that decsion.
The big question is, whether you need an all flash node to host your all flash aggregates. In general I'm a big fan of the idea of having at least a pair of All Flash FAS nodes in your cluster for the following reasons
By dedicating your all flash aggregates to a controller or two, you are ensuring that you have enough FAS hardware resources (e..g CPU and Memory) to really get the best out of the Flash aggregates.
It isolates the performance impact of really bursty and very demanding workloads from the normal workloads on your system, when combined with the non-disruptive workload movement capabilities it gives you an easy way of alleviating performance challenges in your environment when workloads unexpectedly spike due to factors outside your control (e.g. software upgrades, people running massive analytics reports in the middle of busy OLTP times etc).
It allows ONTAP to optimise it'self for low latency workloads giving really predictable performance, something it does well via Quality of Service features in hybrid environments, but giving ONTAP an all flash diet makes it's job much much easier.
The bundle pricing for All Flash FAS is really compelling - more flash, less money 🙂