Seagate Demos World's Fastest Hard Drive, Doubled Performance, HAMR Progress

[photonboy]

Idk. Unless it is new data that is being written, or if it is able to somehow always split the data between the two heads (Which is what I would think kind of like a RAID 0 almost AND NOT I"M NOT SAYING ITS A RAID 0 just that same kind of thinking) then you could get that double data write. But i mean if ALL your data is on on one platter then you are NOT going to hit those speeds. I just see this drive coming out, they say this, you get it, doesn't even come close because of how the hard drive gets used. Unless it knows to split the data between the two heads you will never see that "More than double bandwidth" it is talking about. If it starts to write data, and you need to pull data from that same platter its going to be just a normal drive.

Now what I think would have been a BETTER idea, yes it would be a LOT harder as it would make the driver bigger (Or i think longer) Is to put in a SECOND full head. Each head acts as they do now. This way each head has access to all the data. Now I could see there being issues as well. They both need to write. Well one head will get to a empty sector before the other etc so the hard drive would then have to do a lot of ahead type thinking possible slowing it down but I would see that as a more valuable product even though it would break HDD Physical size standards.

First, Seagate makes the fastest HDD in the world but you have a "BETTER idea"?

And a second head per platter has already been tried. Many things have been tried over the decades. Here's one issue:

"Simultaneously aligning all the heads on all the platters isn't possible because of the incredibly thin data tracks on the platters, so only one of the heads is actively reading or writing data at any given time. That limits read/write throughput and performance with randomly accessed data."

I took that to mean you can't simply split a BYTE of data (eight bits) across eight heads (four platters) thus reading or writing one byte at an entire point in time.

So thus it appears we're limited now by the time the ACTUATOR ARM takes to move a head to a new location. Thus we need a second actuator arm which was an OBVIOUS solution years ago but in practice very, very hard to implement due to the speed and complexity of this.

Just replace "head" with "arm" and your great idea about dual heads is the same concept. You could ALTERNATE the data such that Actuator 1 is reading a bit whilst Actuator 2 is moving to the location of the next bit for the application or data file.

(A general rule-of-thumb with engineering is that if you have a "better idea" than engineers in an area you have little to no experience it's probably either a dumb idea or it's been though of)

UPDATE:
I admit I don't know much about HDD's, so as per the above QUOTE it's not clear to me if they actually do coordinate writing of multiple HEADS currently for sequential data but NOT for random data or always write with only one head. I tried to look that up but couldn't find any obvious answer.

 

[TJohn]

I won't be buying any mechanical drive again. Hasn't that been the goal for 50 years?
The old pre-winchester multi-platter drives in the 770-80s had head pairs for each platter...but the voice-coil actuators could not compete with winchesters.

 

[USAFRet]

I won't be buying any mechanical drive again. Hasn't that been the goal for 50 years?
The old pre-winchester multi-platter drives in the 770-80s had head pairs for each platter...but the voice-coil actuators could not compete with winchesters.

Well...I'll not be filling up my NAS box with SSD's.
4 x 4TB is a bit tough on the old wallet.

The PC's, OTOH, are all SSD only. (except for the old cheesy 9 year old laptop)

 

[jamesltoa4]

Nice idea especially for multimedia and backup storage or virtual systems.
As in things that are created and then either infrequently used, or copied as a sequential read process to another area.
BUT
there seems to be no indication as the the manner that the drive areas would be split - as in
Currently rotating media tends to be faster at the beginning of the drive sector/block 'addresses' so I tend to specify partitions so that the random access and most used files go in a partition at the beginning and the write once - read once large files go at the end - as in 4TB drive 3 partitions OS as first 500, working data as the next 1TB, and the rest for large infrequently accessed files.

That also goes some way to dealing with the NTFS problems where all the partitions file entries are in 1 large hive thatn can take much longer for the OS to work through and rename 1 file than to copy a 10MB file over a USB2 link to another drive.

So first question - are the LBA addresses assigned as separate series for each set of heads so I could split the partitions into blocks using separate actuators for each partition.

Then - these are Seagate drives - Last time I looked at the site the Seagate datasheet for retail drives indicated how many were shipped on a pallet and the pack weight
The documentation available that I managed to get before Seagate declared it to be secret company confidential
included:
The company does NOT indicate -
Which drive will be in a retail USB case
The rotational speed or head movement speeds of drives
Power usage - normal use, startup or idle, or sleep modes

I did get details such as
Speed of the data recording is that defined for the interface -
as in the drive reads and writes at SATA-2 or USB-3 speeds, rpm cache etc. have no bearing on the drives data handling speed.
The error rate for an 8TB drive is 1 in 10^14 bits - which seems to be about expect 1 error in every 10TB written. and 1 unreadable unrecoverable byte means the entire backup set would be suspect.
Max usage of the drive is 55TB a year
Max usage time of the drive is 2400 hours - Yes Seagate drives should only be used for 100 days
Now - I do realise there are drives with better ratings than those generally sold as retail units for general PC use -

But I am not able to recommend those to my small-business clients as I would be expected to do the install, and consequently accept the liability for the expected failures -
A totally indefensible situation as the tech specs for the readily available as retail items drives indicate that as far as backup or general PC use, they are NOT-FIT-FOR-PURPOSE.

Now That is the result of my interaction with Seagate - I do hope that most others have been supplied with much better specifications and documentation.

 

[jamesltoa4]

Re earlier post - typo - missed the second 0 in
The error rate for an 8TB drive is 1 in 10^14 bits - which seems to be about expect 1 error in every 10TB written

 

[cryoburner]

So first question - are the LBA addresses assigned as separate series for each set of heads so I could split the partitions into blocks using separate actuators for each partition.

Considering they are advertising double the sequential throughput of other drives, it seems likely that they are splitting data between the two sets of platters, much like RAID 0, only internally within the drive. In practice, I suspect it would position the data much like any other drive, with the best performance at the beginning of the drive's addressable space in the outer regions of the platters, and both heads generally reading data from a similar region of their respective platters at once. Even on existing multi-platter drives, partitions should be split across all platters.

 

[jamesltoa4]

So first question - are the LBA addresses assigned as separate series for each set of heads so I could split the partitions into blocks using separate actuators for each partition.

Considering they are advertising double the sequential throughput of other drives, it seems likely that they are splitting data between the two sets of platters, much like RAID 0, only internally within the drive. In practice, I suspect it would position the data much like any other drive, with the best performance at the beginning of the drive's addressable space in the outer regions of the platters, and both heads generally reading data from a similar region of their respective platters at once. Even on existing multi-platter drives, partitions should be split across all platters.

 

[jamesltoa4]

Thanks for the response - and I agree that is what I would also expect - HOWEVER, I was considering the detail in the article and what Seagate staff have told me, and what they have removed from their site.
So - reading the bit about being able to address the parts separately, my first question is ARE the LBA's assigned to each track of all the platters working across the discs, meaning that a partition would be using the 2 actuators for the space, or are the LBA's assigned to the tracks of each actuator head set such that a separate partition can be specified for those platters accessed using actuator 1 and another for those using actuator 2.
Having only 1 head for each platter surface - with those heads mounted on a common pillar/bearing means that there is actually little reduction in latency as there is still the wait of up to 1/rpm for the data wanted to come round to the head - while the earlier design of 2 actuator sets meant that, with 2 heads per surface, and those being at about 180° away from each other on the platter, the latency was halved. This design will only (my understanding) slightly reduce the read/write wait time as the in modern drives, head movement time has been so much less than the rpm delay that the drives were (theoretically) able to use data request queue management to decide to reposition the heads to process a read/write from other tracks while waiting for the earlier requested data to come round again.

That would allow for a little bit more data recoverability (in the event of an actuator failure) and better throughput doing sequential read and write.
My understanding of the point of RAID being to give extra speed and address the likelihood of a drive failure. So having the data on 1 drive does not really seem to help with that.
Additionally, having been assured by Seagate customer support and Tech support that the speed of the drive is the interface speed, and cache, rpm and seek time have no bearing on the speed, I do not see how they can say the drive is faster - unless they are using a new interface for the drive connection to the PC -
Note that is based on what Seagate staff told me, NOT on my understanding of the war a hard drive works!

So - the description and my prior contact with Seagate leaves me wondering how much of the claims are actually going to be effectively usable within a windows - PC or server environment.

 

[USAFRet]

My understanding of the point of RAID being to give extra speed and address the likelihood of a drive failure. So having the data on 1 drive does not really seem to help with that.

The 'point' depends on what RAID type you're referring to.
They are NOT all the same
RAID 0, striped. Speed. You're addressing two drives at the same time, so not relying on a single head or actuator arm.
Absolutely zero consideration for drive fail. The data is striped across both drives. Loss of either drive means loss of all data.

RAID 1, mirrored. Redundancy in case of drive fail. Performance is about the same as a single drive, but in the event of a physical drive fail, the system can limp along on the remaining drive. Until you can replace the failed member.
But only helps in case of a physical drive fail. The user and the OS only see a single 'volume'. Any file deletion simply happens on two physical drives at the same time. Delete a file, and it is gone.


This Seagate would seem to be an internal RAID 0 type of thing. If the drive dies, it dies. Your stuff is gone.

 

[cryoburner]

My understanding of the point of RAID being to give extra speed and address the likelihood of a drive failure. So having the data on 1 drive does not really seem to help with that.

Well, RAID 0 is not actually RAID in the traditional sense, in that it doesn't provide any redundancy to protect against drive failure, and actually increases the chance of data on the drives getting lost. It's all about increasing sequential performance by splitting data across two drives and having each drive read part of the data at once. These drives are likely doing something similar, though I suspect they could probably be made more efficient at it than two separate drives with an external controller. Latency probably won't be much better than existing drives though, seeing as there's still only one head for each platter surface. For latency-critical applications, an SSD would probably be better though, anyway.

Additionally, having been assured by Seagate customer support and Tech support that the speed of the drive is the interface speed, and cache, rpm and seek time have no bearing on the speed, I do not see how they can say the drive is faster - unless they are using a new interface for the drive connection to the PC -

Yeah, you were probably communicating with someone who didn't know what they were talking about. : P As for the interface speed, the SATA 3 interface is capable of transferring files from a drive at up to nearly 600 MB/s, while existing hard drives don't get much faster than 200 MB/s. So the interface isn't what's limiting the speed, and there's certainly room to increase sequential performance by reading a file that's split across two platters using two heads at once. It's just not common for modern drives to do that.

Of course, it's difficult to say what real-world performance might be like until these drives get tested by third-party reviewers. Much like with the sequential speeds of existing drives, that "480 MB/s" will undoubtedly be a best-case scenario, and random accesses will likely bring performance far lower, and closer to what other drives can do under many usage scenarios.