Harddrives - 2mb caches verses 8mb caches

  • Thread starter Thread starter Kb
  • Start date Start date
K

Kb

Hi,

I was about to buy a new harddrive, and was wondering if the difference in 2mb and
8mb cache drives was worth the extra. Anyone know what approximate
% speed difference there is (I was looking at Maxtor, Seagate and WD ones)?
And is it it noticable in practice? I was wanting it for my desktop machine,
using mainly desktop apps, programming and the odd game. And is there anything
you can do with a 8mb drive you cant with 2mb drive?

Thanks in advance
 
Check storagereview.com for drive reviews. The price differences
have gotten small, so I would choose one with an 8 meg buffer.
Some 120 gig drives with an 8 meg buffer are priced under $120.
(under $1 a gig!). In addition to buffer size and 7200 rpm rotational
speed,the newer drives have more gigs/platter, which also makes
them faster than older drives.
 
Kb said:
Hi,

I was about to buy a new harddrive, and was wondering if the difference in
2mb and 8mb cache drives was worth the extra. Anyone know what approximate
% speed difference there is (I was looking at Maxtor, Seagate and WD
ones)? And is it it noticable in practice? I was wanting it for my desktop
machine, using mainly desktop apps, programming and the odd game. And is
there anything you can do with a 8mb drive you cant with 2mb drive?

For video captue they help a bunch. For normal uses you'd probably never
notice the small difference in speed. 7200RPM is a good thing to have. The
other point in favor of the 8meg drives is at least with WD you get a 3
instead of a 1 year warranty,
 
Stacey said:
For video captue they help a bunch.

Have you got a link to confirm that? For video capture it shouldn't make any
difference at all. When capturing the video the buffer will fill up
extremely quickly, so you will very soon be constrained to the speed of the
hard drive (as opposed to the interface).

Gareth
 
Gareth said:
Have you got a link to confirm that?

Personal experience. I was dropping frames on a system I built for a guy.
Before I went to a 2 drive setup, I swapped an 8Mb drive in and the problem
was gone. Also on WD's site they recomend the 8Mb drives for DV/file server
applications? Since they do transfer data faster it's got to help!
For video capture it shouldn't make
any difference at all. When capturing the video the buffer will fill up
extremely quickly,

Unless I am totally confused here, it's a =buffer= not storage. Doesn't it
refresh what's in the bufer after that info has been transfered to the
drive? DV needs 3.7Mbs tranfer rate. With a 2Mb drive that's less than 1/2
second of buffering, with the 8Mb drive it's over 2 seconds worth. That's
more than enough to deal with any sort of OS call ect that might occur
during capture. If someone is using 2 drives (one storage, one system) this
wouldn't matter. Also if someone is doing some sort of heavy bandwidth
analog capture, the extra buffer probably isn't enough to matter but most
people doing video computer work are using DV nowadays.
 
Gareth Church wrote:

Yep, that's exaclty right. That's why it makes no difference (assumming
you are capturing >2 seconds worth of video). After a few seconds the
buffer will fill up. But data will still be getting sent to the drive from
the video capturing process. No more of this data can be saved in the
buffer, so in order to keep on accepting new data the stuff in the buffer
must be written to the drive. This means that very quickly the effect of
the buffer will be negated; in order to write a large DV stream you simply
must have a drive that can write (that is physically, to the platters).

OK I realize the drive must be able to write to the disk as fast as the
capture, what I was getting at was if there is an interruption in the
physical write process (that's what ussually causes intermitant frames to
be dropped, a disk call being made while capturing) the cache can handle
the stream until the disk can return to physically capturing the data. The
system can/should recover and refill the cache very fast compared to the
act of physically writting to the disk.

The drive that was having the frame drop problem was the same kind of 7200
80GB WD drive that was tested at over 40Mbs yet was dropping a few frames
here and there even though it was just a 3.7Mbs data stream. Going to an
8meg drive stopped the problem so something was going on? Thanx for the
info as I am curious as to why.
 
Hi,

I was about to buy a new harddrive, and was wondering if the difference in 2mb and
8mb cache drives was worth the extra. Anyone know what approximate
% speed difference there is (I was looking at Maxtor, Seagate and WD ones)?
And is it it noticable in practice? I was wanting it for my desktop machine,
using mainly desktop apps, programming and the odd game. And is there anything
you can do with a 8mb drive you cant with 2mb drive?

Thanks in advance

Around where I live they pretty much only sell 8 meg cache drives in
the larger sizes and none on the really small sizes so its kind of a
moot point . I dont see much of an overlap in my area.

I just bought two of them a 120 meg maxtor and a good deal on another
Maxtor at COMPUSA for $129 200 gig with Promise card I think with 8
meg cache. I still have one of the 60 gigs I bought a few months ago
with the smaller cache which I might build another system with.

Subjectively it does feel faster sometimes but as Ive posted before I
havent really tested it.
 
Stacey said:
Here's what may have been going on?

"The use of cache improves performance of any hard disk, by reducing the
number of physical accesses to the disk on repeated reads and allowing data
to stream from the disk uninterrupted when the bus is busy."

So if the bus is busy (capture or fire wire card) the cache can keep feeding
info to the drive. I do know that this "bus busy" issue is why via chipset
boards have problems with video capture.

The explanation I gave was referring to write-caching. The above description
is for read-caching (where the hard drive says, "ah, he's just read this
sector, there's a fair chance he'll want to read the next sector, so I'll
copy that into the buffer while I'm here"). I suppose there could be
circumstances where this may help with capturing, but I'm having a hard time
thinking up any. In any case, you would be much better off making sure no
process is going to interrupt the capturing process (especially if that
process wants to use the hard disk).
As far as turning off everything, there are some things you can't turn off
that may try to access the drive (even just reading from it) without you
knowing it. Having a larger cache helps keep systems that aren't configured
perfectly from having problems.

I certainly hope not. Sure, there are lots of Windows processes that can't
be terminated. Most of these run in the background, and either a) only do
work when the CPU is idling or b) do work when you actually request it.
Also (I'm trying to get this straight) aren't cached writes "fifo" (first in
first out)

Most explanations of this issue tend to talk about drives using the FIFO
strategy, because it is easiest to explain/understand. Early drives used
FIFO, but I don't believe any manufacturer does any more. Much more advanced
algorithms have been developed now, which seem to offer a better hit-ratio.
Three more caching strategies are listed here:
http://www.pcguide.com/ref/hdd/op/cacheCircuitry-c.html. That site is old
now (by at least a few years), so I wouldn't be surprised if there are new
strategies, or at least more revised versions of these.

It's an interesting point though. Lots of people look at 8-meg drive caches
and think the drive must have good performance, but a drive using a dumb
caching algorithm won't perform as well as a drive using a smarter one.
even when streaming to the drive and not just loaded when first
writing and once they fill up just stop doing anything? I wonder if the
logic on some drives do act that way while others stream the data through
the cache when writing? I can see why they wouldn't bother on reads and
maybe I'm confusing what was going on with that system that was fixed going
to a larger cache drive.

You lost me there! If you are suggesting that the cache is used for both
reads and writes, then you are correct. In fact, they actually leave writes
cached, in case they are needed again soon.

I think you are also asking about what drives do when they fill up. The way
I talked (wrote) about this was probably misleading. As far as I can gather,
the drive doesn't stop using the buffer when it fills up. It will still
place new writes in the buffer, but before it can do that there has to be
room in it. If the buffer is full of data that is waiting to be written to
the hard drive then there is nothing in the buffer that can be removed, so
the new data to be written will have to wait until some of the data in the
buffer has been written in the platters, so that there is room in the buffer
for the new write. Wow, that was a long sentence. Anyway, if you followed
that you will see that the buffer is still used, but in order to use it we
must write something to the platters, removing any advantages the buffer
offered.
I still think it's a good idea given the small difference in price.

As do I. I hope I haven't come across as being too negative. I believe
people place far too much emphasis on this figure, but certainly don't think
it's a bad thing. A drive with an 8-meg buffer can only perform better than
one with a 2-meg buffer (all else being equal).

Gareth
 
Stacey said:
OK I follow that but since the drive can write 30+Mbs and DV only needs
3.7Mbs wouldn't the drive easily be able to deal with keeping the buffer
clean, unless something interrupted it then the buffer/cache could save it
until the drive could catch back up? I'm wondering if when -cache writting-
some drive's logic could do this?

I suppose in this case the size of the drives buffer could make a
difference. A DV stream is atypical in that it is a constant stream of data
that is below the speed that the drive can write to the platters. The
problem comes when another process wants to write to the drive (assumably at
full speed). If the drive has no buffer at all then (the other process)
writing for 1/30th of a second would mean you have missed a frame.

I would still maintain you are better off stopping all the non-essential
processes before capturing (and also give the capturing process a higher
priority), not only because those processes might take up hard disk time but
also CPU time, which could also cause dropped frames. But yes, in the above
circumstance I suppose the size of the buffer could make a difference.

As for the the drive caching algorithm being able to distinguish between
data, and store some data until the drive is free (for instance, the
capturing process is stopped), I don't think that would be possible. I can't
see any way that the hard drive can distinguish between data. It works at a
fairly low level - it gets some data and writes it to the platters. This
sort of thing is typically done at the OS level by assigning priorities to
processes. That way the OS, based on the priorities, can decide how many
resources to give each process. So it could decide which process gets access
to the hard drive and when. Once the data has been sent to the drive,
though, (and is in the drive's buffer) I can't see how the drive can make
any decisions about it. I really don't know for sure, though.
Until the latest drives came out serious
DV uses have always used a system/storage dual drive setup as even with the
OS tweaked and drives that benchmark faster than DV transfer rates, they
would still drop frames intermitantly so something weird is going on.

As you say, DVs data rate is quite low, and all hard drives are able to
write to the platters faster than it, so the 2 drive setup isn't generally
required anymore. As you say dropped frames still happen, but this isn't
necessarily due to the hard disk sub-system. I first started capturing DV on
a PIII 450 which had it's original hard drive (what's that, circa '98/'99?)
and over an hour or so capturing I would typically drop maybe 3 or 4 frames.
Of course, I took time to make sure no extra programs were running in the
background, and disabled the screen saver etc. I believe most problems with
dropped frames are likely to be due to another processes getting in the way,
not because of problems with the disk drive sub-system.
Not at all, I'm just trying to learn something here

Me too. I love discussions like this where you have the motivation to look
things up (to double-check your facts before you write, and also to find new
info) and hear someone else's point of view. Usually people get sick of me
pretty quick though!
as I assumed the cache
is what fixed those frames being dropped and never stopped to wonder why.
At least I thought I understood why before you posted this!

As above, I guess it is possible the larger buffer helped. I'm not convinced
though.But I haven't got a better theory (unless the old drive was using a
low ATA level or something).

Gareth
 
Gareth said:
As you say, DVs data rate is quite low, and all hard drives are able to
write to the platters faster than it, so the 2 drive setup isn't generally
required anymore. As you say dropped frames still happen, but this isn't
necessarily due to the hard disk sub-system. I first started capturing DV
on a PIII 450 which had it's original hard drive (what's that, circa
'98/'99?) and over an hour or so capturing I would typically drop maybe 3
or 4 frames. Of course, I took time to make sure no extra programs were
running in the background, and disabled the screen saver etc. I believe
most problems with dropped frames are likely to be due to another
processes getting in the way, not because of problems with the disk drive
sub-system.

What I think is going on is no matter how much stuff you turn off, there are
things in windows that make short HD calls that you can't stop. 3-4 frames
isn't bad but I like to see zero ;-) I think with these larger cache
drives a "clean" tweaked system isn't as important and is also why people
use a system/capture drive, to avoid the same drive being called when
capture is going on as the hard drive is the slowest part of a system.
As above, I guess it is possible the larger buffer helped. I'm not
convinced though.But I haven't got a better theory (unless the old drive
was using a low ATA level or something).

They were the same WD drives (AFAIK) 7200RPM 80gig ata100, just one was a 2
meg and the other an 8 meg. Thanx for the links and for making me think
about this! :-)
 
Back
Top