On Sun, 11 Feb 2001 10:49:33 +0100, Bernie <[EMAIL PROTECTED]> wrote:
> Clarence wrote:
>> Well SURE. To me, you are saying just what *I* said. That was a LARGE
>> file, you were reading it for the first time essentially, because you
>> were reading only a small amount (your readahead size) at a time, and
>> of course I don't know the extent of your fragmentation, but you must
>> agree that _if_ your disk was badly fragmented most of the readahead
>> data is useless ?
> Since I had created the file before using it (with a restart in betwen of
> course) I had *no* fragmentation on the file.
I guess you didn't see the word "if" above.
> And it's my oppinion that even if it was fragmented I would still gain very
> much from read-ahead (IMHO same or even more). The test I did was so absurd
> so we can't count it of course, but it's my belief that the cache will read
> ahead in the file - not on the disk. The cache on the HD (an unknown
> ammount for me) probably work diffrently. Even if you are correct about
> read-ahead you would still gain from it since you can't have the worst case
> scenario (1 allocated unit here, another in the other part of the HD and
> then the third somewhere else and so on).
>> I DO agree that when readahead CAN do any good, it does a LOT of good.
>> I just don't think it does any good MOST of the time, and on most HDs,
>> (fragmented) it does harm. ;-)
> Most HDs (among us anyway) aren't fragmented IMHO. I don't know about you
> but I use defrag/f on my drives on a regular basis (but I set the BATch
> file up so it doesn't use defrag if nothing has changed since the last time
> thereby saving some time).
Defrag by itself can make the drive slower, because you spread empty
sectors thinly thru it. ;-)
Well, I'm taking part in this experiment right now because I have a very
difficult Quantum HD that I'm trying to drag into the 21st century and
it doesn't want to come. :(
What I mean is, I certainly don't know EVERYTHING that's going on in the
HD/cache world, but I *DO* know you can't say anything definite based on
one experiment or even MANY experiments, if they were all done on one HD.
There is a very large amount of work involved to run even very simple
experiments on several systems if you want meaningful/understandable
results. I'm testing 4 very different computer systems, all with the same
test program and the same caching software, and getting silly results.:((
The 44Mb Miniscribe drive on my 20Mhz '386 gives test times which might
be reasonable. Let's ignore that one for a while.
My 170Mb Maxtor (5x32Mb DOS3.3 partitions) on my 33MHz '486DX at my
office is the problem child because it is the standard the rest should
beat. I get Speedchk times about 27 Sec with a ~600kb Cache (PC-Cache)
using the 3 sector mimimum lookahead and this is on a partition that
has not been de-fragged even once in 10 years !!
I run the same test on the WDigital Caviar 300Mb drive in my P90 at home,
using the same software but cache size increased past the optimum (1320kb)
and I can barely get 26 Secs with PC-Cache. This box uses DOS 5.0. The
partition tested was EMPTY !!! The computer is 3x as fast !!!
I have this recalcitrant Quantum 105Mb drive in a 100 Mhz '586 that is
fully defragged and compacted and almost empty except for DOS 5.0.
There is NOTHING I can do to get that thing under 38 Secs !!
I'm sure it should be as good a drive as the Maxtor, and anyway, I've
got the cache up to 5280kb on a box *3* times as fast but still can't
catch it.
I'll figure out what is going on eventually. All I can say for sure now
is if I quadruple the size of the cache on the P90, it runs 20% slower,
if I set the lookahead on the P90 or '586 to max it makes *NO* difference,
and on the P90 it makes no difference if I frag the drive. (I wrote a
.BAT to fill 7Mb of space and distribute empty sectors throughout it.)
On the Quantum, however, 660kb cache gives 39 sec on a clean drive and
1 min 14 sec on the fragged drive. They both have 64k onboard cache.:(
BTW, last night the the P90 gave me 31 sec with dirty cache and tonite
it makes no difference - 26 sec clean or dirty. Arrrrrggggghhhhhhh!!
To top it all off, I've now seen the DOS clock run BACKWARDS on TWO
machines in between test runs. I lost 45 minutes on the P90 last nite
in about 2 hours. And NO, this effect does not seem to be screwing up
my test timings. <g>
- Clarence Verge
- Still using Arachne V1.62 ....
