[Udpcast] Memory issues creating a PXE boot disk

Andrew Reusch areusch at gmail.com
Fri May 26 03:56:18 CEST 2006


Yes, I certainly can. However, this will take some time. The sender is
running on a big-endian ARM process (the linksys NSLU2) and so I will have
to compile this. Now the receivers are still running on normal x86
computers. In the mean time, I am going to get another computer (x86)
running some version of linux, and then install the latest version of
udpcast on the computer and help with that.

Regardless of that, here are my thoughts. I'm sorry, I realize I wasn't
terribly specific in my last message.

The bug you're describing doesn't sound like the one I'm experiencing at
all...the problem appears on the clients running udp-receiver. I'm not even
sure if it has much to do with udpcast, rather with the way I'm building my
image. Basically, the udp-receiver processes take around 6500 K of memory if
my memory serves (haha, okay never mind). And this stays constant. What does
change is listed in /proc/meminfo. It's the Inactive line, which is what is
steadily increasing and then being flushed. Now if I continually run a sync
every second, it seems to help some. But all that command looks like it's
doing is flushing a bit more memory down to wherever it's supposed to be.
Possibly this has to do with some kind of delayed write in linux? The amount
of inactive memory returns to normal levels after the bash process which
runs udp-receiver terminates.

I'll let you know how udp-sender runs on the x86 machine...any thoughts on
this receiving issue?

Thanks very much for your help,

Andrew

On 5/25/06, Alain Knaff <alain at knaff.lu> wrote:
>
> Andrew Reusch wrote:
> > Hi all,
> >
> > I'm creating my own boot disk to do a re-image from udpcast. I'm a
> little
> > new at this but have been using linux for about a year and a half now as
> my
> > desktop machine. I've had a decent amount of experience coding so I'm no
> > stranger to that either.
> >
> > I've got an boot disk based off of slackware 10.2. The boot disk loads
> up a
> > ramdisk and then mounts the target hard disk on /mnt/target and then
> uses
> > udp-receiver to receive a tarball into that directory. So theoretically
> all
> > data should go to the hard disk. Everything is fine until udp-receiver
> is
> > started. Once it starts, the Inactive memory usage begins steadily
> rising
> > until it has reached a value about 2 mb below the memory limit, where
> the
> > amount of free memory begins oscillating. This amount of inactive memory
> > remains until the bash script that started udpcast exits.
> >
> > The command line used to start udp-receiver is
> >
> > udp-receiver --nokbd 2>/udpcast-log | pv -pets --file size--
> >> /mnt/target/${pkgfn}
> >
> > $pkgfn is the name of the tarball.
> >
> > Normally this is not an issue. But sometimes, the memory peaks out, and
> the
> > kernel kills either udp-receiver or another crucial process. Whoops. The
> > boot disk works fine for 128 MB memory computers, but not on 64 MB
> memory
> > ones.
> >
> > Anybody ever have this happen before? Any ideas??
> >
> > Thanks very much for any help/guidance,
> > Andrew
>
> There was indeed a memory issue in udpcast, which is now fixed in
> version 20060525. However, oddly enough, it was in the sender, not the
> receiver, so I'm not quite sure whether it is indeed the same problem.
>
> During my tests, memory usage stays constant (around 15 megs) even
> during long transfers.  And even with the bug, the excessive amount of
> memory was allocated from the start, rather than accumulating.
>
> Could you give it a try anyways, just in case?
>
> Regards,
>
> Alain
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.udpcast.linux.lu/pipermail/udpcast/attachments/20060525/7c2ee363/attachment.html>


More information about the Udpcast mailing list