Re: xfs and drive speeds

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]



Jodie Reynolds wrote:

>>Actually, I did mis-state one tiny thing - My sysadmin had upgraded it to
SCSI-360 when I wasn't looking, darn him. I discovered that >>when I pulled
the case off to see if there was cache on the controller. It's Atlas-III/10K
and there's 128Meg on the controller.  Internal >>cache-to-drive-cache speed
according to maxtor is 490MByte/sec.

> OK, now you're just showing off :-)

Naw, actually, it's still toooo slow...

>>I  really have no idea how well implemented their caching system is.  But
the  results to me suggest that their cache performs better than >>system
cache and  therefore is slowed by allowing caching elsewhere.

>Could well be. I sanity checked your figures with benchmarks available on
the WEB, and the bigger the filesystem block size (up to 32K), >the faster
the transfer speed. Your test figures do appear to be correct. That Quantum
drive is quite is the business!

They sold it out to Maxtor, btw.  Quantum is only doing tape solutions now,
as far as I know.  What a shame.. and another one bites (bytes) the dust...

>>The winnov board uses a proprietary hardware compression algorithm that is
configurable.  I use a 2:1 compression so that I can get the >>data onto the
drive reliably.  At 2:1 there's basically zero visable loss.

>Do you not convert to mpeg2 or other formats? What is you application -pro
editing?

Actually, we have a proprietary (linux-driven) scheme that allows us to
stream different formats to different people at different bit-rates using a
methodology that I'd have to kill you if I told you about. (Atleast until
the last round of the Provisional Patents are granted...)  We store the raw
video at as high a quality as we can.  From there, if you connect with our
proprietary player, you get our optimized format, from any device, at any
bitrate from 2.6Kbit/sec on up.  If you connect with Windows Media Player,
you get a windows stream.  If you connect with Real, you get a real stream.
If you connect with QuickTime, you get a QuickTime stream, and so on.  All
from the same original video, all optimized to the individual stream, in
real-time, up to 15,000 streams from a single Linux box.  The ideal
situation for us is to get the video onto the machine at the highest
possible quality.  In real-life we use big multi-drive raids that are fast
enough to take the video 1:1 in real time.  On the development machines, we
don't hemmorage that kind of money.   We're not a dot-com company.  [grin]
The Atlas or Cheetah drives are a nice price-point compromise.

>>I have a DC30Pro in my Windows machine.  Not at all impressed.

>The DC30 is a bit of a pain as far as I am concerned. It has poor drivers,
uses a proporietary compression system and you need the same >card to do
playback. Having said that, the problem might just be Windows. I can't say I
get much visible loss out of my DC10+ though. >But then I'm not looking at
it with Pro equipment. It's for home use only.

And it can't be used as a capture board for Windows Media or etc, as far as
I know, it has no VfW drivers.  On the pro-side of the house, I do have a
1080i HDTV and a JVC Reference Monitor, so it is pretty noticable when the
video doesn't capture well.  Especially when using Component Video output to
the Reference Monitor...

>>I've ordered a couple of Cheetah 15,000 RPM SCSI-360 drives..  If you
like, I'll report back their results on the same test.

>That'll be interesting.

I sure hope so!

--- Jodie Reynolds
    Chief Technology Officer
    Interact Devices, Inc.

>Ed-T.


--- Jodie
----- Original Message -----
From:Edward Tandi
To:video4linux-list@xxxxxxxxxx
Sent: Sunday, July 15, 2001 3:39 PM
Subject: Re:  xfs and drive speeds


Jodie,

OK, you are using high-end(-ish) drives. I find the un-cached result
surprising as it appears to be higher than the result for the cached test.
What are you using to do the test?

Also the very high cached read result (looks too high to me) would imply
that the 512MB of RAM you have in the machine is having a large effect on
the results. Could this be affecting the write tests also? Does the Adaptec
card have a cache?

As for what you need for capture; I use a DC10+ that does mjpeg compression.
At full resolution it uses 2GB every 5 minutes. This rounds up to 7MB/s. I
don't know about your winnov card, but my DC10+ doesn't do that well on
compression considering what mpeg1 and mpeg2 can do to it.

Anyway, so why such an interest in this xfs fs format?

Ed-T.

BTW, a single 5K RPM IDE just isn't good enough, and 7.5K RPMs aren't that
much more expensive.


Jodie Reynolds wrote:

I think Eric has a point. Even with you calculations below, I think
thebottleneck is you hard drive. Write rate is slower than read rate.13MB/s
write rate for a single disk is doing very well.Ed.-T.
I agree Write is slower than Read, but his numbers don't seem all that outof
the ball-park, depending on his drive.testing my video capture box (Dual
PIII-1Gig, 512Mb PC133 RAM, dual QuantumAtlas 10,000 RPM drives (one
disabled for test) w/29160 Adaptec SCSI-160controller) I post numbers
consistantly higher than 13MB/sec write, actuallyrather substantially
higher, more than double on a long sustained write(like we'd see for
video) -Test File: "/$$test$$.tst"Test File Size: 6400 MBTesting Uncached
Write Speed....!
Data Transfer: 27.19 MB/s, CPU Load:  0.8%Testing Uncached Read
Speed....Data Transfer: 21.89 MB/s, CPU Load:  0.4%Testing Cached Write
Speed....Data Transfer: 17.89 MB/s, CPU Load:  2.0%Testing Cached Read
Speed....Data Transfer:179.67 MB/s, CPU Load:  0.4%  (PHEW!)I'll also admit
that this is skewed some-what out of the park of commonconsumer machines...
!If we're talking about a 5400 RPM IDE-33 drive, allbets are off. [grin]But
now on to the real question here for a moment:  NTSC resolution is
what,480x704 x24bits x30fps (merged-frame/deinterlaced), right?My numbers
for full-screen, merged-frame/uninterlaced, non-hardware assistedcapture
would then be 480 lines x704 pixels x24 bpp x30 fps or around238Mbit/sec,
which is about 30MBytes/sec, worst-case, right?  Which wouldexplain why I
like my winnov card doing 2:1 hardware compression taking medown to about
15MByte/sec, or enough to reliably ge!
t the data onto thedrive.Did I make some huge mistake here?--- Jodie
Reynolds, CTO     Interact Devices,
Inc._______________________________________________Video4linux-list mailing
listVideo4linux-list@xxxxxxxxxxxxxxx://listman.redhat.com/mailman/listinfo/v
ideo4linux-list





[Index of Archives]     [Linux DVB]     [Video Disk Recorder]     [Asterisk]     [Photo]     [DCCP]     [Netdev]     [Xorg]     [Util Linux NG]     [Xfree86]     [Free Photo Albums]     [Fedora Users]     [Fedora Women]     [ALSA Users]     [ALSA Devel]     [Linux USB]

Powered by Linux