Re: 2.1.40 nfs serving is foobar.
Keith Rohrer (kwrohrer@uiuc.edu)
Tue, 27 May 1997 10:37:26 -0500 (CDT)
> Running 2.1.40, backing up to a syquest disk over NFS I got:
> IP: queue_glue: no memory for gluing queue c0662de0
> IP: queue_glue: no memory for gluing queue c0662de0
> IP: queue_glue: no memory for gluing queue c0662de0
> IP: queue_glue: no memory for gluing queue c0662de0
> IP: queue_glue: no memory for gluing queue c0662de0
>
> (the source machine is running 2.0.30, the syquest machine is running
> 2.1.40)
This isn't NFS-related. I got some of this with the video-on-demand
server I'm working on; it happens sometimes when talking across the
NE2000, but never when talking across the loopback. I'm certain I got
this behavior using TCP (and maybe with UDP too, like the NFS presumably
did); my write() calls were generally about 5k in size, which exceeds
lo's MTU. The 2.1.40 machine was sending the packets to a 2.1.29
(not 39!) machine or to itself.
> This also happened (seemingly repeatedly after a while) on 2.1.39.
I can't confirm or deny this.
> I don't do any large nfs writes to the 2.1.40 machine (I suppose I
> could see if it fails on iozone).
I'm only pumping data at 150k/sec, and the target machine is the only
other box on the wire, so it doesn't take a huge requested DTR...though
I had the server set to send the whole 150k every second...
> Its not happening on 2.0.30.
It's not happening on 2.1.29 either.
> Looking at the code, it has to be a bad occurence (I have
> 24 Mbyte on the machine, and its not doing much other work).
I've got 32 megs.
Keith