Which size of HDDs are supported by the SAS2-846EL Backplane?

urobe

Contributor
Joined
Jan 27, 2017
Messages
113
Hey there,

Can anybody tell me, which size of HDDs are supported by the SAS2-846EL Backplane?

I think the SAS1 version of the backplane was limited to 2 tb drives (at least when all 24 slots were used).

But the SAS2 should at least support 8 tb?

-Tobi
 

artlessknave

Wizard
Joined
Oct 29, 2016
Messages
1,506
a basic google search ought to have answered the question

but eh, here goes.

as I understand it, SAS1 is technically not the limiter, the bit size of the addressing is; any SAS1 controller/expander can be updated to 48+ bit addressing as long as the hardware is capable of handling it, but most companies aren't updating firmware for ancient junk.
SAS 48bit has functionally unlimited drive size with current drive sizes. gonna try some math. might not be perfect but should be illustrative.
32bit LBA~=
2.2​
TB@ 512 byte sectors(really old SAS1 hardware or SAS1 hardware without firmware upgrades)
48bit LBA~=
144,115​
TB@ 512 byte sectors(anything SAS1 with updated 48+bit firmware, anything SAS2+)
64bit LBA~=
8,589,934,592​
TB@ 512 byte sectors(anything SAS1 with updated 64+bit firmware [not sure if these exist], anything SAS2+)
this gets even more silly with 4k sectors (multiple the above by 4)
 

danb35

Hall of Famer
Joined
Aug 16, 2011
Messages
15,504
But the SAS2 should at least support 8 tb?
From personal experience, it does--I have six 8 TB disks on that exact backplane in my FreeNAS box, and they've been there for several months, working just fine.
 

urobe

Contributor
Joined
Jan 27, 2017
Messages
113
Thank you very much for the very detailed information. I did read into it, but gut somewhat conflicting informations. Where 10 large drives were supported but not on all 24 slots, and it wasn't clear which backplane they were talking about.
 

artlessknave

Wizard
Joined
Oct 29, 2016
Messages
1,506
ya, that's all sas1, as long as you skip SAS1 there should be zero problems (iirc the reason 10 large drives works is basically cuz it uses the adressing space from the full 24 drives to kind of cheat)
. also, the direct attach -sas- backplanes, like BPN-SAS-846TQ or BPN-SAS-846A, which are *technically* rated as SAS1, are just SAS wires really, so they work up to SAS3 and have no restrictions since they are not expanders
 

jgreco

Resident Grinch
Joined
May 29, 2011
Messages
18,680
ya, that's all sas1, as long as you skip SAS1 there should be zero problems (iirc the reason 10 large drives works is basically cuz it uses the adressing space from the full 24 drives to kind of cheat)

Uh.... wha, now?

Drives are addressed individually. Either a greater-than-32-bit LBA is correctly passed to the drive by the SAS expander, or it isn't. It has nothing to do with how many drives there are. There is no "adressing space from the full 24 drives".

. also, the direct attach -sas- backplanes, like BPN-SAS-846TQ or BPN-SAS-846A, which are *technically* rated as SAS1, are just SAS wires really, so they work up to SAS3 and have no restrictions since they are not expanders

It would probably be a bad idea to rely on them working at SAS3, but you do have a good chance of getting away with that since SAS is differential.
 

artlessknave

Wizard
Joined
Oct 29, 2016
Messages
1,506
Drives are addressed individually. Either a greater-than-32-bit LBA is correctly passed to the drive by the SAS expander, or it isn't. It has nothing to do with how many drives there are. There is no "adressing space from the full 24 drives".
totally possible. i did say "iirc". if that's the case though, what is the reason for the behavior of a certain number of large drives working but not over that number, that seems to have been reported?
 

jgreco

Resident Grinch
Joined
May 29, 2011
Messages
18,680
totally possible. i did say "iirc". if that's the case though, what is the reason for the behavior of a certain number of large drives working but not over that number, that seems to have been reported?

I've never seen that reported as an actual experience. It invariably traces back to a particular type of crappy eBay listing that is ambiguous and mentions the total capacity a chassis is capable of (i.e. 12-bay == "up to 24TB of storage supported"). That does not mean you can toss three 8TB drives in there. It means you can put up to 12 drives of no more than 2TB each. You can DEFINITELY find reports of people who were disappointed by that too.

This may be further confused because NexentaStor limits (limited?) their community edition to a certain amount of storage (10TB?) before you had to buy their paid product, and there are other examples of capacity-limited storage like that which are not technical limits.

This is confusing to people.

A general-purpose SAS expander chassis would have no reason to have a capacity limit unless you were pulling a NexentaStor and requiring an additional license fee, or if there was an actual technical issue. I would totally believe that if you were buying some brand-name SAS shelf that had a controller in it that there might be a license issue. But for a Supermicro, no. Therefore you need to figure out what the technical issue is, and it turns out that the SAS1 expanders had a 2^32-1 LBA limit. I've heard various stories that this might be remediable via a firmware update, and other expert opinions that this isn't the case, but as I have no firsthand confirmation either way, I simply steer people away from SAS1 expanders and SAS1 HBA's.

The SAS2 and beyond stuff have no such limits. You posted a reasonable-looking chart above; we won't see 144PB drives in the near future, so don't bother pointing out that there's some theoretical limit. :smile: It took us 30 years to go from 1GB to 16TB (~4 orders of magnitude). I can only hope that we experience a flash price crash and SSD density shoots way up, but we're still not going to see 144PB drives in the service life of SAS2 gear which is already a decade old.
 

artlessknave

Wizard
Joined
Oct 29, 2016
Messages
1,506
I've never seen that reported as an actual experience.
hmm, i guess i will flag that memory as corrupted then. too bad I don't have ECC.
remediable via a firmware update,
hmm, swear I read that somewhere...but its probably connected to the previously no longer reliable memory, so, no longer sure.
we won't see 144PB drives in the near future,
technically, we dont foresee this happening in the near future due to current projections an known technologies, but something could change and totally rewrite it with a new invention. the point of pointing it out was to show the absurdity of the max size of sas2+ compared to current and likely future drive sizes.
I can only hope that we experience a flash price crash and SSD density shoots way up,
same. ran across some 7TB SSD...for ONLY $25,000 @ a steal being on sale for $8,000 off.....
 

jgreco

Resident Grinch
Joined
May 29, 2011
Messages
18,680
hmm, swear I read that somewhere...but its probably connected to the previously no longer reliable memory, so, no longer sure.

Okay, but for which silicon?

See, there's a number of SAS expander silicon manufacturers, including PMC Sierra (PM-80xx), LSI (LSISAS2x36), Maxim (MAX70238B), etc. Those are SAS2 part numbers but there were even a few more mfrs back in the SAS1 days. See these things are basically specialized little communications processors. So it's certainly possible that there were firmware updates for one or more of the SAS1 expanders that were available.

technically, we don't foresee this happening in the near future due to current projections an known technologies, but something could change and totally rewrite it with a new invention. the point of pointing it out was to show the absurdity of the max size of sas2+ compared to current and likely future drive sizes.

Still not going to happen. There's a point of impracticality that you will pass.

We passed the point of impracticality for HDD and small file storage years ago. Back in the days of 1GB HDD, running at maybe 100 seeks per second, and approximately two million addressable sectors, you could fill a disk using random sector writes in about 20,000 seconds (quarter of a day). By comparison, a modern 14TB HDD running at maybe 200 seeks per second, the math result is way different. You have twenty seven billion addressable sectors and need about 137 million seconds (1580 days) *just* to fill it. So these are basically impractical for random read/write of sector-sized data, and the only reason HDD still works out is because we usually have large sequential runs of stuff to store.

You can look at SSD on SAS2 to see that there is a practical limit that exists with SAS2. At a generous 550MByte/sec fill rate, you can write 47TB/day or 87PB in the expected 5 year lifespan of a storage device -- doing absolutely no other operations (though SAS2 is bidirectional and dual port so it has possibilities). So you can probably create some hypothetical situation with SAS2 where you have a write-only device that's going constantly at 550MByte/sec for like 10 years and you actually manage to touch each 48-bit LBA once... but I deem this impractical with a "stupid too" bonus multiplier :smile:

same. ran across some 7TB SSD...for ONLY $25,000 @ a steal being on sale for $8,000 off.....

Well the interesting thing is that the vast amount of bits I see being stored on stupid-expensive flash out there are not actually being rewritten often enough to justify the stupid-expensive flash.

I've been deploying consumer-grade flash in hypervisors since 2010-2011. Because most of my applications are "must-run", the typical dodge of just buying a single enterprise SSD and hoping for the best wasn't in the cards. I put them behind an LSI RAID (the actual real ones, not fake-HBA ones) and most of those survive to this day. I think I won, since I was paying well less than half what an enterprise SSD would have been. I do have some workloads that actually require the enterprise SSD, and those get that. *shrug*.

The experience with the 2015 480GB Intel 535's was also enlightening. That year, prices had been plummeting for SSD's, and projections were for this to continue. So I put several TB of 480's out in our data center, two RAID1 datastores with a spare in each hypervisor, knowing that we were going to blow through their 40GB-per-day endurance. I figured on buying new ones in a year or two. Some of those survived, some didn't. I was wrong about future pricing trends but that info came from people more knowledgeable than I about the trends.

Thing is, I'm pretty certain that if larger consumer SSD's were put out, they'd sell, but they might not sell to consumers. Not all enterprise workloads demand 1DWPD. As a matter of fact, I think most of them do NOT. But right now a lot of that is being managed on conventional HDD storage arrays because the price for enterprise is hideous, as you note. Other bits of it are being stored on hideously expensive enterprise SSD.

If you look inside a modern 2.5" SSD you will find it mostly empty air. I wish they'd cram the damn thing full of flash for like a 32TB SSD :smile:
 
Top