TechSpot

Curious about which part of hard drive has bigger impact on performance

By Jskid
Mar 9, 2011
Post New Reply
  1. I notice some hard drives with larger buffers or really big capacities have slower RPM (e.g. 5400). Is the RPM not important or something?

    BTW since there's lots of people with actual problems posting on this forum, is there anything I should put in the title to make clear my questions aren't urgent?
  2. Lokalaskurar

    Lokalaskurar TS Enthusiast Posts: 616

    I would say: for low boot-up times, RPM matter a lot since the buffer needs to be loaded from scratch. But after that, the buffer matters a lot. Of course, what also matters is the data-transfer speed depending on the type of connection and the read/write properties of the hard drive.

    So well, the RPM is in fact important, but hardly noticeable "on the short run." Those extra seconds spent booting and loading files etc will eventually become minutes and maybe hours of your life spent just waiting for something to happen.

    Lowering the RPM is thus an excellent way of reducing the price of big-storage drives of big-buffer drives.

    And what matters the most on the long run, is probably the capacity of the drive. Filling it up too soon is never good for anyone :p
  3. hellokitty[hk]

    hellokitty[hk] Hello, nice to meet you! Posts: 4,367   +125

    I think consumers will be most concerned with RPM; boot times, load times, wait time.
    IMO most important in order is basically sequential read, then random write, then random read, and finally sequential write.
  4. jobeard

    jobeard TS Ambassador Posts: 13,403   +314

    As so many systems have some degree of fragmentaion, ask yourself, what disk I/O happens most frequently? - - random reads!
    After a good defrag, sequental reads should dominate.

    Boot time is nice if it is really low, but as that occurs only ONCE per session, so don't get myopic on that metric - -
    focus on what occurs most frequently - - imo, lauching any program[*].

    Some background on HD performance.
    There are three measurements for a single I/O:
    • move the arm to the proper location, (ie SEEK)
    • let the platter rotate to the right secord (ie rotational delay)
    • DRM the data into (out of) memory, (ie read/write)
    Years gone by when we had multiple platters on one HD, there was a HEAD SWITCH that followed the SEEK.

    If the SEEK costs 100 units, the rotational delay is typically 10 units and the final read/write will cost 1 unit.
    Seek times are relatively fixed, and the read/write time is trivial compared to the other two.
    It ends-up, that rotation delay is inversely related to RPM; the faster the drive, the lower the delay.
    Laptops usually opt for 5400 rpm as it saves battery life.

    There's a great tool Defragler, which allows you to pick precisely which files get compressed into contiguous sectors.
    Run that under an Admin-ID and defrag EVERY DLL or *.EXE that is more than one fragment.
    This will vastly improve [*] noted above.
    Now that the programs have been optimized, run a full defrag to fix the other stuff.

    Oh yea, don't minimize the degradation that a fragmented pagefile will cause you!
    Look for PageDefragand run that once in a while :)
  5. hellokitty[hk]

    hellokitty[hk] Hello, nice to meet you! Posts: 4,367   +125

    That's true, but I think (I will say IMO again) that boot-up times are what people notice the most. Most people are already used to slight delays attributed to slow random read, and won't notice their presence; but they'll probably notice their absence when using a quick SSD.

    That's pretty hard to rate sequential read and random read though; they're both quite important.
  6. jobeard

    jobeard TS Ambassador Posts: 13,403   +314

    if you add SSDs to the mix - - lots of new assumptions arise, some of which have proven false, eg SSDs will never 'wear out'; they have an MTBF that is very similar to good old HDs.

    Without the Seek & Rotational delays implicit in the HD, we all assume that
    a read or write to any location is always at a constant time ~= the DMA time.
    I would really like to see some study data on that assumption :)


Add New Comment

TechSpot Members
Login or sign up for free,
it takes about 30 seconds.
You may also...


Get complete access to the TechSpot community. Join thousands of technology enthusiasts that contribute and share knowledge in our forum. Get a private inbox, upload your own photo gallery and more.