Re: [Xen-ia64-devel] PATCH: xencomm [2]

[Akio Takebe <takebe_akio@xxxxxxxxxxxxxx>]

Hi, Aron

I checked the fragmentation on xen/ia64 with on xen/x86.

The following data is /proc/budddyinfo after starting each daemons.
1. buddyinfo of ia64 at the boottime. (dom0_mem=512MB)
Mon Oct  2 20:42:24 JST 2006
----------iptables-----------
Node 0, zone      DMA      1      2      1      1      1      1      0    
  0      0      1      9 
----------iptables-----------
----------portmap-----------
Node 0, zone      DMA      0      1      1      1      1      0      1    
  0      1      0      1 
----------portmap-----------
----------ssh-----------
Node 0, zone      DMA      1      3      9      3      1      1      1    
  1      0      0      0 
----------ssh-----------
----------sendmail-----------
Node 0, zone      DMA     19      7      5      1      5      1      0    
  1      0      0      0 
----------sendmail-----------
----------haldaemon-----------
Node 0, zone      DMA     20     10      5      1      4      0      1    
  1      0      0      0 
----------haldaemon-----------

2. buddyinfo of x86 at the boottime. (dom0_mem=128MB)

Mon Oct  2 21:22:29 JST 2006
----------iptables-----------
Node 0, zone      DMA     47     11      4      1      0      1      8    
  4      1      0     11 
----------iptables-----------
----------portmap-----------
Node 0, zone      DMA      3      2      0      1      0      0      1    
  0      1      1      9 
----------portmap-----------
----------ssh-----------
Node 0, zone      DMA     23      3      1      1      1      1      1    
  3      2      0      3 
----------ssh-----------
----------sendmail-----------
Node 0, zone      DMA     26      5      4      0      0      0      4    
  3      3      1      1 
----------sendmail-----------
----------haldaemon-----------
Node 0, zone      DMA     44     14     23     14      2      1      1    
  0      1      0      0 
----------haldaemon-----------
----------before stating xend-----------
Node 0, zone      DMA     22     25     24     14      2      1      1    
  0      1      0      0 
----------before stating xend-----------

I think fragmentations of ia64 is a little quickly.
But allocate of oder:8 is hard on also x86 at the starting of xend.

My Xen/IA64 environment have 16 cpus(4 sockets, 8cores, 16threads), 
and Xen/x86 environment have 2core cpus(1 sockets, 2cores).
The difference of the fragmentations may be caused by the number of cpus.
What do you think about it?

Best Regards,

Akio Takebe

>Akio Takebe wrote:  [Fri Sep 29 2006, 11:41:57AM EDT]
>> Hmmm... But this issue was happened on x86-PAE.
>> https://bugzilla.redhat.com/bugzilla/show_bug.cgi?id=201796
>> 
>> Actually with dom0_mem=512M on FC6-ia64,
>> we can success installing blkbk on rare occasions.
>
>On PAE the problem was "fixed" by loading blkbk/netbk in
>/etc/init.d/xend.  This moves the module load early enough in the
>system boot to fix it on that arch.
>
>However on ia64 this isn't sufficient.  We need either for the modules
>to load *earlier* (such as in the initramfs) or we need to use
>dom0_mem=1G (for example)
>
>I have some questions, though:
>
>1. On x86, all of the memory is assigned to dom0, then it balloons to
>   give memory to unpriv'd domains.  Why does ia64 give only 512M to
>   dom0 instead of doing the same as x86?
>
>2. Juan says that he can boot x86 xen with only 128M of memory, and
>   there's still enough physically contiguous to load blkbk and netbk
>   when xend loads.  So why does ia64 need 8x the memory to do avoid
>   fragmentation?  It seems like we must be doing something wrong for
>   memory to fragment so quickly...!
>
>Thanks,
>Aron


_______________________________________________
Xen-ia64-devel mailing list
Xen-ia64-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-ia64-devel