On Sat, Aug 7, 2010 at 7:05 PM, Nirmal Guhan <vavat...@gmail.com> wrote: > On Sat, Aug 7, 2010 at 8:29 AM, Alex Williamson > <alex.william...@redhat.com> wrote: >> On Fri, 2010-08-06 at 17:29 -0700, Nirmal Guhan wrote: >>> On Thu, Aug 5, 2010 at 10:44 PM, Alex Williamson >>> <alex.william...@redhat.com> wrote: >>> > On Thu, Aug 5, 2010 at 12:53 PM, Nirmal Guhan <vavat...@gmail.com> wrote: >>> >> Hi, >>> >> >>> >> Am using Fedora 12 2.6.32.10-90.fc12.i686 on both host and guest. I >>> >> see that the packets destined for a particular port (iperf/5001 if >>> >> that matters) in guest can be captured using "tcpdump" on host whereas >>> >> the reverse is not true i.e I run iperf server on host and tcpdump on >>> >> guest can not read the packets sent to host. Is this expected behavior >>> >> ? >>> > >>> > Yes. >>> >>> So all the packets are recd by the host kernel and sent to guest ? Is >>> this the high level flow? >> >> Yes, when you used bridged/tap networking, all packets first go to the >> host, the bridge, then the guests. >> >>> > >>> >> I have enabled VT-d (through intel_iommu=on) and so was thinking that >>> >> guest will read the packets directly. If this is true, then wonder how >>> >> tcpdump on host can read guest packets ? or is my understanding wrong >>> >> ? Please clarify. >>> > >>> > Enabling VT-d on the host is only the first step, that doesn't >>> > automatically change the behavior of the guest. VT-d allows you to >>> > make use of the -pcidevice (or preferably -device pci-assign) option >>> > to kvm, which exposes a PCI device directly to the guest. For >>> > instance if you have a NIC that you want to dedicate to a guest at PCI >>> > address 00:19.0, you can use "-device pci-assign,host=00:19.0", which >>> > should show up (more than likely at a different PCI address) in the >>> > guest. (you'll have to unbind the device from host drivers, but the >>> > error messages will tell you how to do that) In this model, packets >>> > destined for the guest are only seen by the guest. >>> >>> Thanks. This worked, surprisingly with a performance penalty. The >>> guest ethernet device (eth2 in my case) came up with 10Mb/s speed. I >>> changed the speed to 100Mb/s using ethtool but still the performance >>> (Mbits/sec using iperf) did not improve. Any clues? >> >> What's the device? (lspci -vvv from the host) Link speed shouldn't >> depend on VM performance. What are you using to measure performance? > > It is Intel e1000e driver. > # lspci -vvv > 00:19.0 Ethernet controller: Intel Corporation Device 10ef (rev 06) > Subsystem: Intel Corporation Device 0000 > Control: I/O+ Mem+ BusMaster+ SpecCycle- MemWINV- VGASnoop- > ParErr- Stepping- SERR- FastB2B- DisINTx+ > Status: Cap+ 66MHz- UDF- FastB2B- ParErr- DEVSEL=fast >TAbort- > <TAbort- <MAbort- >SERR- <PERR- INTx- > Latency: 0 > Interrupt: pin A routed to IRQ 41 > Region 0: Memory at ff500000 (32-bit, non-prefetchable) [size=128K] > Region 1: Memory at ff570000 (32-bit, non-prefetchable) [size=4K] > Region 2: I/O ports at f040 [size=32] > Capabilities: [c8] Power Management version 2 > Flags: PMEClk- DSI+ D1- D2- AuxCurrent=0mA > PME(D0+,D1-,D2-,D3hot+,D3cold+) > Status: D0 NoSoftRst- PME-Enable- DSel=0 DScale=1 PME- > Capabilities: [d0] MSI: Enable+ Count=1/1 Maskable- 64bit+ > Address: 00000000fee0f00c Data: 4182 > Capabilities: [e0] PCI Advanced Features > AFCap: TP+ FLR+ > AFCtrl: FLR- > AFStatus: TP- > Kernel driver in use: e1000e > Kernel modules: e1000e > > Am using "iperf" to measure performance with identical invocation for > both pci-passthrough(VT-d) and no pci-passthrough case. Command used : > "iperf -c <addr> -w 16000" <- window size selected was 32K though 16K > was requested. > 0.0-30.0 sec 33.6 MBytes 9.39 Mbits/sec <- guest with pci pass-through + > VT-d > 0.0-30.0 sec 324 MBytes 90.6 Mbits/sec <-- guest without pci pass-through > 0.0-30.0 sec 334 MBytes 93.3 Mbits/sec <--- host > > --Nirmal > > > > > >> Thanks, >> >> Alex
Adding kvm forum back. Please help! Thanks, Nirmal -- To unsubscribe from this list: send the line "unsubscribe kvm" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html