Jump to content
Search In
  • More options...
Find results that contain...
Find results in...
Doom Marine

The uberatomic geekwizard e-peen show-off-your-PC-threads!

What is your PC?  

102 members have voted

  1. 1. What is your PC?

    • Intel Quad Core
      17
    • Intel Dual Core
      28
    • AMD Quad Core
      15
    • AMD Triple Core
      1
    • AMD Dual Core
      14
    • I`m a Mac user, this button turns my computer on!
      1
    • It runs Windows XP and that`s all I`ll ever need!
      8
    • I only play Doom and it`s still the early 90`s!
      3
    • Pentium III all the way Baby!
      2
    • Buttsecks!
      7
    • I stole a Cray supercomputer
      1
    • I have a WOPR taking up most of my downstairs
      4
    • Computer? Isn`t that one of those newfangled electronic calculating gadgets I`ve been hearing about lately?
      1


Recommended Posts

Maes, maybe you should actually try using a dual 10000 RPM drive RAID0 array before getting all long winded. The difference between that and a single 7200 RPM drive is quite noticeable even for a non-gamer such as myself.

Share this post


Link to post

For me, the difference between a single 7200rpm hard drive and two of them in RAID 0 was VERY noticeable in pretty much every task that would normally be waiting on my hard drives.

Share this post


Link to post
Maes said:

Maybe it's just me, but won't decenct caching on many levels, starting from the OS itself, nullify the "zillion file zerging" problem, unless you're accessing many files AND they are cumulatively so large as to trash the cache? There are also solutions like ReadyBoost or eBooster for Windows XP that allow caching some data on a flash/separate HD.

From what I'm seeing on my Disk Activity monitor, caching may help with reading files, but it still won't help with writes. If this was a mechanical drive, that head is going to move back and forth hundreds of times, rearranging data which causes defragmentation.

Notice the Disk Activity image a few posts back, there is nearly zero read on the small files, which is implication that those data are effectively cached by 256 MB of DDR3 on the SSD. Disk Activity writes, on the other hand, is numerous and constant. These random access file writes leads to defragmentation on the disk that no cache can effectively control.

Another limitation to caching is the employment of algorithms that can't account for all usage patterns, which makes it fallible in principle, and that bothers me.

ReadyBoost and EBooster are merely inelegant workarounds with flash drives, and if I had to take it to that length to extend my system's cache with some flash drive dangling out of my USB port, I might as well go the whole nine yards with an honest SSD while upgrading my OS to Windows 7 in the process.

Maes said:

IMHO 10000+ RPM HDs are more of a proof of concept as in "Yeah, if you sacrifice capacity, power consumption and overall lifetime and reliability and manufacture them with tighter tolerances and at a premium, you can end up with a HD that shaves a hair off access times". Pretty much like "audiophiles" are prepared to pay for ridiculously expensive equipment with always diminishing returns.

I think you've understated the margins of improvement of a VRaptor over a 7,200 RPM drive. Not only is the access time faster, but the the areal density (it's in 2.5" platter format) combined with faster rotation gives it a big improvement over most pedestrian 7,200s.

With that said, the VRaptors, just like their namesake, are becoming a dinosaur, it's nowhere as fast as an SSD, and nowhere as generous as a 1TB conventional drive... it's hovering in that limbo niche where one isn't quite sure what to do with the damn thing; it's answering a question nobody is asking =p

Maes said:

If the best possible access times a single disk can deliver WITH NO CACHING is really what's desired then yeah, there's only so much you can achieve with a mechanical hard disk, and any solid state medium will beat it squarely. But that would be like e.g. insisting on getting the fastest possible single-core processor for a given task: unless the problem is broken, ill-posed or pathological, it could probably be served better by less powerful cores working in parallel. And trying to get maximum performance from just one disk IS ill-posed, no matter what.

Real world data shows that current SSDs has a large advantage over hard drives in all levels of read and write, and more importantly, it doesn't take a benchmark to feel the immediate effect of near-instantaneous seek time. If you ever have a chance to use Windows 7 on a SSD for a week, it will become apparent that the SSD, when compared to mechanical drive, is a whole different beast altogether.

Maes said:

I don't think any sane datacenter would purchase either arrays of velociraptors, nor arrays of SSDs to achieve low access times: they'd buy good CONTROLLERS (that means NOT the integrated "SATA RAID" crap on mobos) with plenty of expandable onboard cache RAM, connected to inexpensive disks. With the saved money, they can instead invest on redundancy and multi-level striping, although with a really good controller with NVRAM and autonomous power you can keep thousands of cached read and writes before finally writing stuff to disk.

While we can both agree that VRaptors are out of the equation, I'm not sure if SSDs can be counted out of data centers just yet.

Correct me if I'm wrong, but the last time I checked, onboard cache RAM is way more expensive than SLC drives per GB. Another thing to consider, is that it's been determined that the current generation of SSDs are overall more energy efficient than a "green" drive. The prices are dropping fast too.

--- --- ---

So far, our discussion have been qualitative and is in dire need of some real hard numbers:


4K read is random
4KQD32 = 4K Queue Depth: 32 (measures the 4k cached reading speed)
- 33.06 MB/sec on 4k random read = ~8463 IOPS
- 82.33 MB/sec on 4k random writes = ~21076 IOPS
- 200.0 MB/sec on 4k cached read = ~51200 IOPS
- 135.4 MB/sec on 4k cached write = ~34662 IOPS


4K read is random
4KQD32 = 4K Queue Depth: 32 (measures the 4k cached reading speed)
- 0.805 MB/sec on 4k random read = ~206 IOPS
- 3.032 MB/sec on 4k random writes = ~776 IOPS
- 3.942 MB/sec on 4k cached read = ~1009 IOPS
- 4.316 MB/sec on 4k cached write = ~1104 IOPS


4K read is random
4KQD32 = 4K Queue Depth: 32 (measures the 4k cached reading speed)
- 0.921 MB/sec on 4k random read = ~235 IOPS
- 1.981 MB/sec on 4k random writes = ~507 IOPS
- 2.832 MB/sec on 4k cached read = ~724 IOPS
- 2.002 MB/sec on 4k cached write = ~512 IOPS


4K read is random
4KQD32 = 4K Queue Depth: 32 (measures the 4k cached reading speed)
- 0.711 MB/sec on 4k random read = ~182 IOPS
- 1.628 MB/sec on 4k random writes = ~416 IOPS
- 1.913 MB/sec on 4k cached read = ~489 IOPS
- 1.703 MB/sec on 4k cached write = ~435 IOPS

As the numbers shows, especially the last one, my mechanical drives are generally "fast enough" with caching on, but it's nowhere near the ballpark of my SSD.

Share this post


Link to post

Lots of stuff, with most of which I agree. Just let me "dissent" on a couple:

Doom Marine said:

Correct me if I'm wrong, but the last time I checked, onboard cache RAM is way more expensive than SLC drives per GB.


Huh? "Cache" is only a misnomer in this case, as those controllers generally use standard SDRAM modules, not actual flip-flop based static cache RAM. Sure, they might have some of that too, but when they boast 1 GB or more of "onboard cache" they surely don't mean CPU-grade cache.

Doom Marine said:

Benchmarks


OK, we can definitively agree that a solid-state medium with totally random access will be faster on haphazardly random reads/writes than even the best conventional hard disk.

The question is: how typical is that? What application would require such a sustained read/write performance with those conditions, if not a major datacanter supporting something like Google or Facebook? And then again, resources would probably be better spent optimizing the undelying database layout, instead of just throwing hardware at the problem. Sure, a SSD can be seen as a "RAM like" storage that nullifies the access time problem, but the above is about the only type of application I can think of that would REALLY need an SSD (or alternatively, large amounts of cache RAM and VERY delayed write policy).

Let's see some others:

  • Loading an OS: it's not like that on an unfragmented drive. Running it from an SSD will just mask the problem.
  • Loading applications and games: nope, similar reasons to above.
  • Running applications and games: shouldn't be. If I find a game that's made out of 200000 tiny files however (and there are some), I'd hesitate to install it: the designers should have came up with a better way of organizing resources.
  • Recording/playing video and large streaming data: that's where large, space-generous HDs are best at. SSD space premium is a serious handicap here, e.g. in some products like Flash-based camcorders.
Then again, this is a thread about hardware e-peen...not about the most cost effective business-grade compromises ;-)

Share this post


Link to post
Maes said:

  • Loading an OS: it's not like that on an unfragmented drive. Running it from an SSD will just mask the problem.

From my experience, unfragmented drives and OS don't exist in the same place. The day after my OS partition is defragmented, it just reverts back to ~20% fragmentation from random writes and reallocation of files, and this is with pagefile.sys disabled. I'm not clear about the masked problem you're referring to...

Maes said:

  • Recording/playing video and large streaming data: that's where large, space-generous HDs are best at. SSD space premium is a serious handicap here, e.g. in some products like Flash-based camcorders.

Mechanical drives won't be obsolete for a long, foreseeable time. SSDs, of course, are supplementary to getting the best out of both worlds.

Maes said:

Then again, this is a thread about hardware e-peen...not about the most cost effective business-grade compromises ;-)

I'm surprised by the number of members having quad cores, I thought everyone on Doomworld is still in the 90's. Statistically speaking, 8% of those who voted are still in the XP era, and 7% shunned e-peen in favor of buttsecks... whether those persons want to be the pitcher or catcher is not made so clear =p

Share this post


Link to post

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now
×