Jump to content


Photo

Speed difference with more platters.


  • You cannot start a new topic
  • Please log in to reply
27 replies to this topic

#1 spinal discord

spinal discord

    Member

  • Member
  • 3 posts

Posted 06 December 2002 - 09:32 AM

Let's say a manufacturer releases several models of a new harddrive line with 1, 2, and 3 platters. Are the 2 and 3 platter versions 2 and 3 times (respectively) faster than the 1 platter version? By faster, I mean the time to transfer a block of sequential data - disregarding seek time. I tried to find such data on this site, but it seems just the "flagship" models of greatest capacity are reviewed.

#2 Mickey

Mickey

    Member

  • Patron
  • 2,276 posts

Posted 06 December 2002 - 11:58 AM

If the density is the same and the mechanics are the same, you shouldn't see any significant difference between 1, 2, and 3 disks. Regardless how many disks you have, you're only reading/writing from one head at a time, not all of them concurrently.

You might see a slight performance improvement with the multi-disk versions because you will have more cases that you can do a head-switch between platters to find your data instead of seeking to a new cylinder and waiting for the servo to settle enough to start reading, but this would be nowhere near 2 to 3 times improvement.

#3 ncogneto

ncogneto

    Member

  • Member
  • 60 posts

Posted 06 December 2002 - 12:52 PM

If the density is the same and the mechanics are the same


If the density is the same then the drives would have different capacties. not quite sure how the mechanics culd be the same by any stretch as would not differnt numbers of platters require different mechanics?

While your answer is correct as you understand the question, I am not sure that this is what he meant to ask. Let me add a twist to the question as you understand it to conver both possibilities.

Assume that drive A and drive B have the same rated capacity. Drive A is a 2 platter design with higher density. Drive B although using lower density platters, has 4 platters as opposed to drive A's 2. The mechanics of each drive are comparable in the speed aspect. What can we expect when comparing the two drives?

#4 Mickey

Mickey

    Member

  • Patron
  • 2,276 posts

Posted 06 December 2002 - 01:51 PM

Assume that drive A and drive B have the same rated capacity. Drive A is a 2 platter design with higher density. Drive B although using lower density platters, has 4 platters as opposed to drive A's 2. The mechanics of each drive are comparable in the speed aspect. What can we expect when comparing the two drives?


I was sloppy with my earlier answer, for which I apologize. I should have said similar mechanical platform. When going between different platter configurations on the same platform, the only thing that *usually* changes is the number of heads on the actuator assembly. Some designs use balance weights to counteract the effect of not having as many heads. Others use an entirely different actuator.

To simplify the firmware and servo design, you usually don't want drastically different mechanical performance between your configurations (i.e. one config has much lower move time for the actuator than another one). In this sense, it's "the same mechanics."

Thus, if the mechanics are the same from a move time standpoint, and the spin speed is the same, with the only difference being different numbers of platters and areal density, I would expect the one with greater areal density to produce better performance. The actual total rated capacity is immaterial in this case.

#5 Voodoo Rufus

Voodoo Rufus

    Member

  • Member
  • 2 posts

Posted 06 December 2002 - 02:40 PM

So from my understanding drives will write on platter full first, then switch to the second platter next. So there should be a little burp in the read times from switching platters right?

Why don't drives do internal striping of the discs? They could get pretty fast then.

#6 blakerwry

blakerwry

Posted 06 December 2002 - 03:11 PM

So from my understanding drives will write on platter full first, then switch to the second platter next. So there should be a little burp in the read times from switching platters right?  

Why don't drives do internal striping of the discs? They could get pretty fast then.



No. Say you have a 3 platter drive, all data will be written from the outside of the platters to the inside of the platters and is spread equally across all platters.


When you look at an STR(sequential transfer rate) benchmark such as HDtach, ATTO, etc..., you sill see a sloped line that should be reletively smooth with few and minor jumps in STR.

If drives operated the way you said(fill a platter then move to the next) you would start off at a high number and go down as you get to the end of a platter. Then you would see the graph suddenly pop back up to the original high number, and again go down to the minimum number as you reached the end of the second platter... this would happen until you ran out of platters...
Posted Image

#7 blakerwry

blakerwry

Posted 06 December 2002 - 03:23 PM

as for the original question...

Greater density platters (read larger capacity per platter) means a higher STR.

Having more platters means you have a longer ways to go before your STR drops off.
Example: you have a 40 and a 120 GB drive using same density platters. The max STR for each drive is 60mb/sec and the min STR is 40mb/sec.

The 40GB drive will go from a max of 60 to a min of 40mb/sec in a 40gb transfer.

However, it will take 120GB's of data for this to happen for the 120GB drive.

One situation this would matter is if you have real time data that needs to be written to the HDD and you canot have this value fall below 50MB/sec. If you have 40GB of data to store, this would not be possible on the 40GB drive. But it would be possible on the 120GB drive.



If you have 3 drives of equal capacity, and each has a different number of platters then that can only mean that the platter density is getting smaller as the number of platters increases. Remember: lower density == slower STR's.




So, will having more platters make a drive faster(STR wise)? ...that depends on how much data you have on the drive compared to the drive's capacity.

Will having higher density platters make your drive faster? Yes.

Will a drive with less platters be faster than a drive with more platters if the capacities are the same? Yes.
Posted Image

#8 sechs

sechs

    Member

  • Member
  • 1,525 posts

Posted 06 December 2002 - 03:25 PM

Yes, the drive writes through one cylinder on each platter surface before moving on to the next.

Some manufacturers rate drives with fewer platters to have faster access times. Presumably they've gone through the effort of tuning the firmware for the lighter head assembly.

#9 Mickey

Mickey

    Member

  • Patron
  • 2,276 posts

Posted 06 December 2002 - 04:26 PM

Why don't drives do internal striping of the discs? They could get pretty fast then.


Technically difficult and financially impractical. The slight offset between each head with respect to its peers, plus the slight shift due to temperature changes and the like, means you can't have all the heads working at the exact same time for all cylinders. Going with multiple actuators could solve this, but then you've greatly increased the complexity of the drive and made it cost a lot more.

#10 ehurtley

ehurtley

    Member

  • Member
  • 870 posts

Posted 06 December 2002 - 05:34 PM

Why don't drives do internal striping of the discs? They could get pretty fast then.


Technically difficult and financially impractical. The slight offset between each head with respect to its peers, plus the slight shift due to temperature changes and the like, means you can't have all the heads working at the exact same time for all cylinders. Going with multiple actuators could solve this, but then you've greatly increased the complexity of the drive and made it cost a lot more.


Um, but that's exactly what they DO. (Read blakerwry's posts for more info.)
Motorola 68000@8Mhz | 1.5MB 30-pin DRAM | Single 800kB 3.5" Floppy
Intel Core Duo@2GHz | 2GB DDR2-667 SDRAM | Seagate 7200RPM 100GB 2.5" HD
Intel Core i7@4GHz | 6 GB DDR3-1600 SDRAM | Seagate 7200RPM 1 TB 3.5" HD

#11 Chew

Chew

    Member

  • Member
  • 1,975 posts

Posted 06 December 2002 - 06:58 PM

Why don't drives do internal striping of the discs? They could get pretty fast then.


Technically difficult and financially impractical. The slight offset between each head with respect to its peers, plus the slight shift due to temperature changes and the like, means you can't have all the heads working at the exact same time for all cylinders. Going with multiple actuators could solve this, but then you've greatly increased the complexity of the drive and made it cost a lot more.


Um, but that's exactly what they DO. (Read blakerwry's posts for more info.)


I think you misunderstood blakewry's post. What happens is the first part of the data is written through the first head, the next part through the next head, and so on. Once all heads have written to their matching tracks, it goes back to the first head and starts on the next track. Data is not simultaneously being written through all the heads.

#12 Voodoo Rufus

Voodoo Rufus

    Member

  • Member
  • 2 posts

Posted 06 December 2002 - 07:21 PM

So each platter gets a chunk of data written to it, but they read in the same way? Then why are drives like the WD800 slower than the WD2000 in the beginning of the drive and overall?

#13 spinal discord

spinal discord

    Member

  • Member
  • 3 posts

Posted 06 December 2002 - 08:44 PM

Why don't drives do internal striping of the discs? They could get pretty fast then.


Technically difficult and financially impractical. The slight offset between each head with respect to its peers, plus the slight shift due to temperature changes and the like, means you can't have all the heads working at the exact same time for all cylinders. Going with multiple actuators could solve this, but then you've greatly increased the complexity of the drive and made it cost a lot more.


Thank you (and everyone) for the very informative replies. Now that I know multiple heads do not operate simultaneously and a brief reason why, the question then becomes "how would this cost increase compare to buying two hard drives and putting them in a RAID0 configuration (assuming equal capacity in the end)?" I'd gladly pay extra for a harddrive with a sort of internal RAID than to implement it myself. In fact, Serial ATA has a lot of bandwidth and the next version will have much more. Besides increasing the density of the platters and their rotational speed, I see this as one "easy" way to fill all of this bandwidth.

#14 MaxtorSCSI

MaxtorSCSI

    Member

  • Member
  • 346 posts

Posted 06 December 2002 - 08:56 PM

HDD performance is dictated by three primary design criteria: recording density, rotational speed, and seek time.

Recording density dictates how much data can fit on a single track. A higher recording density design will have more data per track than a lower density design. This is important, because...

...Rotational speed dictates how many tracks can be read per second. A higher speed design will spin the disk more times per second than a lower speed design. For a given recording density, the higher RPM drive will be able to read the track it's on more times per second, which means it can transfer more data to/from that track in that second. Of course, reading the same track over and over again is of limited value, even if it does provide the fastest transfer of data. To be of value to the typical user, the drive also has to be able to seek...

...Seek time controls how fast the head can move from one location to another. For sequential transfers, we care most about one-track seek (and also head switch, since a head switch involves a short seek that's very much like a one-track seek). When the drive is seeking from one track to another, or switching heads, it can't be reading data. As a result, seek time represents overhead that degrades peak achievable transfer rates. Given the same density and RPM, longer seek times mean more time spent not reading/writing, which means lower over all transfer rate.

Within a given product family, SCSI HDDs are generally "common designs". For example, Atlas 10K IV (shameless plug: currently, the fastest 10K drive available according to SR) comes in 3 capacity points, 36G, 73G, and 146G. These drives share identical electronics, nearly identical firmware and very similar mechanics.

Mechanically, the three capacity points differ in the number of platters and heads (1&2, 2&4, or 4&8 respectively), as well as some optimizations for the rotary actuator (mass reductions for the lower head-count designs). Since the rotary actuators used by each capacity point have different masses/intertias, these differences affect the mechanical properties of the actuators, and the voice coil motors that moves them. The firmware's servo controller is "tuned" with different parameters for each capacity point so that performance is optimal for each.

So, where is all this leading? Back to the original questions:

Are the 2 and 3 platter versions 2 and 3 times (respectively) faster than the 1 platter version?  By faster, I mean the time to transfer a block of sequential data - disregarding seek time.


The answer is "yes & no".

The recording densities are the same for each of the 3 capacity points, the additional capacity is achieved by providing more tracks/cylinder. Since the amount of data on a single track is the same for a given cylinder, regardless of the number of platters, the "on-track" (meaning, no seeks or head switches) transfer rates will be identical regardless of the number of platters. *However*, seek times are NOT the same for all 3 capacity points. The lower capacity drive has fewer heads, so it has a lower inertia rotary actuator, so it can generally be made to seek faster. Since *real* data transfers usually have a need to seek, the lower-capacity drives *can* deliver higher performance by virtue of their faster seek times (the 36G A10KIV has an average seek time that's about 10% faster than the 146G version).

You might be tempted to buy a lower capacity drive because of this, but like so many other things in life, there's a catch. The length (circumference) of a data track varies as a function of the radius (that 2piR thing). This means that inner radius tracks have less data than outer radius tracks because they're shorter. Since transfer rate is a function of how much data is recorded on a track, inner tracks by necessity have lower transfer rates. As a result, in some regards a higher capacity drive can perform better than a lower capacity drive. When you transfer a 36GB file to/from a 36GB drive, you use the whole disk, and are on the slowest inner tracks by the time you get to the end of the file. If you transfer the same 36GB file to a 146GB drive, you're still at least 1/4 of the way in from the outer radius at the end of file (it's actually better than that, since area varies by the *square* of radius). The transfer rates at the outer radii are much higher than at the inner radii. The 146G drive performs better in this transfer than the 36G drive.

Storage Review (and many other bechmarkers) choose the highest-capacity designs to review in an attempt to provide the most even comparison between different suppliers products. However, you can usually count on the lower-capacity designs to perform better in workloads that involve predominantly random seeks because they usually have faster move times. The higher capacity designs provide better transfer rates as a function of GBs-used-on-the-drive (but not % of capacity) because of their higher data density.

I'd gladly pay extra for a harddrive with a sort of internal RAID than to implement it myself.


Drives can only read/write to a single head at a time because they only have a single Channel and Preamp (multi channel designs are *very* expensive). As far as I'm aware, no one currently makes a HDD that can simultaneously R/W multiple heads. Such a design would be cost prohibitive. This means you can't have an internal RAID design with any of today's HDDs (I think Seagate may have experimented with a dual-actuator design a few years back, that could read/write two channels simultaneously, essentially doing what you suggest. However, the intention was to reduce rotational latency more than to improve data transfer rate. Having to sets of heads 180 out of phase with each other is the equivalent of spinning the disk pack twice as fast, it cuts the average time to access a given sector by 50%. For random workloads, rotational latency is the primary limiting factor for performance so cutting it by 50% makes a huge difference. But, it wasn't cost competetive).

#15 Mickey

Mickey

    Member

  • Patron
  • 2,276 posts

Posted 06 December 2002 - 09:18 PM

Within a given product family, SCSI HDDs are generally "common designs".  For example, Atlas 10K IV (shameless plug: currently, the fastest 10K drive available according to SR) comes in 3 capacity points, 36G, 73G, and 146G.  These drives share identical electronics, nearly identical firmware and very similar mechanics.


The same holds true in the IDE world. Nothing like keeping as much stuff common as possible to help keep costs down.

Impressive post, MaxtorSCSI. :)

#16 Chew

Chew

    Member

  • Member
  • 1,975 posts

Posted 06 December 2002 - 09:23 PM

Thanks go to MaxtorSCSI for explaining more completely what others have tried to explain...

A question though. Do the lower head count drives truly improve seek times, even one-track seeks, by a full 10%? Is this a (well-)educated guess, or have you seen the numbers?

I'm curious on your thoughts on the 180GB vs 200GB question. I'm of the belief that the 66.6GB/platter gains it's capacity above the 60GB/platter by increasing the number of tracks, rather than the sectors per track (or both). But would that be by increasing the density of the tracks, or using more physical space on a platter?

#17 MaxtorSCSI

MaxtorSCSI

    Member

  • Member
  • 346 posts

Posted 07 December 2002 - 10:10 AM

A question though.  Do the lower head count drives truly improve seek times, even one-track seeks, by a full 10%?  Is this a (well-)educated guess, or have you seen the numbers?


I work in Design Engineering, in Maxtor's Server Products Group (the SCSI division). At least when it comes to Maxtor SCSI products, nothing I post is a guess! 8O

I'm curious on your thoughts on the 180GB vs 200GB question.  I'm of the belief that the 66.6GB/platter gains it's capacity above the 60GB/platter by increasing the number of tracks, rather than the sectors per track (or both).  But would that be by increasing the density of the tracks, or using more physical space on a platter?


Tracks Per Inch (TPI) is a "mechanical parameter" of the disk's design. It's a function of the physical specifications of the head (the spec's for the width of the read and write elements) and is set in the manufacturing process by a machine called a "servo track writer" (STW). TPI never varies within a given product family.

However, not all heads are created equally. We're talking about physical structures measured in widths of only a few angstroms. Fabrication of these structures is difficult to control "perfectly". A few molecules difference from one device to another can measurably affect performance. As a result, some heads will be able to achieve higher Bits Per Inch (BPI) than others. If you try to control the population of heads to a tight enough specification that all the heads will be equal performers, you end up throwing away a lot of otherwise perfectly good heads simply because they don't quite "make the cut". This can raise the costs of these components significantly (and disk manufacturing is all about costs).

We play a "statistical game" in the manufacturing process, pushing better performing heads to higher BPI and helping lower performing heads by reducing BPI. This allows us to yield more drives, significantly lowering costs for the end user (and also improving over all quality, since fewer drives need to be dissassembled and reworked to replace their heads).

The variation from head to head is generally quite small, and the statistical distribution is "normal", so on average drives come out of the process at their target capacities. Differences in capacity can be achieved either by allowing the drive to record at a higher BPI, or by allowing the drive access to more tracks, or by limiting the drive's access to fewer tracks.

The differences you see in in-family product capacities are the result of the statistics of the process.

#18 Chew

Chew

    Member

  • Member
  • 1,975 posts

Posted 07 December 2002 - 07:30 PM

Ok, so TPI is where the extra capacity comes from, as I believed. I just had the source of the increased TPI backwards.

So it's safe to say the 200GB and 180GB drives should have almost identical performance. I guess the 200GB may have the smallest advantage in terms of seek times due to the higher track density, but I would imagine it would be so small that it falls within the margin of error in a benchmark.

Thanks again MaxtorSCSI for a clear, concise post :)

#19 blakerwry

blakerwry

Posted 08 December 2002 - 01:10 AM

actually, I think that the seeks might be worse... as the heads would have to be more presise... People are demanding capacity at the price of performance. This seems to be a common situation in the newest super-high-capacity drives.
Posted Image

#20 tygrus

tygrus

    Member

  • Member
  • 116 posts

Posted 09 December 2002 - 07:41 AM

There will always be differences between the 'Benchmark' performance and real life usage.

Testing two harddrives (of the same family) at 100% capacity will often show a slower seek time for the bigger brother.

Testing a 200GB drive and its smaller brother of 80GB with a test file of 40GB will probably show the 200GB as better.

Here is a quick summary off the top of my head.
Each taken with all else being equal (average access time = seek + avg. rotational latency(=0.5 x 60 / rpm)).

* More tracks (TPI) = more capacity, more possible head positioning, requires more precise head placment, possibly slower acces time, faster STR if partly used, average STR (100% drive) no difference (<1%).

* More Bits per inch (linear not ^2, BPI) = more capacity, higher STR, less head movement for a specific size transfer (eg. 100MB on all tested drives), very little effect on average access time, requires better head/electronics to handle density and bit rate.

* More platters/heads = larger head assembly/motors, greater power consumption / drive runs hotter, larger inertia (motor/heads starts & stops slower), limits head speed, less T-T (track-to-Track) movements for a given file (eg. 100MB), larger capacity, slight difference in STR (

* smaller disk diameter = faster possible average access time, faster possible RPM, lower disk mass -> less power -> less heat, less capacity, smaller STR start/end ratio -> higher average STR.

* faster RPM = more power -> more heat, more noise, faster STR possible but limited by head/electronics, hard to maintain high BPI, faster average access times.

* more independent heads/actuators: Thats enough for now. IBM tried it a long time ago. Better to just buy two normal drives and run as RAID1 or RAID0

I probably missed a bit but you get the general idea.

#21 blakerwry

blakerwry

Posted 09 December 2002 - 08:36 AM

I think it might be good to point out that most drives use a compromise of 2 or more of these items to achieve their performance...

ex: SCSI drives use smaller platters to lower seeks and create less heat... then they increase RPM's to increase STR's... which pretty much cancels out any heat benefits of smaller platters.

IDE drives have RPM's in the slower specturm but maintain high STR's because they increase density (Bit per inch).
Posted Image

#22 Jan Kivar

Jan Kivar

    Member

  • Member
  • 1,190 posts

Posted 10 December 2002 - 03:05 AM

I read from somewhere that 15k SCSI drives use 2.5" media because the platter structure of 3.5" media is not rigid enough to withstand the stress that the high rotational speed causes. So one must use 2.5" media in high RPM drives. I think that 10k drives don't have platters as big as 7.2k IDE (of SCSI) drives have. Physical sizes that is. Care to share info of the physical media (platter) sizes of Maxtor HDs, MaxtorSCSI?

The main reason to use 15k rotational speed is to lower the average rotational latency. There has been little change to seek times throughout the past 5 years (given a size of media), so "best" solution to achieve faster access times is to lower the latency, thus increase RPM.

The downside to high rotational speed is that one must use low platter density (BpI), as the head cannot read from/write to platter fast enough to maintain 1:1 interleave. (MaxtorSCSI, correct me if I'm wrong...)

Cheers,

Jan

#23 Mickey

Mickey

    Member

  • Patron
  • 2,276 posts

Posted 10 December 2002 - 11:37 AM

15K drives use 2.5 inch media. 10K drives use 3 inch media. 7200 rpm drives and slower use 3.5 inch media. This is largely due to flutter effects from the high spin speeds (think of a sheet of paper flapping in the breeze, then imagine if you shorten it how the magnitude of the flapping at the end is smaller). Another way to counteract this is with thicker media, but this adversely affects shock/vibe performance and reduces the number of platters you can fit in a drive.

#24 MaxtorSCSI

MaxtorSCSI

    Member

  • Member
  • 346 posts

Posted 11 December 2002 - 08:42 PM

Those last few posts are pretty much 100% correct. I was going to say much of the same stuff, but there hardly seems any point!

:wink:

#25 blakerwry

blakerwry

Posted 12 December 2002 - 02:56 AM

Dont forget guys, that a smaller physical patter means that the heads won't have to seek as far on a full stroke or half stroke seek... this probably helps decrease seek times a bit.
Posted Image



0 user(s) are reading this topic

0 members, 0 guests, 0 anonymous users