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

[Borut Mrak <b@xxxxxxxxxx>]

Mark Williamson wrote:
> YOu need to set a dom0_mem on the Xen command line, as described by the other 
> poster.
This did not work for us. It did improve the situation, but we still have 
crashes twice a week (instead of daily).
HW:
Intel Xeon (both 50xx and 51xx exhibit the same problem)
Intel 5000 chipset (motherboards from Intel and Tyan)
Areca PCIe RAID (1200 series)
8GB of RAM (around 6 used in 8 VMs, including dom0_max_mem=512MB)

SW:
Debian Etch
Xen 3.0.3
Kernel 2.6.17 (the one in Debian), patched with:
  * arcmsr driver from Areca
  * newer e1000 driver from Intel (integrated PCIe NICs on these boards
    are not supported on 2.6.17)

> If you don't do this dom0 takes all the system memory and balloons itself down 
> in order to start domains.  ISTR the ballooning confuses Linux's memory 
> allocator somehow, and this shows up impacting the net driver.
As you see from above, no balooning is going on here. dom0_max_mem is set 
to 512MB, and all other domains combined still leave at least a GB or two 
free.
> Setting dom0_mem explicitly on the Xen commandline means that the auto 
> ballooning isn't necessary (although you may still want to switch off 
> auto-ballooning and/or set a new minimum dom0 memory level so that dom0 will 
> *never* get shrunk, even if you run out of other memory)
> Does this make some sense?  Hopefully it'll fix your problem, please let us 
> know.
Thank you for this suggestion. I will try setting dom0-min-mem to 0 and 
see what happens. I'll report here.
Is it necessary to reboot the whole machine for this setting to take 
effect, or is restarting xend enough?
regards,
Borut Mrak.

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