[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

RE: [Xen-devel] Dom0 Hang for large VCPU counts > PCPU


  • To: "Zulauf, John" <john.zulauf@xxxxxxxxx>, xen-devel@xxxxxxxxxxxxxxxxxxx
  • From: "Woller, Thomas" <thomas.woller@xxxxxxx>
  • Date: Sun, 28 Jan 2007 12:52:10 -0600
  • Delivery-date: Sun, 28 Jan 2007 10:55:06 -0800
  • List-id: Xen developer discussion <xen-devel.lists.xensource.com>
  • Thread-index: Acc/6WSeIoDi05TDRUmkOjJugvswfgABJXXgAD0PtgAAh799cA==
  • Thread-topic: [Xen-devel] Dom0 Hang for large VCPU counts > PCPU

More info.  Using c/s 13628 64b smp hv, with SMP HVM suseLinux10 and
Opensuse10.2 64b guests, w/ gfx enabled.  
Guest config parms: 32 VCPUS, 2000 memory, pae/acpi/apic=1,
shadow_memory=512, no vif line
Machine has 8 physical cores in AMD-V system, w/ 16G physical RAM.

SuseLinux10 (8 PCPU) = boots 32 VCPUs without issue w/ standard guest
kernel config 

SuseLinux10 (2 PCPU maxcpus=2) = boots 32 VCPUs without issue (quite a
bit slower) w/ standard guest kernel config, guest very unresponsive,
and dom0 very slow response but better than guest.  Vcpu-list shows that
all 32 VCPUs are running though.  Unable to login to guest after 10
minutes, after entering login/passwd.

OpenSuse10.2 = hangs on boot with standard guest kernel boot options 
        adding acpi=off allows boot
        adding clocksource=acpi_pm did not help, still hangs on boot
(black screen, no splash ever displayed)

I'll run some overnight tests on the 8 PCPU SUSE10 setup, and see if it
is stable wrt >16 VCPUs running.
tom

> -----Original Message-----
> From: Zulauf, John [mailto:john.zulauf@xxxxxxxxx] 
> Sent: Thursday, January 25, 2007 6:43 PM
> To: Woller, Thomas; xen-devel@xxxxxxxxxxxxxxxxxxx
> Subject: RE: [Xen-devel] Dom0 Hang for large VCPU counts > PCPU
> 
> Further testing:
> 
> A brief update the DomU "crash for large VCPU count" 
> 
> For PCPU == 2 (single Core 2 Duo/965 platform) host machine 
> hangs for with VCPU's > 20.  
> 
> For PCPU == 8 (Dual quadcore/Bensley)
> 
> we have tested up to 24 VCPU successfully. However, we've 
> seen FC6 rebooting the DomU sporadically with VCPU > 16.  The 
> suggested clock_source option has no effect.
> 
> 
> -----Original Message-----
> From: Woller, Thomas [mailto:thomas.woller@xxxxxxx]
> Sent: Wednesday, January 24, 2007 11:35 AM
> To: Zulauf, John; xen-devel@xxxxxxxxxxxxxxxxxxx
> Subject: RE: [Xen-devel] Dom0 Hang for large VCPU counts > PCPU
> 
> Not sure if this is useful.  We have a box with 8 cores, and 
> can run 32 VCPUs without issue on AMD-V with suse10 64b smp guest.
> This data though is from around january 8th, so it's a bit 
> stale.  I don't have the exact c/s that the tests were run 
> on, or the guest config parms, but I think it was with 6Gig 
> of RAM for the guest. :P  I can try this guest again next day 
> or 2 if useful.
> tom
> 
>  XEND_DEBUG = 1
> Name                              ID  VCPU   CPU State   Time(s) CPU
> Affinity
> Domain-0                           0     0     0   -b-     
> 172.5 any cpu
> Domain-0                           0     1     1   -b-      
> 60.0 any cpu
> Domain-0                           0     2     3   -b-      
> 25.2 any cpu
> Domain-0                           0     3     3   r--      
> 10.2 any cpu
> Domain-0                           0     4     2   -b-       
> 8.7 any cpu
> Domain-0                           0     5     5   -b-       
> 6.8 any cpu
> Domain-0                           0     6     6   -b-       
> 3.6 any cpu
> Domain-0                           0     7     0   -b-       
> 4.8 any cpu
> suse10_x64_smp                     4     0     6   -b-     
> 144.2 any cpu
> suse10_x64_smp                     4     1     4   -b-      
> 36.3 any cpu
> suse10_x64_smp                     4     2     4   -b-      
> 26.5 any cpu
> suse10_x64_smp                     4     3     3   ---     
> 883.8 any cpu
> suse10_x64_smp                     4     4     5   r--     
> 885.6 any cpu
> suse10_x64_smp                     4     5     6   ---     
> 883.6 any cpu
> suse10_x64_smp                     4     6     7   ---     
> 884.2 any cpu
> suse10_x64_smp                     4     7     2   ---     
> 884.4 any cpu
> suse10_x64_smp                     4     8     4   ---     
> 886.8 any cpu
> suse10_x64_smp                     4     9     7   r--     
> 885.6 any cpu
> suse10_x64_smp                     4    10     6   ---     
> 885.2 any cpu
> suse10_x64_smp                     4    11     4   r--     
> 884.0 any cpu
> suse10_x64_smp                     4    12     0   r--     
> 884.6 any cpu
> suse10_x64_smp                     4    13     3   ---     
> 883.7 any cpu
> suse10_x64_smp                     4    14     1   ---     
> 887.0 any cpu
> suse10_x64_smp                     4    15     1   ---     
> 884.7 any cpu
> suse10_x64_smp                     4    16     0   ---     
> 885.5 any cpu
> suse10_x64_smp                     4    17     7   ---     
> 884.2 any cpu
> suse10_x64_smp                     4    18     2   r--     
> 885.9 any cpu
> suse10_x64_smp                     4    19     1   ---     
> 886.0 any cpu
> suse10_x64_smp                     4    20     6   ---     
> 885.7 any cpu
> suse10_x64_smp                     4    21     6   ---     
> 886.4 any cpu
> suse10_x64_smp                     4    22     2   ---     
> 885.6 any cpu
> suse10_x64_smp                     4    23     5   ---     
> 888.9 any cpu
> suse10_x64_smp                     4    24     4   ---     
> 885.0 any cpu
> suse10_x64_smp                     4    25     0   ---     
> 885.2 any cpu
> suse10_x64_smp                     4    26     4   ---     
> 885.0 any cpu
> suse10_x64_smp                     4    27     7   ---     
> 885.1 any cpu
> suse10_x64_smp                     4    28     4   ---     
> 882.4 any cpu
> suse10_x64_smp                     4    29     6   ---     
> 884.1 any cpu
> suse10_x64_smp                     4    30     5   ---     
> 883.6 any cpu
> suse10_x64_smp                     4    31     1   r--     
> 885.3 any cpu
> 
> 
> ________________________________
> 
>       From: xen-devel-bounces@xxxxxxxxxxxxxxxxxxx
> [mailto:xen-devel-bounces@xxxxxxxxxxxxxxxxxxx] On Behalf Of 
> Zulauf, John
>       Sent: Wednesday, January 24, 2007 12:57 PM
>       To: xen-devel@xxxxxxxxxxxxxxxxxxx
>       Subject: [Xen-devel] Dom0 Hang for large VCPU counts > PCPU
>       
>       
> 
>       We have been experimenting with large VCPU counts >> 
> PCPU and have succeeded in hanging Dom0 in or during /sbin/loader.
> 
>        
> 
>       Using xen-3.0.4-testing with an HVM booting the FC6 DVD on a
> Core2 Duo (i.e. 2 PCPU) on a 965 chipset
> 
>        
> 
>       VCPU=17 - works
> 
>       VCPU=20 - works (takes a very long time)
> 
>       VCPU=24 - lockup (whole machine, yes I mean Dom0)
> 
>       VCPU=31 - same
> 
>       VCPU=32 - same
> 
>        
> 
>       We've noted in the Xen l-apic code a hard 32 CPU limit 
> (a uint32 used as an l-apic (vcpu?) bitmask), but this looks 
> to be unrelated.
> 
>        
> 
>       John Zulauf
> 
>       Intel Corporation
> 
>        
> 
> 
> 



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


 


Rackspace

Lists.xenproject.org is hosted with RackSpace, monitoring our
servers 24x7x365 and backed by RackSpace's Fanatical Support®.