[Xen-devel] Re: [Xen-users] "Memory squeeze in netback driver"

[Harald Koenig <H.Koenig@xxxxxxxxxxxxxxxxxxxx>]

Hi,

today my XEN server run into the same problem: many kernel messages in dom0

        xen_net: Memory squeeze in netback driver.
        printk: 4 messages suppressed.

and all 13 domU networks are dead.

this is a dual-dual-core-operon (4 CPUs) running SUSE 10.1 in dom0
(xen 3.0.2_09763-0.8 from SUSE) with 8GB ram and Broadcom BCM5704 Gigabit 
Ethernet
on Supermicro H8DA8.

is there any chance to avoid rebooting dom0 ?

is it possible to avoid this problem in the future ?


On Oct 19, Petersson, Mats wrote:

> dom0_mem shouldn't be added to the vmlinuz line, but to the xen.gz line
> - which I think is the reason your dom0 is showing 12GB. Not sure if
> that's got anything to do with the above problem tho'. 

I don't think so because in my setup I'm passing dom0_mem to xen:

    kernel /boot/xen.gz dom0_mem=500M com1=38400,8n1 console=vga,com1 
sync_console
    module /boot/vmlinuz-xen root=/dev/sda1 showopts console=ttyS0,38400 
console=tty0


but checking the time stamps in syslog when those messages started continuesly
(there are some few "Memory squeeze in netback" events from time to time)
it looks like this might got triggered by the creation of a new domU
which caused the dom0 being shrinked from 500 MB to ~290 MB 
(dom0-min-mem is 196).  shortly after that, dom0 was reset to 500 MB.

here is a short histogram on the number of "Memory squeeze in netback driver" 
messages per hour (with those "printk: XXX messages suppressed" added correctly)

          day  HH count
        Nov 20 16 1
        Nov 20 17 1
        Nov 20 18 350
        Nov 20 19 273
        Nov 20 20 277
        Nov 20 21 268
        Nov 20 22 250
        Nov 20 23 275
        Nov 21 00 1195
        Nov 21 01 3600
        Nov 21 02 3600
        Nov 21 03 3600
        Nov 21 04 3595
        Nov 21 05 3600
        ...

around 18 o'clock (6 pm;) the dom0 memory resize happend. looks like this 
was an initial trigger for the problem. 

the external connection got lost around 00:45 am this moring, which matches
the increase of those netback messages from about every 15 secs to every second.


do you need any more information ?


on xen-devel I found only two mails about "Memory squeeze in netback driver"
from Isaku Yamahata <yamahata@xxxxxxxxxxxxx> with a patch which removes
that kernel printk():

        Attachment #2: 9897:5fcebfab4e31_netback_autotranslate_mode.patch

        # HG changeset patch
        # User yamahata@xxxxxxxxxxxxx
        # Node ID 5fcebfab4e317f1db0b0379ac4be99d8cdb9b2b2
        # Parent  a19cc748469e99eae7f289f3beb03206982b9ba2
        change semantics of grant transfers for vp guests so that the
        operation automatically gets you a fresh page at the same
        pseudo-physical address as Keir suggested.
        PATCHNAME: netback_autotranslate_mode



any ideas or comments ?

thanks,

Harald Koenig
-- 
"I hope to die                                      ___       _____
before I *have* to use Microsoft Word.",           0--,|    /OOOOOOO\
Donald E. Knuth, 02-Oct-2001 in Tuebingen.        <_/  /  /OOOOOOOOOOO\
                                                    \  \/OOOOOOOOOOOOOOO\
                                                      \ OOOOOOOOOOOOOOOOO|//
Harald Koenig                                          \/\/\/\/\/\/\/\/\/
science+computing ag                                    //  /     \\  \
koenig@xxxxxxxxxxxxxxxxxxxx                            ^^^^^       ^^^^^

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-devel