WD drive with 8MB cache - is the hype justified?

[Tea]

I've been watching the hysterical praise the 8MB cache WD 120GB drive has been getting around the traps, and wondering if it is really half as good as people are making out. It just seems a little too good to be true. How the hell can an IDE drive with a 7200 RPM spindle and a seek time in the 9ms range hope to compete with a 15,000 SCSI drive in the sub-4ms class?

I've been curious about that for quite some time, but had no opportunity to find out for myself, as no-one is buying drives in that size range around here. I mean, I've sold about 4 60GB drives and exactly two 80GB Western Digitals. Everything else is 40GB and below. So I don't get to play with the big ones.

Last night I found myself having to install apps onto my G: partition, which is supposed to be my "miscelaneous and nearly empty, put things that don't belong anywhere else" partition on my home box - because I'm almost out of room on the other drive letters. The XP 1700 has two drives at present, an old 9.1GB Quantum Atlas IV 7200 RPM SCSI and my beloved 18GB Cheetah X15.

I need about 20GB more than that. Or else to have a proper tidy-up. Easier to just add an extra drive. Ideally, I'd buy a 36GB X15-36LP but it would be completely stupid to waste that much money on (depending on which way you look at it) saving some tedious disc management, or putting in a very high-performance drive for long-term storage of the stuff I have that almost never gets accessed.

The sensible thing, then, would be to take a 20GB 5400 RPM Samsung home and add it on, or better yet, make it a 40GB drive and replace the Atlas - I could use the Atlas to replace the 4.3GB Fireball in my burner box, which is the only non-SCSI system I have at work. For this long-term storage stuff, speed is completely unimportant, and reliability matters. (Yeah, I have nearly everything on CD - but which CD? )

No prizes for guessing then. I just ordered myself a 120GB Western Digital with the 8MB cache. I have absolutely no idea what I'm going to do with 120GB of extra storage! (Or "only" 100GB if they are out of the 120s - I said on the P/O to just send whichever one they had in stock.)

But I'll be very interested indeed to see how it stacks up against a real performer. WD12000JB vs Mark 1 X15. I'm looking forward to this!

 

[timwhit]

Be sure to let us know what you think Tannin!

 

[Tea]

Who cares what Tannin thinks?

I'll let you know what I think - which is much more important. :wink:

 

[Prof.Wizard]

It just seems a little too good to be true. How the hell can an IDE drive with a 7200 RPM spindle and a seek time in the 9ms range hope to compete with a 15,000 SCSI drive in the sub-4ms class?

I don't think anyone said that. (And if he did, he was wrong!)

It can compete with 7200 or 10000 RPM SCSI drives, but not with 15000 for sure...

Oh, and the hype is true. The drive rocks and I still can't understand why other companies haven't announced models bearing an increased cache.

 

[LiamC]

Cache

What is the purpose of cache? To stop the processor going to memory or to stop the hard drive from actually having to read the disk.

What does this mean? It means that more (or more effective) cache will hide the seek penalties incurred in reading the disk - this translates into the drive appearing to be better at seek tasks than it actually is.

What do SCSI drives have over IDE? One advantage is that they are faster -> lower seek times.

Can it make a difference? <gratuitous pimpage>

http://users.bigpond.net.au/liamc/q22001/zdrevisitedp2.htm

</gratuitous pimpage>

have a good hard look at the WD Protégé and the Seagate U5 scores. Both have the same platter density, ~access times and interface. The Protégé has a 2MB cache and the U5 has 512KB. Have a look at the performance numbers returned. While 5 or so points in Winstone don't seem like much, that's almost two speed grades of CPU -> going from a
1GHz Athlon to a 1200. And this speed up happened only from the writing to and reading from the disk part of the Winstone tests.

Also have a look at the BB and JB scores on SR. The 8MB models post some impressive numbers.

Will you notice this in the real world? Depends on what you do. I bought 100GB model because it was the fastest IDE disk and my 60GXP was making me nervous Have I noticed a difference? Not really. But then one machine has 512MB and the other has 768MB - so I'm not disk limited.

It boots a little faster, but that could be wishful thinking on my part. Winbench 99 says it's faster. The only time I notice it is when I run IBM's VisualAge COBOL compiler - because it will write the OBJ files for each program back to disk in order, and if you compile projects with many files (like I do) the writing to disk really slows things down. But last year I was bringing a lot of work home - now I'm not. For gaming it shouldn't matter - any game worth its salt runs from memory, so a fast disk is only going to influence load times. For photo/movie editing, I would reckon it would be worth looking at.

Was it worth it? Dunno. Before memory became cheap I would have said yes. But then again, the premium over the ordinary 100GB monster was only about AUS$80 (US$40) so I was happy to pay the extra.

Is it a replacement for fast SCSI? - beats me, I've never been able to afford SCSI!

0.02

 

[CougTek]

http://users.bigpond.net.au/liamc/q22001/zdrevisitedp2.htm

Nice information there, especially the benchmark showing the difference between a single drive setup and one with the swap file ran on a different drive. Interesting.

 

[Tea]

I think I might have done a rather stupid thing tonight. I got my drive today - a 100GB JB, they must not have had stock of the 120GB version - and, of course, I raced home with it and plugged it in right away. Now that I've put precious data on it, it ocurrs to me that it's quite possibly in the worst possible place I could put a drive for testing.

My home machine runs W2K at present, which I am unfamiliar with, and it only does very light duty stuff - mostly just web surfing. So how the #^&$# am I going to even notice how fast the dam thing is? I should have resisted the temptation of the immedate and kept it around the workshop for a while, done some test installs on it, got a feel for it. Or else stuck it in my office server, which does do some moderately hard work from time to time, and runs OS/2, which I have used for long enough to become sensitized to quite small differences.

For what it's worth, it is clearly faster than the Atlas IV I replaced with it - though that's probably more a transfer rate thing than anything else, as the second drive in this machine tends to just store the larger, seldom accessed files. It cooks along very nicely, in fact.

Just as a really rough and ready benchmark, I copied my E: drive (2nd partition on the Cheetah) over to the corresponding position on the second partition of the JB, then formatted the E: drive, and copied everything back again. About 4.2GB of assorted stuff, 51,000 files or so. That left me with two 100% defragged fresh drives. To flush the caches, I switched off, powered up again, and then used the W2K "search" function to search for "*". Took 21 seconds on the JB, 18 seconds on the X15. That was impressive for an IDE drive! (And damn it, I wish I'd done it on the Quantum now, and I wish I had, say, a WD400BB and a 5400 RPM Spinpoint here to try it with those as well. (NTFS, 512k clusters, W2K's index feature turned off.) (I changed the cluster size later.)

Then, I did one of my favourite hard drive torture tests. Start a massive XCOPY going (or drag and drop in this case) and try to do something else that is disc intensive. I used Z Tree to log various of the other partitions (i.e., the ones that were not involved in the XCOPY - thus making the seek mechanism work really hard doing near-full-track seeks constantly). It's hard to tell with Win2K because it doesn't multi-task very well anyway, but once again it was the most impressive performance I have ever seen from an IDE drive, and it wasn't completely shamed by the X15. Slower, yes, but in the ball park.

The JB doesn't sound as sweet as the X15 of course. But that is a matter of conditioning. Anything that costs as much as an X15 sounds good by definition!

I need to use the JB as a boot drive, ideally in my office OS/2 box. Does this mean I need to buy another one? Maybe I'll take that 40GB 4500 RPM Samsung home after all, and take this baby back to the office. It's wasted here.

 

[Mercutio]

What I'd rather see is the results of Tannin's search for '*' on a 'JB and on one of the new 10k Atlases... 'cause if it takes twice the spindle speed on an x15 to make it ~ 20% faster, and nothing else compares (as SR's benchmarks suggest), well, that kind of changes my opinion about SCSI workstations...

 

[timwhit]

Tannin, Did you get a chance to check out the WD drive as a boot drive? What did you think?

 

[Tea]

No Tim. Don't think I will, actually. Not unless I pull it out and replace it with a smaller, lower-performance drive, which I may well do, as I need neither speed nor space for this task. The JB is wasted in this machine. Then I could play with it in some other machine. The best would be to slip it into my other XP 1700, the one at the office. Finding the time to do that might be hard though.

 

[Adcadet]

Tea -
I realize that you aren't claiming that your tests are perfect, but I feel the need to point this out.

Somehow I think your xcopy tests don't really do much in the way of a benchmark since they mainly test STR, for which the larger cache won't do a thing for. I suspect the same test with a 1000BB would have given the same results.

 

[Tea]

Actually, Adcadet, I'm not so sure about that. If I were to simply time a big XCOPY, yes. But that's not what I do. I'm doing the big XCOPY to move my existing data over to the new drive, and then rearrange thing till I have my system more or less organised again, and then attempting to keep on working normally while it happens. I don't particularly care how long the XCOPY takes, as it's a task that takes too long to sit and wait for, and one that I'd have to do more frequently or time with a stop watch to get a proper feel for anyway. What I'm interested in is the ability I have to go on working normally while the XCOPY takes place.

That, to me, is the real test of a good system: put it under heavy load and still be able to use it just the same as normal. Or as close as possible to that, anyway.

All this is to say that it's not a test of transfer rate nearly so much as it's a test of access time and drive logic. Expecting a drive to do heavy read and/or write work in one partition and still be able to slip over to my current working partition and service my read requests (as I surf or play back an MPEG or whatever) is asking a lot. But the really good systems make quite a reasonable stab at it. In general, SCSI drives beat IDE ones for this, fast-seeking drives beat slower ones, and OS/2 thrashes Windows - though the NT family products are at least somewhere on the same street, Windows 9.X was hopeless.

Would a 1000BB produce the same not-bad but not-fantastic result as the JB? I am not sure. I'd have to try it. And I am not going to buy a second unnecessary 100GB drive just to satisfy my curiosity. And in any case, I'd have to run much more careful benchmarks than this seat-of-the-pants stuff, and I don't have the time right now.

But speaking of the seat-of-the-pants, quite a lot of the original gloss has worn off my JB now. It's quick, but it is becoming more and more obvious as I get settled in with it that it doesn't have the same snap as the X15. It's fine once it gets started, but I'm noticing quite a few tasks that begin with a momentary stutter: opening a big folder full of JPEGs, for example. There is a pause of perhaps a half-second or a second which is distinctly noticable on the JB, but not present with the Cheetah. And come to think of it, I didn't used to get that stuff with the Atlas IV either. But perhaps that was because at 9.1GB the Atlas didn't have all that much stuff on it anyway.

In short, the way I'm starting to feel about it, the JB is not fast enough for a $500 drive, and it's starting to bug me just a little. (This is to ignore the benefit of the massive storage space it offers - but to me, space doesn't really matter - I only needed 20GB.) I think I'll do one of two things: sell it and put in a Spinpoint 5400 - which will be quite a bit slower, of course, but won't bug me because I won't expect it to go at anything remotely like the speed of the X15 and it will have cost me just spare change, or else sell the JB and buy another SCSI drive next time I see an offer that's too hard to refuse.

X15-36LP?
A remaindered X15 Mark 1 at a bargain price?
Wait for the X15 Mark III?
Maybe an Atlas 10K III?

Does anyone have both a 10K III and an X15? If so, how do you feel about them?

Guess I'll take a Spinpoint home and see how slow it feels, then decide. Or maybe I'll just keep the JB. It's a darn good drive, after all.

 

[CougTek]

I have both an Atlas 10K III and a Cheetah X15-36LP, but they are now both to rest because the motherboard I used them on is R.I.P. and past its warranty. I didn't used them in very intensive tasks so its hard to tell if the Cheetah was all that much quicker than the Atlas.

I had an odd problem with ATTO on these drives. My small files tranfer rates were poor on the X15-36LP (poor on all three partitions with two different cluster sizes) but reached 40-55MBps (W/R) in the lower half part of the test (the bigger files). The Atlas 10K III however, had the opposite results. It reached 55MBps at halfway into the test and busted 85MBps (in the reads) at the end. The write transfers also were too good to be true with a maximum of ~50MBps. I got the transfer speed of RAID 0 using a single drive. Something was very wrong, but I didn't had the time to figure what it was before the mobo died.

 

[Tea]

I know what you mean, Coug. Many are the small problems and the big mysteries that I just leave because I don't have the time - or, to be more to the point, don't feel motivated to make the time.

Such as proper benchmarks on the JB I used to have.

Yes, used to have. I had thought to take a 40GB Spinpoint 5400 home tonight to replace the JB with. But Westan were out of stock of Spinpoints and sent me a substitute - bloody Proteges. Sluggish things, and dearer than the Samsungs too. Nothing actually wrong with them, just that I have not the slightest intention of putting up with a 12.5ms seek time in my own machine - not even for a rarely-used data drive.

So, seeing as I had a 60GB Spinpoint on the shelf, I grabbed that instead. Just finished copying the data over, gave it a test run plugged in side-by side with the JB. Slower, but you know there ain't that much in it. There is a different quality to the way it goes about its work also. Where the JB hesitates more obviously and then delivers all in a rush, the Spinpoint feels mor "rounded" if you like to put it that way. The JB was all go or all stop, the Samsung seems to glide from the one state to the other less perceptably. Perhaps that is the combined effect of (a) the bigger cache and (b) the Spinpoint only being on an ATA-33 cable (seeing as I had borrowed the secondary controller for the data transfer from my CD burner. (Now that I've pulled the JB out, of course, it's on the ATA-100 as primary master.)

I'll go a day or two first but at this stage I think I'll keep the Spinpoint. Now that I have the chance to move a whole lot of junk off the X15, I can look forward to overall improved performance.

One last thought while I'm on the subject. Seems to me that, at least in the case of a data drive which holds mostly smaller files, the key performance number remains access time. This is why the Spinpoint doesn't feel all that much slower than the JB - it's only 1ms slower, so far as access time goes. By rights, the Cheetan ought to feel much faster than either, and it does, though not as much as I'd have expected - but then I'm booting off it, so I guess comparing a different partition of my boot drive to a single in-use partition of my data drive is not a fair test.

If I had to quantify the differences, I guess I'd score the drives like this - giving the X15 an arbitary 10 out of 10 -

X15 - 10
JB - 7
Spinpoint - 6

The difference between the Spinpoint and the JB is, in this application, maybe half as noticable as the difference between the JB and the X15.

 

[Santilli]

and, where in all this did you put the pagefile?

I've used a faster ide drive for a storage pagefile, on some machines, and been happy with it. Don't you think this is a better idea, even if the startup drive is an X 15?

gs

 

[Tannin]

I'm not sure that the pagefile has any effect on this machine, Greg. It's so ridiculously over-powered for the work it does already that fine tuning is just silly. So, for that matter, is spending all that money on the CPU and hard drives for it! But for what it's worth, I left the page file on the X15 until just after I had put the JB in. At that point I realised that I had heaps of space everywhere except on my C: partition, which at 2.5GB total had only 200MB or so free. I never quite trust Partition Magic, and I don't know how to clone a W2K install by hand (though I could do Win95 or DOS or OS/2 in my sleep) so I was reluctant to make it bigger, and too lazy to reinstall.

But I fiddled around for 10 minutes wondering how to move the swapfile, finally discovered that it was in the only place it never occured to me to look - i.e., the same place it is in Win95/98! Put it on the JB, now it's on the Samsung 5400. But I have not noticed any performance difference. The stuff this machine does, it probably doesn't get that many hits anyway.

As a rule of thumb, I'm in favour of putting the pagefile on a seperate drive, even a moderately low-performance one, but it seems to matter less and less as we fit more and more RAM. Back in 16MB days, it was vital!

 

[Santilli]

Tannin: I agree

"As a rule of thumb, I'm in favour of putting the pagefile on a seperate drive, even a moderately low-performance one, but it seems to matter less and less as we fit more and more RAM. Back in 16MB days, it was vital!"

One of the machines around here is a P2 400 mhz, Intel mobo.
With 384 mb ram, 2000 os, it really doesn't hit the page file much, which is good, because with that mobo, and the perifs, I can't put another drive on any channels, and, I'm out of PCI slots.

The faster the hard drive in it, the less time it hits the drive, period, and, this makes a big difference.

But, in the other machine, 1.4 gig Athlon, the hard drives are hit so rarely, with 512 mb of ram, and 105 mb/sec raid, that putting the pagefile on the raid works just fine.

gs

 

[Bozo]

My system has 640meg/RAM. I'm running XP Pro and in task manager it shows how much PageFile is in use. Just starting the computer and opening IE, it uses 144meg of paging. Seems like a lot, but then XP has a lot of bloat.

The operating system is on a WD200BB hard drive. The PageFile is on a 3Ware RAID 0 setup. I've had the PageFile on the boot drive and didn't see much difference. But, it's only a PIII-800.

Bozo