Mechanical HDD speed

[Deleted member 24505]

Are mechanical HDD's read/write limited by the disc speed or electronically? Just wondering.

 

[Shrek]

I believe it is disk speed and data density, although write can be slowed by shingling.

Of course, if one remains within the cache size, one can exceed this limit.

 

[Deleted member 24505]

I guess it is non viable to make faster mechanical drives then?

 

[Shrek]

I think 15,000 rpm hard drives exist but increasing the data density might be a better way to get better transfer speed.

Even better, put in lots of heads that can work in parallel.

Or have a large (8GB) solid state cache.

 

[dragontamer5788]

Are mechanical HDD's read/write limited by the disc speed or electronically? Just wondering.

If you make a 7200 rpm drive, it will always spin at 7200 RPM. (It can't change speed). Historically, drives went to 10,000 RPM or 15,000 RPM or beyond, but SSDs were a better idea. 7200 is the fastest we got today, and some people prefer 5400 RPM so that the drives make less noise and use less power.

All hard drives pretty much read/write at the rotation speed. Every hard drive has a number of "bits per inch", and they can read/write to it at that speed. (The outer-edge has faster bits, because of how radius / circles work / simple geometry... while the inner-edge has slower bits)

So mechanical-ish. 7200 RPM isn't a hard limit, its just what people are happy with in practice. If you want faster, SSDs exist.

---------

CD-ROMs and DVDs decided to go with "bits-per-degree of rotation", so that the inner-bits have the same speed as the outer-bits (consistent read/write speeds is more important for movies and other such products). But that means that the "outer-bits" take up more space than the inner bits. Just a funny quirk that shows you how these little engineering decisions make a physical effect on the real world.

 

[Shrek]

Imagine 5 platters with all 10 heads working in parallel; one could get a lot of extra speed that way.

 

[dragontamer5788]

Imagine 5 platters with all 10 heads working in parallel; one could get a lot of extra speed that way.

Parallel heads on different platters makes sense (5 heads on 5 platters).

Multiple heads on a single platter looks like an incredibly difficult problem however (ex: 2-heads on one platter). How do you know if the information you read was for head#1 or for head#2 to read? What if head#2 is 5-microns to the left and is therefore reading data about 500kb away than expected? How do you keep the heads synchronized? Or if they're unsynchronized, how do you ensure that they won't get in each others way? What if one head is writing and the other is reading, would the data be borked?

--------

Seagate is beginning to split up the heads so that they can work independently.

This will be seen as two hard drives on one cable to the operating system... and that's really all this is. Instead of having a 8-platter drive with 8-parallel heads, it can be seen as 2x 4-platter drives with 4-heads each.

This saves on on motor / case / packaging costs and increases the amount of "random-I/O" the drive can do, but only across two different sets of data. Its a bit complicated but it should get the job done if we rewrite our software to take advantage of it...

I guess it is non viable to make faster mechanical drives then?

Faster in what way?

Hard drives have plenty of bandwidth. The issue is that people want more random iops, which is limited by the physical arm moving to-and-from the right positions on the platter, as well as the ~4-milliseconds it takes for a 7200 RPM drive to spin around to the correct location to read/write.

We can fix that by spinning faster and adding more read/write heads, but that increases cost and complexity. Furthermore, this would only double or quadruple random-iops... SSDs are 1000x more IOPS than a hard drive. Just making things 2x faster or 8x faster isn't going to be competitive against SSDs.

 

[Mussels]

It's access time, rotation speed and data density

If the density increases, they can read faster per sector
If the rotation and access times speed up, they can change between sectors faster.

Having to wait for parts to physically move, is simply a huge hamper to speed.

 

[cvaldes]

The biggest limitation with rotational HDDs is the fact that the drive head can only be in one place at a time.

That's why Oracle DBAs and adult site operators had massive disk arrays back in the Nineties.

Kevin Looney who wrote Oracle administration manuals in that era proposed something like a 22 disk solution for the Oracle RDBMS (circa version 8). The operating system would go on one disk and each of the 22 major Oracle files would go on a separate disk instead of everything on the one disk. That way you don't have to wait for a log write to finish before you wrote to a rollback segment.

SSDs don't have this limitation.

There's a point of diminishing returns in increasing rotational speed, stacking platters, or changing the diameter of the platter. Remember that at the same rotational speed, the center of the disc moves much faster than the edge of the disc in linear speed. Increasing the diameter to something like 8" might seem like a good idea but then you run into challenges in moving the drive head fast enough.

This leads us to how drive performance is measured. A spinner today isn't so bad for a linear read/write of a large contiguous file, let's say a 10GB video file. However HDDs are vastly inferior for random access operations. So depending on the test, sometimes the HDDs don't look so bad. Other situations they are terrible.

 

[Deleted member 24505]

I get it, just wondered. I guess mechanical has reached a pinnacle. I only use them as external now anyways, way too slow.

 

[dragontamer5788]

I get it, just wondered. I guess mechanical has reached a pinnacle. I only use them as external now anyways, way too slow.

Oh its not so bad. Just get one of these babies.

SC946ED-R2KJBOD | 4U | Chassis | Products | Super Micro Computer, Inc.

946ED-R2KJBOD

www.supermicro.com

When you have 90 hard drives working in parallel, they are in the aggregate pretty fast.

 

[RJARRRPCGP]

Actually, Western Digital was manufacturing Velociraptor HDDs that use about the same amount of amperes (IIRC) (or watts) as a 7K HDD! The solution, apparently, was to shrink down to 2.5-inches, just like laptop HDDs.

The heatsink kit is what makes them fit in a still standard 3.5-inch bay.

The new ones, aren't like the ones of the "first decade".