James Miller
2008-Oct-15 20:20 UTC
[Xen-users] QueryPerformaceCounter and TSC returns bad results with Windows HVM
Hi everyone,
We have a server with 2 sockets, 4 cores each, running CentOS release 5
(Final) and Xen 3.0.3. On that server we have 7 Windows 2003 DomUs.
We''ve locked each one to a specific core and each has a single CPU HAL.
We have an application which uses QueryPerformanceCounter. On very rare
occasions a call to QPC returns a bad result causing our processes to
disconnect from the server.
I''ve read about some bugs with HPC
http://bugzilla.xensource.com/bugzilla/show_bug.cgi?id=1291
I''ve also read something about not using APCI somewhat mitigates this
issue.
Could someone tell me if this issues is being or has been resolved by
later versions of Xen?
I have, below, a sample configuration of a Windows HVM DomU as well as
the output of xm vcpu-list. Please let me know if there is any
additional information I can provide to help resolve this issue and I
greatly appreciate all your help.
Sample config
import os, re
arch = os.uname()[4]
if re.search(''64'', arch):
arch_libdir = ''lib64''
else:
arch_libdir = ''lib''
name = ''evalvm01.vm''
kernel = ''/usr/lib/xen/boot/hvmloader''
builder=''hvm''
boot=''c''
memory = 2048
vif = [ ''type=ioemu, vifname=vifevalvm01, mac=aa:00:ef:39:a0:08
,bridge=xenbr1'' ]
device_model = ''/usr/'' + arch_libdir +
''/xen/bin/qemu-dm''
vnc=1
sdl=0
vcpus = 1
cpus ="1"
vnclisten=''0.0.0.0''
vncpasswd=''xxxxxxxxxx''
disk = [ ''phy:/dev/vg01/evalvm01,ioemu:hda,w'']
acpi=1
vncunused=1
vncdisplay=1
xm vcpu-list
Name ID VCPUs CPU State Time(s) CPU
Affinity
Domain-0 0 0 0 r-- 2794806.2 0
Domain-0 0 1 - --p 2.0 any cpu
Domain-0 0 2 - --p 1.1 any cpu
Domain-0 0 3 - --p 1.4 any cpu
Domain-0 0 4 - --p 0.9 any cpu
Domain-0 0 5 - --p 0.9 any cpu
Domain-0 0 6 - --p 0.8 any cpu
Domain-0 0 7 - --p 0.9 any cpu
evalvm01.vm 118 0 1 -b- 27903.4 1
evalvm02.vm 112 0 2 -b- 10507.8 2
evalvm03.vm 113 0 3 -b- 2445.3 3
evalvm04.vm 116 0 4 -b- 79872.1 4
evalvm05.vm 119 0 5 -b- 38717.1 5
evalvm06.vm 115 0 6 -b- 98750.3 6
evalvm07.vm 117 0 7 r-- 65222.2 7
Jim
_______________________________________________
Xen-users mailing list
Xen-users@lists.xensource.com
http://lists.xensource.com/xen-users
James Miller
2008-Oct-16 14:47 UTC
2nd request -- Re: [Xen-users] QueryPerformaceCounter and TSC returns bad results with Windows HVM
James Miller wrote:> Hi everyone, > > We have a server with 2 sockets, 4 cores each, running CentOS release > 5 (Final) and Xen 3.0.3. On that server we have 7 Windows 2003 DomUs. > We''ve locked each one to a specific core and each has a single CPU > HAL. We have an application which uses QueryPerformanceCounter. On > very rare occasions a call to QPC returns a bad result causing our > processes to disconnect from the server. > I''ve read about some bugs with HPC > http://bugzilla.xensource.com/bugzilla/show_bug.cgi?id=1291 > I''ve also read something about not using APCI somewhat mitigates this > issue. > > Could someone tell me if this issues is being or has been resolved by > later versions of Xen? > > I have, below, a sample configuration of a Windows HVM DomU as well as > the output of xm vcpu-list. Please let me know if there is any > additional information I can provide to help resolve this issue and I > greatly appreciate all your help. > > > Sample config > import os, re > arch = os.uname()[4] > if re.search(''64'', arch): > arch_libdir = ''lib64'' > else: > arch_libdir = ''lib'' > name = ''evalvm01.vm'' > kernel = ''/usr/lib/xen/boot/hvmloader'' > builder=''hvm'' > boot=''c'' > memory = 2048 > vif = [ ''type=ioemu, vifname=vifevalvm01, mac=aa:00:ef:39:a0:08 > ,bridge=xenbr1'' ] > device_model = ''/usr/'' + arch_libdir + ''/xen/bin/qemu-dm'' > vnc=1 > sdl=0 > vcpus = 1 > cpus ="1" > vnclisten=''0.0.0.0'' > vncpasswd=''xxxxxxxxxx'' > disk = [ ''phy:/dev/vg01/evalvm01,ioemu:hda,w''] > acpi=1 > vncunused=1 > vncdisplay=1 > > xm vcpu-list > Name ID VCPUs CPU State Time(s) CPU > Affinity > Domain-0 0 0 0 r-- 2794806.2 0 > Domain-0 0 1 - --p 2.0 any cpu > Domain-0 0 2 - --p 1.1 any cpu > Domain-0 0 3 - --p 1.4 any cpu > Domain-0 0 4 - --p 0.9 any cpu > Domain-0 0 5 - --p 0.9 any cpu > Domain-0 0 6 - --p 0.8 any cpu > Domain-0 0 7 - --p 0.9 any cpu > evalvm01.vm 118 0 1 -b- 27903.4 1 > evalvm02.vm 112 0 2 -b- 10507.8 2 > evalvm03.vm 113 0 3 -b- 2445.3 3 > evalvm04.vm 116 0 4 -b- 79872.1 4 > evalvm05.vm 119 0 5 -b- 38717.1 5 > evalvm06.vm 115 0 6 -b- 98750.3 6 > evalvm07.vm 117 0 7 r-- 65222.2 7 > > > Jim > >Hi everyone, I hate to be a pest but this is becoming a _REAL_ show stopper for us going forward with using Xen. I would appreciate any thoughts, if I''m asking in the wrong way or not providing enough info _PLEASE_ let me know. About the only other thing I can offer is that the problem happens very infrequently but when it does the process will report, for example, that it went 2,451,046ms w/out talking to the server that''s like ~40min meaning the QPC gave an erroneous response. Almost as if a 64bit counter (I think it''s a 64bit counter) was rolling over?.! Anyway, thanks in advance for any suggestions or assistance Jim _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
James Miller
2008-Oct-21 22:03 UTC
3rd Request Re: 2nd request -- Re: [Xen-users] QueryPerformaceCounter and TSC returns bad results with Windows HVM
James Miller wrote:> James Miller wrote: >> Hi everyone, >> >> We have a server with 2 sockets, 4 cores each, running CentOS release >> 5 (Final) and Xen 3.0.3. On that server we have 7 Windows 2003 >> DomUs. We''ve locked each one to a specific core and each has a >> single CPU HAL. We have an application which uses >> QueryPerformanceCounter. On very rare occasions a call to QPC >> returns a bad result causing our processes to disconnect from the >> server. >> I''ve read about some bugs with HPC >> http://bugzilla.xensource.com/bugzilla/show_bug.cgi?id=1291 >> I''ve also read something about not using APCI somewhat mitigates this >> issue. >> >> Could someone tell me if this issues is being or has been resolved by >> later versions of Xen? >> >> I have, below, a sample configuration of a Windows HVM DomU as well >> as the output of xm vcpu-list. Please let me know if there is any >> additional information I can provide to help resolve this issue and I >> greatly appreciate all your help. >> >> >> Sample config >> import os, re >> arch = os.uname()[4] >> if re.search(''64'', arch): >> arch_libdir = ''lib64'' >> else: >> arch_libdir = ''lib'' >> name = ''evalvm01.vm'' >> kernel = ''/usr/lib/xen/boot/hvmloader'' >> builder=''hvm'' >> boot=''c'' >> memory = 2048 >> vif = [ ''type=ioemu, vifname=vifevalvm01, mac=aa:00:ef:39:a0:08 >> ,bridge=xenbr1'' ] >> device_model = ''/usr/'' + arch_libdir + ''/xen/bin/qemu-dm'' >> vnc=1 >> sdl=0 >> vcpus = 1 >> cpus ="1" >> vnclisten=''0.0.0.0'' >> vncpasswd=''xxxxxxxxxx'' >> disk = [ ''phy:/dev/vg01/evalvm01,ioemu:hda,w''] >> acpi=1 >> vncunused=1 >> vncdisplay=1 >> >> xm vcpu-list >> Name ID VCPUs CPU State Time(s) CPU >> Affinity >> Domain-0 0 0 0 r-- 2794806.2 0 >> Domain-0 0 1 - --p 2.0 any cpu >> Domain-0 0 2 - --p 1.1 any cpu >> Domain-0 0 3 - --p 1.4 any cpu >> Domain-0 0 4 - --p 0.9 any cpu >> Domain-0 0 5 - --p 0.9 any cpu >> Domain-0 0 6 - --p 0.8 any cpu >> Domain-0 0 7 - --p 0.9 any cpu >> evalvm01.vm 118 0 1 -b- 27903.4 1 >> evalvm02.vm 112 0 2 -b- 10507.8 2 >> evalvm03.vm 113 0 3 -b- 2445.3 3 >> evalvm04.vm 116 0 4 -b- 79872.1 4 >> evalvm05.vm 119 0 5 -b- 38717.1 5 >> evalvm06.vm 115 0 6 -b- 98750.3 6 >> evalvm07.vm 117 0 7 r-- 65222.2 7 >> >> >> Jim >> >> > > Hi everyone, > > I hate to be a pest but this is becoming a _REAL_ show stopper for us > going forward with using Xen. I would appreciate any thoughts, if I''m > asking in the wrong way or not providing enough info _PLEASE_ let me > know. > > About the only other thing I can offer is that the problem happens > very infrequently but when it does the process will report, for > example, that it went 2,451,046ms w/out talking to the server that''s > like ~40min meaning the QPC gave an erroneous response. Almost as if > a 64bit counter (I think it''s a 64bit counter) was rolling over?.! > > Anyway, thanks in advance for any suggestions or assistance > > Jim > > > >Soooo one of our programmers wrote a little program to test the issue and here''s the results: C:\RS>qpc_test.exe FREQUENCY = 3579545 cycles per second (369E99) COUNTER = 26206348760 cycles (61A0525D8) THRESHOLD exceeded: diff= 8589970453, threshold = 500, threshold * freq / 1000 1789772 THRESHOLD exceeded: last_counter = 749471789887, this_Counter = 758061760340 CAN ANYONE PLEASE COMMENT ON THIS ISSUE? _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users