On Thu, Apr 24, 2008 at 12:02 AM, Sudev Barar <[EMAIL PROTECTED]> wrote:
> On 23/04/2008, David Burgess <[EMAIL PROTECTED]> wrote:
> > I have 4 thin clients booting pxe on realtek 10/100 onboards through an
> > unmanaged netgear prosafe gigabit switch. If I run a youtube video on
> one
> > client, video and audio are reasonably smooth. If I play the same
> youtube
> > video on all 4 clients simultaneously the video gets much slower. If I
> > full-screen a couple of those then the audio starts to break up. For
> this
> > reason I'm fairly confident that this rules out the thin client
> hardware; it
> > must be a bottleneck in the server or network, right?
>
> If you are running these four youtube sessions off the internet then
> your upside connection would be a bottleneck. Each session creates a
> individual pipe.
>
> For testing you can do a local broadcast, say using vlc, and check.
Yes, I should have explained that the youtube video is fully downloaded on
all the clients. Once it has played through I can hit the play again button
on the clients and repeat the process.
Also, when all four clients are playing the video fullscreen with broken
sound, the video is not only refreshing slowly (about 1 fps), but it's also
playing in slow motion: rather than dropping frames the video is lagging
further and further behind the audio. Then when I switch some or all of the
clients back to the smaller embedded video size, the video plays fast
forward until it catches up to the audio, which is no longer broken.
I'm pretty well convinced the bottleneck is client-side, I just can't figure
out why, unless gnome system monitor is reporting wildly incorrect
throughput. As I said, it's showing a sine wave peaking about 28 MB/s (~228
mbps). Even if I'm pushing a steady 230 mbps, I should theoretically be
capable of more. Anyway, that sine wave peaks well over 3 mbps when the only
traffic on the machine is an internet download and my router confirms that
my internet connection peaks at 3 mbps. This would suggest that the area
under the sine wave is more representative of my actual sustained
throughput, rather than the peak, and my 4 thin clients are probably pulling
an aggregate of something well less than 228 mbps.
I'm wondering if that's all the cpu can handle. Realtek cards are known for
their lack of cpu offloading, so perhaps my celeron 2.5 GHz is pinned trying
to play flash while the realtek is pulling a solid 30 mbps. I know somebody
in the m0n0wall forums reported his router maxing out at 12 mbps throughput
on realtek cards and achieving numbers several times higher than that (maybe
even wire speed? I'll have to find that post) after switching to intel
cards.
I'm going to try sticking an intel GBE into one of the clients and see what
difference.
Thanks for your help.
db
-------------------------------------------------------------------------
This SF.net email is sponsored by the 2008 JavaOne(SM) Conference
Don't miss this year's exciting event. There's still time to save $100.
Use priority code J8TL2D2.
http://ad.doubleclick.net/clk;198757673;13503038;p?http://java.sun.com/javaone
_____________________________________________________________________
Ltsp-discuss mailing list. To un-subscribe, or change prefs, goto:
https://lists.sourceforge.net/lists/listinfo/ltsp-discuss
For additional LTSP help, try #ltsp channel on irc.freenode.net
