[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index] Re: [Xen-devel] performace issue when turn on apicv
I will setup a test environment for xen upstream version but need some time. Meanwhile, can you give me some idea about what MAY cause this problem? I have been using xentrace to trace the problem, from what I see, the apicv feature itself works apicv=1 2583096 VMEXIT 19027420184 TSC WRMSR 459708 VMEXIT 6924749392 TSC External interrupt 293818 VMEXIT 451974088 TSC Virtualized EOI 843 VMEXIT 54729244 TSC I/O instruction 3260 VMEXIT 15979024 TSC Control-register accesses 1345 VMEXIT 2199736 TSC Exception or non-maskable interrupt (NMI) 39 VMEXIT 1516768 TSC EPT violation 54 VMEXIT 891712 TSC VMCALL 205 VMEXIT 370864 TSC CPUID apicv=0 3416159 VMEXIT 20929093044 TSC WRMSR 1098428 VMEXIT 11029334704 TSC External interrupt 41128 VMEXIT 64360924 TSC Interrupt window 664 VMEXIT 49245372 TSC I/O instruction 3221 VMEXIT 20116036 TSC Control-register accesses 1401 VMEXIT 2280412 TSC Exception or non-maskable interrupt (NMI) 39 VMEXIT 1581428 TSC EPT violation 53 VMEXIT 749588 TSC VMCALL 205 VMEXIT 355500 TSC CPUID 113 VMEXIT 298568 TSC RDMSR RDMSR gone,so "APIC Register Virtualization" works apicv=1 IRQ IRQ_MOVE_CLEANUP_VECTOR( 32): 21 IRQ LOCAL_TIMER_VECTOR( 249): 423804 IRQ CALL_FUNCTION_VECTOR( 251): 1171 IRQ EVENT_CHECK_VECTOR( 252): 1130 IRQ INVALIDATE_TLB_VECTOR( 253): 1 apicv=0 IRQ IRQ_MOVE_CLEANUP_VECTOR( 32): 22 IRQ LAST_DYNAMIC_VECTOR( 223): 27 IRQ LOCAL_TIMER_VECTOR( 249): 448057 IRQ CALL_FUNCTION_VECTOR( 251): 1173 IRQ EVENT_CHECK_VECTOR( 252): 608024 vmexit caused by External interrupt: EVENT_CHECK_VECTOR reduced a lot, so "Virtual Interrupt Delivery" and "Posted Interrupt Processing" works, I guess. I think the problem is not caused by apicv itself, maybe some other logic has conflict with apicv. On 2015/6/11 15:35, Zhang, Yang Z wrote: Liuqiming (John) wrote on 2015-06-11: > Hi, > > Recently I encounter a strange performance problem with APIC virtualization. > > My host has Intel(R) Xeon(R) CPU E7-4890 v2 CPU installed which support > APIC virtualization and x2apic, and there are 4 socket * 15 > cores_per_socket = 60 core available for VM. There is also a SSD disk on > host and the host support vt-d, so I can passsthrough this SSD to VM. > > A VM created with 60 vcpus, 400G memory and SSD device assigned. > I pin these vcpus 1:1 to phisical cpu, and in this VM only keep 15 vcpus > online. > The problem is: when apicv turn on, a significant performace decrease > can be observed and it seems related to cpu topology. > > I had test follow cases > apicv=1: > 1) ONLINE VCPU {0,1,2,3,4,5,6,7,8,9,10,11,12,13,14} > PIN TO > PCPU {0,1,2,3,4,5,6,7,8,9,10,11,12,13,14} > 2) ONLINE VCPU {0,5,9,13,17,21,25,29,33,37,41,45,49,53,57} > PIN TO > PCPU {0,5,9,13,17,21,25,29,33,37,41,45,49,53,57} > apicv=0: > 3) ONLINE VCPU {0,1,2,3,4,5,6,7,8,9,10,11,12,13,14} > PIN TO > PCPU {0,1,2,3,4,5,6,7,8,9,10,11,12,13,14} > 4) ONLINE VCPU {0,5,9,13,17,21,25,29,33,37,41,45,49,53,57} > PIN TO > PCPU {0,5,9,13,17,21,25,29,33,37,41,45,49,53,57} > the result is (the lower the better): > 1) 989 > 2) 2209 > 3) 978 > 4) 1130 > > It is a database testcase running on suse11sp3 system in the VM, and I > had traced that "read" and "write" syscall get much slower in 2) case. > > I have disabled NUMA in BIOS, so it seems apicv cause this bad > performance when using cpus in different nodes. > > Can any one shed some light on this? > > Btw, I am using xen 4.1.5 version with apicv backported, so I am not > sure whether something broken when backporting or just apicv behaves > this way. Can you retest it based on upstream Xen? Just as you suspected, your backporting may be the culprit. > > > Best regards, Yang _______________________________________________ Xen-devel mailing list Xen-devel@xxxxxxxxxxxxx http://lists.xen.org/xen-devel
|
Lists.xenproject.org is hosted with RackSpace, monitoring our |