I get that someone put a lot of effort into making NetApp "products" easier to size and buy and that there are some actual advantages to having heads and software targeted use single disk types, but this feels like the old "R" and "F" days.
For medium-sized customers that have a more evolutionary path with FAS systems, the split in the SLC/QLC vs. controller compatibility is frustrating. Fork-lift upgrades are not economical and while I'd gladly dump FSAS for QLC, I want to keep using the millions invested in SLC disks.
The beauty of NetApp always used to be (with the exception of the R series) that disks and controllers could have relatively separate life cycles and that gave everyone flexibility to scale the one or the other along the way. Generally, the constraints on QLC compatibility are most annoying, but the hard A vs. C split is there too, and I don't yet know what the new FAS systems will support, but I'm a bit frustrated by the choices I currently have.
</end_rant>
#Rant: A, C, FAS constraints
1 messages · Page 1 of 1 (latest)
While I agree you can still manage all A-Series, C-Series and FAS nodes in one big cluster. (ASA excluded)
I haven't dug into the controller constraints for MCC yet, they used to be "same controllers per site", but this adds to the complexity perhaps as well
You understand that in the A/C series, ONTAP essentially bypasses thousands of lines of code for spinning disk? SSD can attach to fas. You can certainly have a mixed fas with SSD and the SSD can be an aggregate or part of a hybrid aggregate. They are just slower.
I have not heard they in’s and out’s of exactly why the split for A/C series but it likely along the same line as code optimization. When the unit boots it dictates what it is and ONTAP sees that and uses optimized code paths based on what the unit is programmed to be
I believe I touched on that in my original post. But if I can (and do) have SLC SSD and HDD+Flash-Cache on FAS now (even as sub-optimal as that may be), why exclude QLC? Why can't A-series take both QLC and SLC? Why isn't there a "B-Series"?
As OG1 stated, you are able to setup a cluster (does not have to be MCC) with a mixture og F, A and C series, you just cannot mix the disks attached to the HA-clusters that have specific constraints...
if it was for the same price as one cluster pair with QLC and SLC, that might be an argument
and in this case, it has to be MCC
iirc, that was the "Maverick" project ca. 2012. I'm aware of this. Writing to cell-based storage is a different ballgame. LSI knew this a long time ago too, as did anyone who programmed for embedded devices with flash storage
When the unit boots it dictates what it is and ONTAP sees that and uses optimized code paths based on what the unit is programmed to be
it wouldn't be that hard to dynamically check on boot what types of devices are attached, and reconfigure the system accordingly. If QLC are attached -> boot in C-Series personality, if TLC are attached, boot in A-Series personality, and in all other configs boot in FAS personality.
I'm pretty sure the reason it was done with a bootarg, and with a non-changeable one, is purely marketing reasons...
They didn't exclude QLC from FAS they excluded all NVMe-based SSD storage from FAS. While you can attached TLC-based SAS-shelves you can't attach any NVMe-shelves (either TLC or QLC).
For your second question: I guess it was for the same reason why NetApp initially named their first QLC-SSD storage system FAS 500f and not AFF. Because they feared allowing "inferior" SSDs into the almighty fast AFF-brand might cause confusion (that's what I heard at least). Some customers might say "I bought an AFF and it so slow, I get no sub-ms latency" while not understanding they bought QLC-SSDs. Separating it into A- and C-Series was the better move imo.
But yes, marketing most definitely had a say in this decision.
I know that you are able to convert a FAS27xx to A220 by setting a BIOS variable... but as far as I remember you were not able to convert it back again... 🙂 Not sure why.. also not sure if you had to remove the 1TB NVME cache installed in the FAS (which I don't think the A220 has)...
Are you 100% on that? I think the A400 can have both SAS12 and 100G NVME shelfs attached? Maybe not supported in production?
QLC and SLC have different performance profiles. That's why you don't see them mixed.
It's not noticeable in most cases, but on edge cases it is.
That might be why, but I don't want to have that be the official stance, just my personal observation. I see I'm the only employee responding and I don't feel qualified to answer this.
Nooo, I meant that FAS can only connect SAS-based storage. While AFF can connect SAS- and NVMe-based storage.
As far as a I understood it it was no decision against QLC for FAS, but a decision against any NVMe-based shelves - doesn't matter if they house TLC or QLC SSDs.
That is 100% correct.
Any aff unit can attach NVMe tlc drives (not capacity based/qlc nvme) as well as regular SSD. It is strongly recommended not to mix in the same aggregate. If they are mixed, it should have separation at least at the raid group.
This hasn’t been true for a long time. The only platform this was actually supported on was the Fas80x0 (8020/8040/8060) series. After a specific bios update it became impossible to change as the loader variable is read only
i simply want to move from large capacity FSAS to large capacity SSD while not losing the SLC SSD's I have, and if HDD and SSD works well enough on FAS, then I'll be happy to take the hit with mixing TLC/QLC and SLC drives on one controller (and, of course, never in a single aggregate, but I think the available sizes make this mostly a non-starter anyway)
it would seem that the code should at some point be optimized for the base disk on each aggregate and not one optimization "personality" per controller, even if that is a lot of refacturing.
i'm not trying to take anyting away from anyone else who needs top end, ASA functionality or anything. There's just a gaping hole here that I'd love to see filled.
It can still be set but not reset. This is how spare parts work, you get "unbranded" FAS controllers and depending on what controller you're replacing you have 1 chance to set the correct variable. Once set it cannot be reset (technically it can, we had received a "branded" part as spare where we could reset the flag with NetApp engineering, but it is not something an enduser can or should do)
But that is a specific supported scenario. You cannot just take a field deployed FAS unit and convert it.
Very valuable discussion, I learned quite a few things👍
You might be able to do this.
Talk to the account team and see what they can offer. Maybe you could get a fPVR.
Yeah, I'm working that angle as well. Requests have been submitted. Crossing my fingers that the FAS line will still be the Swiss Army Knife of controllers
I've submitted a Future Enhancement PVR for support for QLC on future FAS systems. Current ones do not support the NS224 NVMe shelves.
This is a very complicated discussion as it becomes one of product management, sales reporting requirements, and not one of only technical capability unfortunately.