When do you think we will see Helium filled 2.5" hard drives?

[Excessi0n]

https://patents.google.com/patent/US6546499B1/en

^^^^ This patent is called "Redundant array of inexpensive platters (RAIP)".

It was filed back in 1999, but I'm thinking with Helium this could be a lot more interesting.

For example, lets say that Internal RAID-0 is possible, how fast do the platters really need to be spinning to saturate SATA 6 Gbps?

SAS 24 Gbps?

U.2? (PCIe 4.0, PCIe 5.0, etc)

Power consumption at these speeds?

I thought HDD manufacturers already did that?

 

[cbn]

I thought HDD manufacturers already did that?

Only one head (which can either read or write) is active at a time, rather than all of them simultaneously.

 

[cbn]

Some hypothetical speeds (which assume perfect scaling) I came up for "Redundant array of inexpensive platters (RAIP)" using current platter densities and spindle speeds:



1.) 3.5" 7200 rpm Helium eight platter (1.5TB PMR platters @ 250 MB/s read) = 16 surfaces (ie, 16 heads reading) x 250 MB/s = 4,000 MB/s



2.) 2.5" 10,000 rpm (non-Helium) four platter ( 600GB PMR platters @ 279 MB/s read) = eight surfaces (ie, eight heads reading) x 279 MB/s = 2,232 MB/s.

Assuming the current four platter Seagate 10,000 2.4TB SAS drive (above) could be expanded to six platter with Helium here is what it would get:

2.5" 10,000 rpm Helium six platter ( 600GB platters @ 279 MB/s read) = 12 surfaces (ie, 12 heads reading) x 279 MB/s = 3,348 MB/s



3.) 2.5" 5400 rpm (non-Helium) five platter (1TB SMR platters @ 140 MB/s read) = ten surfaces (ie, ten heads reading) x 140 MB/s = 1,400 MB/s

Assuming the current five platter Seagate drive above could be expanded to seven platter with Helium here is what it would get:

2.5" 5400 rpm Helium seven platter (1TB SMR platters @ 140 MB/s read) = fourteen surfaces (ie, fourteen heads reading) x 140 MB/s = 1,960 MB/s



P.S. SAS 24 Gbps is actually 19.2 Gbps according to this article. Dual port SAS 24 Gbps would therefore be 38.4 Gbps. (38.4 Gbps = 4,800 MB/s or 4.8 GB/s)

 

[cbn]

Regarding "Redundant array of inexpensive platters (RAIP)" I do wonder about the impact of the (hypothetical) on drive controller heating up.

Looking at the numbers in the previous post, there is a lot of throughput possible......

 

[Charlie22911]

In pure sequential performance yes, however the spindles and heads are fixed relative to themselves. My (uneducated) guess is that seek times will remain unchanged or perhaps worse, which is probably why it hasn’t been done yet.

 

[cbn]

In pure sequential performance yes, however the spindles and heads are fixed relative to themselves. My (uneducated) guess is that seek times will remain unchanged or perhaps worse, which is probably why it hasn’t been done yet.

If seek times were unchanged, the IOPS would still increase by whatever number of read/write heads the drive had. (So for a drive with 8 heads the IOPS would increase by x8 and for a drive with 16 heads the IOPS would increase by x16).

And even if the seek times got a little worse, the IOPS would still improve.

 

[cbn]

Here is the bottom of an Intel 750 2.5" NVMe SSD:

And the thickness is 15mm:

That is a good amount of cooling for a high throughput device that using NAND.

How do you think the heat would compare for a high throughput device using spinning platters?

 

[cbn]

A size comparison of two 2.5" hard drives (A 5400 rpm 500GB SSHD and a 15,000 rpm 600GB SAS drive) from this tear down article (Picture gallery with 40 images here) :

It takes 3 of the smaller 200GB platters (from the 15K SAS drive) to exceed the capacity that one 500GB platter (from the SSHD).

The following reasons given on this page explain why:

The Laptop drive spins at 5400RPM and contains 500GB of storage on one platter, the Enterprise drive contains 600GB storage over 3 platters. The platters are smaller on the Enterprise drive because it spins nearly 3 times the speed at 15,000RPM. Smaller diameter disks are less resistance to vibration. Vibration is a very important factor in server drives because if you have too much vibration induced by so many drives squeezed into the server chassis; vibrations can magnitude to the point that the head can no longer stay in line to the data tracks. By lowering the aerial density on the platters, a wider data tracks can be made, this makes it better for the head to stay on track at higher vibrations and higher disk RPM speeds, it also increases the data integrity required in mission critical applications that can be found in servers. The enterprise drive also features a much heavier/studier base plate. This also helps against drive vibration and also heat dissipation where multiple drives are mounted close together.

So with vibration being such an issue in a server rack perhaps this one reason why we don't yet see "Redundant array of inexpensive platters (RAIP)"? (ie, Keeping all read/write heads in contact simultaneously is more difficult than we think).

But perhaps with Helium (reducing platter flutter) it would do-able with the "thick and small diameter platters" at one of the lower levels of industry standard RPMs?

 

[cbn]

In pure sequential performance yes, however the spindles and heads are fixed relative to themselves. My (uneducated) guess is that seek times will remain unchanged or perhaps worse, which is probably why it hasn’t been done yet.

If seek times were unchanged, the IOPS would still increase by whatever number of read/write heads the drive had. (So for a drive with 8 heads the IOPS would increase by x8 and for a drive with 16 heads the IOPS would increase by x16).

And even if the seek times got a little worse, the IOPS would still improve.

Thinking about this more, I think the bigger problem(s) would be 1.) heat on the controller from having a high throughput (especially with older process tech) combined with the 2.) difficulty of all heads staying on track at the same time.

So that means that the RPM of the platters would likely have to decrease for at least those two reasons.

Low RPM will increase the latency.

Conceptually, latency is rather simple to understand; it is also easy to calculate. The faster the disk is spinning, the quicker the correct sector will rotate under the heads, and the lower latency will be. Sometimes the sector will be at just the right spot when the seek is completed, and the latency for that access will be close to zero. Sometimes the needed sector will have just passed the head and in this "worst case", a full rotation will be needed before the sector can be read. On average, latency will be half the time it takes for a full rotation of the disk.

For very small or small files I do wonder what the overall impact would be since the file is now read in parallel (due the striping across platter surfaces) rather than serially (as in a conventional hard drive). I suppose it matters what the file size is, the RPM and how many heads/platter surfaces are involved?