I just read on blocksandfiles that Seagate started to double the heads in Exos X18 series, branded as 2X18. But they say that aplications must be updated to address the new hardware properly. Will they be supported by Storj? Is there a benefit? I’m interested in buying these new toys for future upgrades of my nodes.
Thanks!
2 Likes
storagenode uses OS to access disks, so the question would be - is it supported by your OS?
I believe there would be no difference though.
I wonder if it is supported by OS correctly - by my logic (could be wrong) it is a kind of raid in a sense that there are 2 separate writes/reads at the same time, so data should be written accordingly in separate pieces? Is it being handled by HDD controller perhaps?
As far as I know, those drives are seen as two logical units (half capacity each) by the OS. You can create a RAID0 from them and get a “single drive”. For that you can use your OS if it supports that or you may need a RAID controller.
In theory, you could also use the two halves as separate drives and run two nodes on them, but I do not recommend this since it is very likely that the whole drive would fail instead of one half at a time.
Even in this case I would suggest to do not use RAID0, however, I will agree, that running two nodes is still some kind of not good too.
I hoped it would be a one disk.
RAID0 would be appropriate for this type of drive. The drive has a common motor, but two actuators. If it dies, it is extremely likely that both halves will die at the same time (it could develop a bad head or platter, but that is less likely than a head crash etc).
By joining both halves in RAID0 you get the same reliability as with a normal single-actuator drive, because, essentially, a normal drive is a “RAID0” of however many heads it has, the difference is that in the new drive both halves can seek independently.
Of course RAID options with actual redundancy would be better, but also would require multiple drives, just like with the normal drives.
2 Likes
I believe, this is was a goal to create such a centaur.
To make it reliable you need to buy at least two such drives 
I think the goal is to make drive read write faster with cheap price, as you need to add only second head and some electronics all other is same cost. but you have 2x bigger read write speed
1 Like
As far as I have read only the SAS drive. The SATA one should present itself as a asingle drive.
1 Like
Number of heads is the same, the difference is that the head stack is split into two independent parts, so minimal additional parts to pretty much double the IOPS of the drive. The drive is basically two thin regular drives on top of one another.
In which case there would be no difference to the OS or the node software and the drive would do the RAID0 internally.
1 Like
Q: How does an Exos® 2X SATA configuration differ from a SAS configuration?
A: For the SAS configuration, each actuator is assigned to a logical unit number (LUN 0
and LUN 1). For example, one 18TB SAS drive will present itself to the operating system
as two 9TB devices that the operating system can address independently, as it would
with any other HDD.
The Exos® 2X SATA configuration will present itself to the operating system as one logical
device since SATA does not support the concept of LUNs. The user must be aware that
the first 50% of the logical block addresses (LBAs) on the device correspond to one
actuator and the second 50% of the LBAs correspond to the other actuator.
1 Like
From that it sounds like you will only get double iops when using both sections at once. So unless the OS knows to spread the data out like that it probably doesn’t make much difference?
1 Like
Yeah, it’s just two thin drives with a common spindle motor.
This, however,
means that in order to be able to get the IOPS from the drive you have to create two partitions and then combine them in RAID0. Otherwise, until you fill the drive up to more than half, you will only be using the one actuator.
1 Like
Sounds very first gen to me. They do so much with HDD controllers these days, I wouldn’t be surprised if future generation SATA versions would deal with this internally automatically. At least when they start offering more consumer oriented HDD’s with this feature.
There are two obvious ways to combine the two halves:
- The first X TB is first half, the other X TB is the other half
- Internal RAID0 with some fixed stripe width.
Both options have their pros and cons, but the first one allows users to partition the drive and use both halves independently (though I wonder how do you know exactly where is the LBA0 of the second half).
I’m sure that is documented somewhere. I understand that for exos drives it makes sense to leave it up to the end user. But it wouldn’t surprise me if they start using this tech in ironwolf or other more consumer HDD’s. I’m pretty certain they will just internally raid and obfuscate it to the OS then. Kind of how SMR was initially host managed and basically all moved to drive managed now.
There are still sweet things coming to HDDs, they will not be killed by SSDs anytime soon, like DVDs where killed by USD sticks. 2x heads count was one, but HAMR (heat assisted magnetic recording) is another and promises big increase in capacity. 50TB per drive in 2024.
But for STORJ usecase I think it might make sense to use a 2X18 drive. If the drive is filled to more than 50% with node data, new data is written to “drive 2” while egress-reads are mostly done on “drive 1”. This reduces the IO pressure on the single drives.
Also when the drive is almost 100% full, things like file walker or GC could finish in half the time of a normal drive.
I just ordered a refurbished 2X18 SATA drive and will do some tests
I assume you propose some kind of concatenation of the two parts? Normal RAID0 would write to both parts at the same time. I think it can be done with lvm though.
However, most of egress comes from recently-uploaded files (let your node run out of space for new data and watch egress drop), so I don’t think it would help much. The full part of the drive would stay mostly idle, while the part with empty space would have both uploads and downloads.
hmm yes that’s true, but some old data also gets deleted from time to time which makes space for new (hot) data on the first (half) drive 
And it would still lead to half of the file walker runtime I think.
Afaik the 18TB SATA Version only presents one drive to the OS, so the drive controller will do the RAID stuff.