Re: [OmniOS-discuss] The ixgbe driver, Lindsay Lohan, and the Greek economy
Hello all, Well, I no longer blame the ixgbe driver for the problems I'm seeing. I tried Joerg's updated driver, which didn't improve the issue. So I went back to the drawing board and rebuilt the server from scratch. What I noted is that if I have only a single 1-gig physical interface active on the ESXi host, everything works as expected. As soon as I enable two interfaces, I start seeing the performance problems I've described. Response pauses from the server that I see in TCPdumps are still leading me to believe the problem is delay on the server side, so I ran a series of kernel dtraces and produced some flamegraphs. This was taken during a read operation with two active 10G interfaces on the server, with a single target being shared by two tpgs- one tpg for each 10G physical port. The host device has two 1G ports enabled, with VLANs separating the active ports into 10G/1G pairs. ESXi is set to multipath using both VLANS with a round-robin IO interval of 1. https://drive.google.com/file/d/0BwyUMjibonYQd3ZYOGh4d2pteGs/view?usp=sharing This was taken during a write operation: https://drive.google.com/file/d/0BwyUMjibonYQMnBtU1Q2SXM2ams/view?usp=sharing I then rebooted the server and disabled C-State, ACPI T-State, and general EIST (Turbo boost) functionality in the CPU. I when I attempted to boot my guest VM, the iSCSI transfer gradually ground to a halt during the boot loading process, and the guest OS never did complete its boot process. Here is a flamegraph taken while iSCSI is slowly dying: https://drive.google.com/file/d/0BwyUMjibonYQM21JeFZPX3dZWTg/view?usp=sharing I edited out cpu_idle_adaptive from the dtrace output and regenerated the slowdown graph: https://drive.google.com/file/d/0BwyUMjibonYQbTVwV3NvXzlPS1E/view?usp=sharing I then edited cpu_idle_adaptive out of the speedy write operation and regenerated that graph: https://drive.google.com/file/d/0BwyUMjibonYQeWFYM0pCMDZ1X2s/view?usp=sharing I have zero experience with interpreting flamegraphs, but the most significant difference I see between the slow read example and the fast write example is in unix`thread_start -- unix`idle. There's a good chunk of unix`i86_mwait in the read example that is not present in the write example at all. Disabling the l2arc cache device didn't make a difference, and I had to reenable EIST support on the CPU to get my VMs to boot. I am seeing a variety of bug reports going back to 2010 regarding excessive mwait operations, with the suggested solutions usually being to set cpupm enable poll-mode in power.conf. That change also had no effect on speed. -Warren V -Original Message- From: Chris Siebenmann [mailto:c...@cs.toronto.edu c...@cs.toronto.edu] Sent: Monday, February 23, 2015 8:30 AM To: W Verb Cc: omnios-discuss@lists.omniti.com; c...@cs.toronto.edu Subject: Re: [OmniOS-discuss] The ixgbe driver, Lindsay Lohan, and the Greek economy Chris, thanks for your specific details. I'd appreciate it if you could tell me which copper NIC you tried, as well as to pass on the iSCSI tuning parameters. Our copper NIC experience is with onboard X540-AT2 ports on SuperMicro hardware (which have the guaranteed 10-20 msec lock hold) and dual-port 82599EB TN cards (which have some sort of driver/hardware failure under load that eventually leads to 2-second lock holds). I can't recommend either with the current driver; we had to revert to 1G networking in order to get stable servers. The iSCSI parameter modifications we do, across both initiators and targets, are: initialr2tno firstburstlength 128k maxrecvdataseglen 128k[only on Linux backends] maxxmitdataseglen 128k[only on Linux backends] The OmniOS initiator doesn't need tuning for more than the first two parameters; on the Linux backends we tune up all four. My extended thoughts on these tuning parameters and why we touch them can be found here: http://utcc.utoronto.ca/~cks/space/blog/tech/UnderstandingiSCSIProtocol http://utcc.utoronto.ca/~cks/space/blog/tech/LikelyISCSITuning The short version is that these parameters probably only make a small difference but their overall goal is to do 128KB ZFS reads and writes in single iSCSI operations (although they will be fragmented at the TCP layer) and to do iSCSI writes without a back-and-forth delay between initiator and target (that's 'initialr2t no'). I think basically everyone should use InitialR2T set to no and in fact that it should be the software default. These days only unusually limited iSCSI targets should need it to be otherwise and they can change their setting for it (initiator and target must both agree to it being 'yes', so either can veto it). - cks On Mon, Feb 23, 2015 at 8:21 AM, Joerg Goltermann j...@osn.de wrote: Hi, I think your problem is caused by your link properties or your switch settings. In general the standard ixgbe seems to perform
Re: [OmniOS-discuss] The ixgbe driver, Lindsay Lohan, and the Greek economy
Hi, I think your problem is caused by your link properties or your switch settings. In general the standard ixgbe seems to perform well. I had trouble after changing the default flow control settings to bi and this was my motivation to update the ixgbe driver a long time ago. After I have updated our systems to ixgbe 2.5.8 I never had any problems Make sure your switch has support for jumbo frames and you use the same mtu on all ports, otherwise the smallest will be used. What switch do you use? I can tell you nice horror stories about different vendors - Joerg On 23.02.2015 10:31, W Verb wrote: Thank you Joerg, I've downloaded the package and will try it tomorrow. The only thing I can add at this point is that upon review of my testing, I may have performed my pkg -u between the initial quad-gig performance test and installing the 10G NIC. So this may be a new problem introduced in the latest updates. Those of you who are running 10G and have not upgraded to the latest kernel, etc, might want to do some additional testing before running the update. -Warren V On Mon, Feb 23, 2015 at 1:15 AM, Joerg Goltermann j...@osn.de mailto:j...@osn.de wrote: Hi, I remember there was a problem with the flow control settings in the ixgbe driver, so I updated it a long time ago for our internal servers to 2.5.8. Last weekend I integrated the latest changes from the FreeBSD driver to bring the illumos ixgbe to 2.5.25 but I had no time to test it, so it's completely untested! If you would like to give the latest driver a try you can fetch the kernel modules from https://cloud.osn.de/index.__php/s/Fb4so9RsNnXA7r9 https://cloud.osn.de/index.php/s/Fb4so9RsNnXA7r9 Clone your boot environment, place the modules in the new environment and update the boot-archive of the new BE. - Joerg On 23.02.2015 02:54, W Verb wrote: By the way, to those of you who have working setups: please send me your pool/volume settings, interface linkprops, and any kernel tuning parameters you may have set. Thanks, Warren V On Sat, Feb 21, 2015 at 7:59 AM, Schweiss, Chip c...@innovates.com mailto:c...@innovates.com wrote: I can't say I totally agree with your performance assessment. I run Intel X520 in all my OmniOS boxes. Here is a capture of nfssvrtop I made while running many storage vMotions between two OmniOS boxes hosting NFS datastores. This is a 10 host VMware cluster. Both OmniOS boxes are dual 10G connected with copper twin-ax to the in rack Nexus 5010. VMware does 100% sync writes, I use ZeusRAM SSDs for log devices. -Chip 2014 Apr 24 08:05:51, load: 12.64, read: 17330243 KB, swrite: 15985KB, awrite: 1875455 KB Ver Client NFSOPS Reads SWrites AWrites Commits Rd_bw SWr_bw AWr_bwRd_t SWr_t AWr_t Com_t Align% 4 10.28.17.105 0 0 0 0 0 0 0 0 0 0 0 0 0 4 10.28.17.215 0 0 0 0 0 0 0 0 0 0 0 0 0 4 10.28.17.213 0 0 0 0 0 0 0 0 0 0 0 0 0 4 10.28.16.151 0 0 0 0 0 0 0 0 0 0 0 0 0 4 all 1 0 0 0 0 0 0 0 0 0 0 0 0 3 10.28.16.175 3 0 3 0 0 1 11 04806 48 0 0 85 3 10.28.16.183 6 0 6 0 0 3 162 0 549 124 0 0 73 3 10.28.16.180 11 0 10 0 0 3 27 0 776 89 0 0 67 3 10.28.16.176 28 2 26 0 0 10 405 02572 198 0 0 100 3 10.28.16.178 46064602 4 0 0 294534 3 0 723 49 0 0 99 3 10.28.16.179 49054879 26 0 0 312208 311 0 735 271 0 0 99 3 10.28.16.181 55155502 13 0 0 352107 77 0
Re: [OmniOS-discuss] The ixgbe driver, Lindsay Lohan, and the Greek economy
Hi, I remember there was a problem with the flow control settings in the ixgbe driver, so I updated it a long time ago for our internal servers to 2.5.8. Last weekend I integrated the latest changes from the FreeBSD driver to bring the illumos ixgbe to 2.5.25 but I had no time to test it, so it's completely untested! If you would like to give the latest driver a try you can fetch the kernel modules from https://cloud.osn.de/index.php/s/Fb4so9RsNnXA7r9 Clone your boot environment, place the modules in the new environment and update the boot-archive of the new BE. - Joerg On 23.02.2015 02:54, W Verb wrote: By the way, to those of you who have working setups: please send me your pool/volume settings, interface linkprops, and any kernel tuning parameters you may have set. Thanks, Warren V On Sat, Feb 21, 2015 at 7:59 AM, Schweiss, Chip c...@innovates.com wrote: I can't say I totally agree with your performance assessment. I run Intel X520 in all my OmniOS boxes. Here is a capture of nfssvrtop I made while running many storage vMotions between two OmniOS boxes hosting NFS datastores. This is a 10 host VMware cluster. Both OmniOS boxes are dual 10G connected with copper twin-ax to the in rack Nexus 5010. VMware does 100% sync writes, I use ZeusRAM SSDs for log devices. -Chip 2014 Apr 24 08:05:51, load: 12.64, read: 17330243 KB, swrite: 15985KB, awrite: 1875455 KB Ver Client NFSOPS Reads SWrites AWrites Commits Rd_bw SWr_bw AWr_bwRd_t SWr_t AWr_t Com_t Align% 4 10.28.17.105 0 0 0 0 0 0 0 0 0 0 0 0 0 4 10.28.17.215 0 0 0 0 0 0 0 0 0 0 0 0 0 4 10.28.17.213 0 0 0 0 0 0 0 0 0 0 0 0 0 4 10.28.16.151 0 0 0 0 0 0 0 0 0 0 0 0 0 4 all 1 0 0 0 0 0 0 0 0 0 0 0 0 3 10.28.16.175 3 0 3 0 0 1 11 04806 48 0 0 85 3 10.28.16.183 6 0 6 0 0 3 162 0 549 124 0 0 73 3 10.28.16.180 11 0 10 0 0 3 27 0 776 89 0 0 67 3 10.28.16.176 28 2 26 0 0 10 405 02572 198 0 0 100 3 10.28.16.178 46064602 4 0 0 294534 3 0 723 49 0 0 99 3 10.28.16.179 49054879 26 0 0 312208 311 0 735 271 0 0 99 3 10.28.16.181 55155502 13 0 0 352107 77 0 89 87 0 0 99 3 10.28.16.184 12095 12059 10 0 0 763014 39 0 249 147 0 0 99 3 10.28.58.1154016040 1166354 53 191605 474 202346 192 96 144 83 99 3 all 42574 33086 2176354 53 1913488 1582 202300 348 138 153 105 99 On Fri, Feb 20, 2015 at 11:46 PM, W Verb wver...@gmail.com wrote: Hello All, Thank you for your replies. I tried a few things, and found the following: 1: Disabling hyperthreading support in the BIOS drops performance overall by a factor of 4. 2: Disabling VT support also seems to have some effect, although it appears to be minor. But this has the amusing side effect of fixing the hangs I've been experiencing with fast reboot. Probably by disabling kvm. 3: The performance tests are a bit tricky to quantify because of caching effects. In fact, I'm not entirely sure what is happening here. It's just best to describe what I'm seeing: The commands I'm using to test are dd if=/dev/zero of=./test.dd bs=2M count=5000 dd of=/dev/null if=./test.dd bs=2M count=5000 The host vm is running Centos 6.6, and has the latest vmtools installed. There is a host cache on an SSD local to the host that is also in place. Disabling the host cache didn't immediately have an effect as far as I could see. The host MTU set to 3000 on all iSCSI interfaces for all tests. Test 1: Right after reboot, with an ixgbe MTU of 9000, the write test yields an average speed over three tests of 137MB/s. The read test yields an average over three tests of 5MB/s. Test 2: After setting ifconfig ixgbe0 mtu 3000, the write tests yield 140MB/s, and the read tests yield 53MB/s. It's important to note here that if I cut the read test short at only 2-3GB, I get results upwards of 350MB/s, which I assume is local cache-related distortion. Test 3: MTU of 1500. Read tests are up to 156 MB/s. Write tests yield
Re: [OmniOS-discuss] The ixgbe driver, Lindsay Lohan, and the Greek economy
Thank you Joerg, I've downloaded the package and will try it tomorrow. The only thing I can add at this point is that upon review of my testing, I may have performed my pkg -u between the initial quad-gig performance test and installing the 10G NIC. So this may be a new problem introduced in the latest updates. Those of you who are running 10G and have not upgraded to the latest kernel, etc, might want to do some additional testing before running the update. -Warren V On Mon, Feb 23, 2015 at 1:15 AM, Joerg Goltermann j...@osn.de wrote: Hi, I remember there was a problem with the flow control settings in the ixgbe driver, so I updated it a long time ago for our internal servers to 2.5.8. Last weekend I integrated the latest changes from the FreeBSD driver to bring the illumos ixgbe to 2.5.25 but I had no time to test it, so it's completely untested! If you would like to give the latest driver a try you can fetch the kernel modules from https://cloud.osn.de/index.php/s/Fb4so9RsNnXA7r9 Clone your boot environment, place the modules in the new environment and update the boot-archive of the new BE. - Joerg On 23.02.2015 02:54, W Verb wrote: By the way, to those of you who have working setups: please send me your pool/volume settings, interface linkprops, and any kernel tuning parameters you may have set. Thanks, Warren V On Sat, Feb 21, 2015 at 7:59 AM, Schweiss, Chip c...@innovates.com wrote: I can't say I totally agree with your performance assessment. I run Intel X520 in all my OmniOS boxes. Here is a capture of nfssvrtop I made while running many storage vMotions between two OmniOS boxes hosting NFS datastores. This is a 10 host VMware cluster. Both OmniOS boxes are dual 10G connected with copper twin-ax to the in rack Nexus 5010. VMware does 100% sync writes, I use ZeusRAM SSDs for log devices. -Chip 2014 Apr 24 08:05:51, load: 12.64, read: 17330243 KB, swrite: 15985 KB, awrite: 1875455 KB Ver Client NFSOPS Reads SWrites AWrites Commits Rd_bw SWr_bw AWr_bwRd_t SWr_t AWr_t Com_t Align% 4 10.28.17.105 0 0 0 0 0 0 0 0 0 0 0 0 0 4 10.28.17.215 0 0 0 0 0 0 0 0 0 0 0 0 0 4 10.28.17.213 0 0 0 0 0 0 0 0 0 0 0 0 0 4 10.28.16.151 0 0 0 0 0 0 0 0 0 0 0 0 0 4 all 1 0 0 0 0 0 0 0 0 0 0 0 0 3 10.28.16.175 3 0 3 0 0 1 11 04806 48 0 0 85 3 10.28.16.183 6 0 6 0 0 3 162 0 549 124 0 0 73 3 10.28.16.180 11 0 10 0 0 3 27 0 776 89 0 0 67 3 10.28.16.176 28 2 26 0 0 10 405 02572 198 0 0 100 3 10.28.16.178 46064602 4 0 0 294534 3 0 723 49 0 0 99 3 10.28.16.179 49054879 26 0 0 312208 311 0 735 271 0 0 99 3 10.28.16.181 55155502 13 0 0 352107 77 0 89 87 0 0 99 3 10.28.16.184 12095 12059 10 0 0 763014 39 0 249 147 0 0 99 3 10.28.58.1154016040 1166354 53 191605 474 202346 192 96 144 83 99 3 all 42574 33086 2176354 53 1913488 1582 202300 348 138 153 105 99 On Fri, Feb 20, 2015 at 11:46 PM, W Verb wver...@gmail.com wrote: Hello All, Thank you for your replies. I tried a few things, and found the following: 1: Disabling hyperthreading support in the BIOS drops performance overall by a factor of 4. 2: Disabling VT support also seems to have some effect, although it appears to be minor. But this has the amusing side effect of fixing the hangs I've been experiencing with fast reboot. Probably by disabling kvm. 3: The performance tests are a bit tricky to quantify because of caching effects. In fact, I'm not entirely sure what is happening here. It's just best to describe what I'm seeing: The commands I'm using to test are dd if=/dev/zero of=./test.dd bs=2M count=5000 dd of=/dev/null if=./test.dd bs=2M count=5000 The host vm is running Centos 6.6, and has the latest vmtools installed. There is a host cache on an SSD local to the host that is also in place. Disabling the host cache didn't immediately have
Re: [OmniOS-discuss] The ixgbe driver, Lindsay Lohan, and the Greek economy
By the way, to those of you who have working setups: please send me your pool/volume settings, interface linkprops, and any kernel tuning parameters you may have set. Thanks, Warren V On Sat, Feb 21, 2015 at 7:59 AM, Schweiss, Chip c...@innovates.com wrote: I can't say I totally agree with your performance assessment. I run Intel X520 in all my OmniOS boxes. Here is a capture of nfssvrtop I made while running many storage vMotions between two OmniOS boxes hosting NFS datastores. This is a 10 host VMware cluster. Both OmniOS boxes are dual 10G connected with copper twin-ax to the in rack Nexus 5010. VMware does 100% sync writes, I use ZeusRAM SSDs for log devices. -Chip 2014 Apr 24 08:05:51, load: 12.64, read: 17330243 KB, swrite: 15985KB, awrite: 1875455 KB Ver Client NFSOPS Reads SWrites AWrites Commits Rd_bw SWr_bw AWr_bwRd_t SWr_t AWr_t Com_t Align% 4 10.28.17.105 0 0 0 0 0 0 0 0 0 0 0 0 0 4 10.28.17.215 0 0 0 0 0 0 0 0 0 0 0 0 0 4 10.28.17.213 0 0 0 0 0 0 0 0 0 0 0 0 0 4 10.28.16.151 0 0 0 0 0 0 0 0 0 0 0 0 0 4 all 1 0 0 0 0 0 0 0 0 0 0 0 0 3 10.28.16.175 3 0 3 0 0 1 11 04806 48 0 0 85 3 10.28.16.183 6 0 6 0 0 3 162 0 549 124 0 0 73 3 10.28.16.180 11 0 10 0 0 3 27 0 776 89 0 0 67 3 10.28.16.176 28 2 26 0 0 10 405 02572 198 0 0 100 3 10.28.16.178 46064602 4 0 0 294534 3 0 723 49 0 0 99 3 10.28.16.179 49054879 26 0 0 312208 311 0 735 271 0 0 99 3 10.28.16.181 55155502 13 0 0 352107 77 0 89 87 0 0 99 3 10.28.16.184 12095 12059 10 0 0 763014 39 0 249 147 0 0 99 3 10.28.58.1154016040 1166354 53 191605 474 202346 192 96 144 83 99 3 all 42574 33086 2176354 53 1913488 1582 202300 348 138 153 105 99 On Fri, Feb 20, 2015 at 11:46 PM, W Verb wver...@gmail.com wrote: Hello All, Thank you for your replies. I tried a few things, and found the following: 1: Disabling hyperthreading support in the BIOS drops performance overall by a factor of 4. 2: Disabling VT support also seems to have some effect, although it appears to be minor. But this has the amusing side effect of fixing the hangs I've been experiencing with fast reboot. Probably by disabling kvm. 3: The performance tests are a bit tricky to quantify because of caching effects. In fact, I'm not entirely sure what is happening here. It's just best to describe what I'm seeing: The commands I'm using to test are dd if=/dev/zero of=./test.dd bs=2M count=5000 dd of=/dev/null if=./test.dd bs=2M count=5000 The host vm is running Centos 6.6, and has the latest vmtools installed. There is a host cache on an SSD local to the host that is also in place. Disabling the host cache didn't immediately have an effect as far as I could see. The host MTU set to 3000 on all iSCSI interfaces for all tests. Test 1: Right after reboot, with an ixgbe MTU of 9000, the write test yields an average speed over three tests of 137MB/s. The read test yields an average over three tests of 5MB/s. Test 2: After setting ifconfig ixgbe0 mtu 3000, the write tests yield 140MB/s, and the read tests yield 53MB/s. It's important to note here that if I cut the read test short at only 2-3GB, I get results upwards of 350MB/s, which I assume is local cache-related distortion. Test 3: MTU of 1500. Read tests are up to 156 MB/s. Write tests yield about 142MB/s. Test 4: MTU of 1000: Read test at 182MB/s. Test 5: MTU of 900: Read test at 130 MB/s. Test 6: MTU of 1000: Read test at 160MB/s. Write tests are now consistently at about 300MB/s. Test 7: MTU of 1200: Read test at 124MB/s. Test 8: MTU of 1000: Read test at 161MB/s. Write at 261MB/s. A few final notes: L1ARC grabs about 10GB of RAM during the tests, so there's definitely some read caching going on. The write operations are easier to observe with iostat, and I'm seeing io rates that closely correlate with the network write
Re: [OmniOS-discuss] The ixgbe driver, Lindsay Lohan, and the Greek economy
Is the issue here only related to iSCSI? We've used the X520's for NFS for a couple of years and it has worked really well for us. Not sure if this is an accurate test, but iperf shows the following results for me: Over 1GbE: [ ID] Interval Transfer Bandwidth [ 3] 0.0-10.0 sec981 MBytes823 Mbits/sec Over 10GbE on the same machines: [ ID] Interval Transfer Bandwidth [ 3] 0.0-10.0 sec 9.42 GBytes 8.09 Gbits/sec I could be going in the wrong direction here, but I was curious as well since we rely on 10G heavily. Chris On Sun, Feb 22, 2015 at 8:54 PM, W Verb wver...@gmail.com wrote: By the way, to those of you who have working setups: please send me your pool/volume settings, interface linkprops, and any kernel tuning parameters you may have set. Thanks, Warren V On Sat, Feb 21, 2015 at 7:59 AM, Schweiss, Chip c...@innovates.com wrote: I can't say I totally agree with your performance assessment. I run Intel X520 in all my OmniOS boxes. Here is a capture of nfssvrtop I made while running many storage vMotions between two OmniOS boxes hosting NFS datastores. This is a 10 host VMware cluster. Both OmniOS boxes are dual 10G connected with copper twin-ax to the in rack Nexus 5010. VMware does 100% sync writes, I use ZeusRAM SSDs for log devices. -Chip 2014 Apr 24 08:05:51, load: 12.64, read: 17330243 KB, swrite: 15985KB, awrite: 1875455 KB Ver Client NFSOPS Reads SWrites AWrites Commits Rd_bw SWr_bw AWr_bwRd_t SWr_t AWr_t Com_t Align% 4 10.28.17.105 0 0 0 0 0 0 0 0 0 0 0 0 0 4 10.28.17.215 0 0 0 0 0 0 0 0 0 0 0 0 0 4 10.28.17.213 0 0 0 0 0 0 0 0 0 0 0 0 0 4 10.28.16.151 0 0 0 0 0 0 0 0 0 0 0 0 0 4 all 1 0 0 0 0 0 0 0 0 0 0 0 0 3 10.28.16.175 3 0 3 0 0 1 11 04806 48 0 0 85 3 10.28.16.183 6 0 6 0 0 3 162 0 549 124 0 0 73 3 10.28.16.180 11 0 10 0 0 3 27 0 776 89 0 0 67 3 10.28.16.176 28 2 26 0 0 10 405 02572 198 0 0 100 3 10.28.16.178 46064602 4 0 0 294534 3 0 723 49 0 0 99 3 10.28.16.179 49054879 26 0 0 312208 311 0 735 271 0 0 99 3 10.28.16.181 55155502 13 0 0 352107 77 0 89 87 0 0 99 3 10.28.16.184 12095 12059 10 0 0 763014 39 0 249 147 0 0 99 3 10.28.58.1154016040 1166354 53 191605 474 202346 192 96 144 83 99 3 all 42574 33086 2176354 53 1913488 1582 202300 348 138 153 105 99 On Fri, Feb 20, 2015 at 11:46 PM, W Verb wver...@gmail.com wrote: Hello All, Thank you for your replies. I tried a few things, and found the following: 1: Disabling hyperthreading support in the BIOS drops performance overall by a factor of 4. 2: Disabling VT support also seems to have some effect, although it appears to be minor. But this has the amusing side effect of fixing the hangs I've been experiencing with fast reboot. Probably by disabling kvm. 3: The performance tests are a bit tricky to quantify because of caching effects. In fact, I'm not entirely sure what is happening here. It's just best to describe what I'm seeing: The commands I'm using to test are dd if=/dev/zero of=./test.dd bs=2M count=5000 dd of=/dev/null if=./test.dd bs=2M count=5000 The host vm is running Centos 6.6, and has the latest vmtools installed. There is a host cache on an SSD local to the host that is also in place. Disabling the host cache didn't immediately have an effect as far as I could see. The host MTU set to 3000 on all iSCSI interfaces for all tests. Test 1: Right after reboot, with an ixgbe MTU of 9000, the write test yields an average speed over three tests of 137MB/s. The read test yields an average over three tests of 5MB/s. Test 2: After setting ifconfig ixgbe0 mtu 3000, the write tests yield 140MB/s, and the read tests yield 53MB/s. It's important to note here that if I cut the read test short at only 2-3GB, I get results upwards of 350MB/s, which I assume is local cache-related
Re: [OmniOS-discuss] The ixgbe driver, Lindsay Lohan, and the Greek economy
I can't say I totally agree with your performance assessment. I run Intel X520 in all my OmniOS boxes. Here is a capture of nfssvrtop I made while running many storage vMotions between two OmniOS boxes hosting NFS datastores. This is a 10 host VMware cluster. Both OmniOS boxes are dual 10G connected with copper twin-ax to the in rack Nexus 5010. VMware does 100% sync writes, I use ZeusRAM SSDs for log devices. -Chip 2014 Apr 24 08:05:51, load: 12.64, read: 17330243 KB, swrite: 15985KB, awrite: 1875455 KB Ver Client NFSOPS Reads SWrites AWrites Commits Rd_bw SWr_bw AWr_bwRd_t SWr_t AWr_t Com_t Align% 4 10.28.17.105 0 0 0 0 0 0 0 0 0 0 0 0 0 4 10.28.17.215 0 0 0 0 0 0 0 0 0 0 0 0 0 4 10.28.17.213 0 0 0 0 0 0 0 0 0 0 0 0 0 4 10.28.16.151 0 0 0 0 0 0 0 0 0 0 0 0 0 4 all 1 0 0 0 0 0 0 0 0 0 0 0 0 3 10.28.16.175 3 0 3 0 0 1 11 04806 48 0 0 85 3 10.28.16.183 6 0 6 0 0 3 162 0 549 124 0 0 73 3 10.28.16.180 11 0 10 0 0 3 27 0 776 89 0 0 67 3 10.28.16.176 28 2 26 0 0 10 405 02572 198 0 0 100 3 10.28.16.178 46064602 4 0 0 294534 3 0 723 49 0 0 99 3 10.28.16.179 49054879 26 0 0 312208 311 0 735 271 0 0 99 3 10.28.16.181 55155502 13 0 0 352107 77 0 89 87 0 0 99 3 10.28.16.184 12095 12059 10 0 0 763014 39 0 249 147 0 0 99 3 10.28.58.1154016040 1166354 53 191605 474 202346 192 96 144 83 99 3 all 42574 33086 2176354 53 *1913488* 1582 202300 348 138 153 105 99 On Fri, Feb 20, 2015 at 11:46 PM, W Verb wver...@gmail.com wrote: Hello All, Thank you for your replies. I tried a few things, and found the following: 1: Disabling hyperthreading support in the BIOS drops performance overall by a factor of 4. 2: Disabling VT support also seems to have some effect, although it appears to be minor. But this has the amusing side effect of fixing the hangs I've been experiencing with fast reboot. Probably by disabling kvm. 3: The performance tests are a bit tricky to quantify because of caching effects. In fact, I'm not entirely sure what is happening here. It's just best to describe what I'm seeing: The commands I'm using to test are dd if=/dev/zero of=./test.dd bs=2M count=5000 dd of=/dev/null if=./test.dd bs=2M count=5000 The host vm is running Centos 6.6, and has the latest vmtools installed. There is a host cache on an SSD local to the host that is also in place. Disabling the host cache didn't immediately have an effect as far as I could see. The host MTU set to 3000 on all iSCSI interfaces for all tests. Test 1: Right after reboot, with an ixgbe MTU of 9000, the write test yields an average speed over three tests of 137MB/s. The read test yields an average over three tests of 5MB/s. Test 2: After setting ifconfig ixgbe0 mtu 3000, the write tests yield 140MB/s, and the read tests yield 53MB/s. It's important to note here that if I cut the read test short at only 2-3GB, I get results upwards of 350MB/s, which I assume is local cache-related distortion. Test 3: MTU of 1500. Read tests are up to 156 MB/s. Write tests yield about 142MB/s. Test 4: MTU of 1000: Read test at 182MB/s. Test 5: MTU of 900: Read test at 130 MB/s. Test 6: MTU of 1000: Read test at 160MB/s. Write tests are now consistently at about 300MB/s. Test 7: MTU of 1200: Read test at 124MB/s. Test 8: MTU of 1000: Read test at 161MB/s. Write at 261MB/s. A few final notes: L1ARC grabs about 10GB of RAM during the tests, so there's definitely some read caching going on. The write operations are easier to observe with iostat, and I'm seeing io rates that closely correlate with the network write speeds. Chris, thanks for your specific details. I'd appreciate it if you could tell me which copper NIC you tried, as well as to pass on the iSCSI tuning parameters. I've ordered an Intel EXPX9502AFXSR, which uses the 82598 chip instead of the 82599 in the X520. If I get
[OmniOS-discuss] The ixgbe driver, Lindsay Lohan, and the Greek economy
Hello all, Each of the things in the subject line are: 1: Horrendously broken 2: Have an extremely poor short-term outlook 3: Will take a huge investment of time by intelligent, dedicated, and insightful people to fix. It's common knowledge that the ixgbe driver in omnios/illumos/opensolaris is broke as hell. The point of this message is not to complain. The point is to pass on a configuration that is working for me, albeit in a screwed-up degraded fashion. I have four ESXi 5.5u2 host servers with 1 Intel PCI-e quad-gigabit NIC installed in each. Three of the gigabit ports on each client are dedicated to carry iSCSI traffic between each host and a single storage server. The storage server is based on a Supermicro X10SLM-F mainboard, which has three PCI-e slots. Two of the slots are used for storage controllers, and a single slot is used for an Intel X520 dual-port fiber 10G NIC. Previously, I had a single storage controller and two quad-gig NICs installed in the storage server, and was able to get close to line-rate on multipath iSCSI with three host clients. But when I added the fourth, I upgraded to 10G. After installation and configuration, I observed all kinds of bad behavior in the network traffic between the hosts and the server. All of this bad behavior is traced to the ixgbe driver on the storage server. Without going into the full troubleshooting process, here are my takeaways: 1: The only tuning factor that appears to have significant effect on the driver is MTU size. This applies to both the MTU of the ixgbe NIC as well as the MTU of the 1-gig NICs used in the hosts. 2: I have seen best performance with the MTU on the ixgbe set to 1000 bytes (yes, 1k). The MTU on the ESXi interfaces is set to 3000 bytes. 3: Setting 9000 byte MTUs on both sides results in about 150MB/s write speeds on a a linux vmware guest running a 10GB dd operation. But read speeds are at 5MB/s. 4: Testing of dd operations on the storage server itself shows that the filesystem is capable of performing 500MB/s+ reads and writes. 5: After setting the MTUs listed in point 2, I am able to get 270-300MB/s writes on the guest OS, and ~200MB/s reads. Not perfect, but I'll take it. 6: No /etc/system or other kernel tunings are in use. 7: Delayed ACK, Nagle, and L2 flow control tests had no effect. 8: pkg -u was performed before all tests, so I should be using the latest kernel code, etc. 9: When capturing traffic on omnios, I used the CSW distribution of tcpdump. It's worth noting that unlike EVERY ... OTHER ... IMPLEMENTATION ... of tcpdump I've ever used (BSD flavors, OSX, various linux distros, various embedded distros), libpcap doesn't appear to get individual frame reports from the omnios kernel, and so aggregates multi-frame TCP segments into a single record. This has the appearance of 20-60kB frames being transmitted by omnios when reading a packet capture with Wireshark. I cannot tell you how irritating this is when troubleshooting network issues. 10: At the wire level, the speed problems are clearly due to pauses in response time by omnios. At 9000 byte frame sizes, I see a good number of duplicate ACKs and fast retransmits during read operations (when omnios is transmitting). But below about a 4100-byte MTU on omnios (which seems to correlate to 4096-byte iSCSI block transfers), the transmission errors fade away and we only see the transmission pause problem. I'm in the process of aggregating the 10G ports and performing some IO testing with the vmware IO performance tool. That should show the performance of the 10G NIC when both physical ports are in use, and hopefully get me some more granularity on the MTU settings. If anyone has a list of kernel tuning parameters to test, I'm happy to try them out and report back. I've found a variety of suggestions online, but between Illumos, solaris, openindiana, Nexenta, opensolaris, etc, the supported variables are, um, inconsistent. -Warren V ___ OmniOS-discuss mailing list OmniOS-discuss@lists.omniti.com http://lists.omniti.com/mailman/listinfo/omnios-discuss
Re: [OmniOS-discuss] The ixgbe driver, Lindsay Lohan, and the Greek economy
You should PLEASE share your original note with the illumos developer's list. Also, please keep in mind that it is VERY possible you're seeing crosscall effects where the interrupt-servicing CPU core is not on the same PCIe bus as where you're card is plugged in. I never got a chance to perform these tests fully when I *had* ixgbe HW handy, but I observed bizarre improvements, or the disappearance of bizarre effects, if I: - disabled the HT-inspired CPUs (psradm -f HT CPUs) - Disabled one of the two CPUs (again, using psradm). You may wish to try messing around with what OS-reported CPUs are on your Romley (Xeon E5) system. I will also note that it's high time for illumos to pull in the ixgbe updates from upstream. Intel is NOT being very helpful here, partially because of fear-of-Oracle, and partially because there aren't enough illumos customers to make a dent in their HW sales. In the past, illumos developers have found the time to yank in the newest driver updates from upstream. That hasn't happened recently. For the record, OmniTI might be able to contribute here IF AND ONLY IF a sufficiently paying customer motivates us. I suspect the same answer (sufficiently paying customer) applies to engineers from any other illumos shop as well. Dan ___ OmniOS-discuss mailing list OmniOS-discuss@lists.omniti.com http://lists.omniti.com/mailman/listinfo/omnios-discuss
Re: [OmniOS-discuss] The ixgbe driver, Lindsay Lohan, and the Greek economy
On Feb 20, 2015, at 9:10 PM, Dan McDonald dan...@omniti.com wrote: I never got a chance to perform these tests fully when I *had* ixgbe HW handy, but I observed bizarre improvements, or the disappearance of bizarre effects, if I: - disabled the HT-inspired CPUs (psradm -f HT CPUs) - Disabled one of the two CPUs (again, using psradm). You may wish to try messing around with what OS-reported CPUs are on your Romley (Xeon E5) system. To see the layouts, the psrinfo(1M) command is your friend, especially psrinfo -vp. Dan ___ OmniOS-discuss mailing list OmniOS-discuss@lists.omniti.com http://lists.omniti.com/mailman/listinfo/omnios-discuss
Re: [OmniOS-discuss] The ixgbe driver, Lindsay Lohan, and the Greek economy
After installation and configuration, I observed all kinds of bad behavior in the network traffic between the hosts and the server. All of this bad behavior is traced to the ixgbe driver on the storage server. Without going into the full troubleshooting process, here are my takeaways: [...] For what it's worth, we managed to achieve much better line rates on copper 10G ixgbe hardware of various descriptions between OmniOS and CentOS 7 (I don't think we ever tested OmniOS to OmniOS). I don't believe OmniOS could do TCP at full line rate but I think we managed 700+ Mbytes/sec on both transmit and receive and we got basically disk-limited speeds with iSCSI (across multiple disks on multi-disk mirrored pools, OmniOS iSCSI initiator, Linux iSCSI targets). I don't believe we did any specific kernel tuning (and in fact some of our attempts to fiddle ixgbe driver parameters blew up in our face). We did tune iSCSI connection parameters to increase various buffer sizes so that ZFS could do even large single operations in single iSCSI transactions. (More details available if people are interested.) 10: At the wire level, the speed problems are clearly due to pauses in response time by omnios. At 9000 byte frame sizes, I see a good number of duplicate ACKs and fast retransmits during read operations (when omnios is transmitting). But below about a 4100-byte MTU on omnios (which seems to correlate to 4096-byte iSCSI block transfers), the transmission errors fade away and we only see the transmission pause problem. This is what really attracted my attention. In our OmniOS setup, our specific Intel hardware had ixgbe driver issues that could cause activity stalls during once-a-second link heartbeat checks. This obviously had an effect at the TCP and iSCSI layers. My initial message to illumos-developer sparked a potentially interesting discussion: http://www.listbox.com/member/archive/182179/2014/10/sort/time_rev/page/16/entry/6:405/20141003125035:6357079A-4B1D-11E4-A39C-D534381BA44D/ If you think this is a possibility in your setup, I've put the DTrace script I used to hunt for this up on the web: http://www.cs.toronto.edu/~cks/src/omnios-ixgbe/ixgbe_delay.d This isn't the only potential source of driver stalls by any means, it's just the one I found. You may also want to look at lockstat in general, as information it reported is what led us to look specifically at the ixgbe code here. (If you suspect kernel/driver issues, lockstat combined with kernel source is a really excellent resource.) - cks ___ OmniOS-discuss mailing list OmniOS-discuss@lists.omniti.com http://lists.omniti.com/mailman/listinfo/omnios-discuss
Re: [OmniOS-discuss] The ixgbe driver, Lindsay Lohan, and the Greek economy
Hello All, Thank you for your replies. I tried a few things, and found the following: 1: Disabling hyperthreading support in the BIOS drops performance overall by a factor of 4. 2: Disabling VT support also seems to have some effect, although it appears to be minor. But this has the amusing side effect of fixing the hangs I've been experiencing with fast reboot. Probably by disabling kvm. 3: The performance tests are a bit tricky to quantify because of caching effects. In fact, I'm not entirely sure what is happening here. It's just best to describe what I'm seeing: The commands I'm using to test are dd if=/dev/zero of=./test.dd bs=2M count=5000 dd of=/dev/null if=./test.dd bs=2M count=5000 The host vm is running Centos 6.6, and has the latest vmtools installed. There is a host cache on an SSD local to the host that is also in place. Disabling the host cache didn't immediately have an effect as far as I could see. The host MTU set to 3000 on all iSCSI interfaces for all tests. Test 1: Right after reboot, with an ixgbe MTU of 9000, the write test yields an average speed over three tests of 137MB/s. The read test yields an average over three tests of 5MB/s. Test 2: After setting ifconfig ixgbe0 mtu 3000, the write tests yield 140MB/s, and the read tests yield 53MB/s. It's important to note here that if I cut the read test short at only 2-3GB, I get results upwards of 350MB/s, which I assume is local cache-related distortion. Test 3: MTU of 1500. Read tests are up to 156 MB/s. Write tests yield about 142MB/s. Test 4: MTU of 1000: Read test at 182MB/s. Test 5: MTU of 900: Read test at 130 MB/s. Test 6: MTU of 1000: Read test at 160MB/s. Write tests are now consistently at about 300MB/s. Test 7: MTU of 1200: Read test at 124MB/s. Test 8: MTU of 1000: Read test at 161MB/s. Write at 261MB/s. A few final notes: L1ARC grabs about 10GB of RAM during the tests, so there's definitely some read caching going on. The write operations are easier to observe with iostat, and I'm seeing io rates that closely correlate with the network write speeds. Chris, thanks for your specific details. I'd appreciate it if you could tell me which copper NIC you tried, as well as to pass on the iSCSI tuning parameters. I've ordered an Intel EXPX9502AFXSR, which uses the 82598 chip instead of the 82599 in the X520. If I get similar results with my fiber transcievers, I'll see if I can get a hold of copper ones. But I should mention that I did indeed look at PHY/MAC error rates, and they are nil. -Warren V On Fri, Feb 20, 2015 at 7:25 PM, Chris Siebenmann c...@cs.toronto.edu wrote: After installation and configuration, I observed all kinds of bad behavior in the network traffic between the hosts and the server. All of this bad behavior is traced to the ixgbe driver on the storage server. Without going into the full troubleshooting process, here are my takeaways: [...] For what it's worth, we managed to achieve much better line rates on copper 10G ixgbe hardware of various descriptions between OmniOS and CentOS 7 (I don't think we ever tested OmniOS to OmniOS). I don't believe OmniOS could do TCP at full line rate but I think we managed 700+ Mbytes/sec on both transmit and receive and we got basically disk-limited speeds with iSCSI (across multiple disks on multi-disk mirrored pools, OmniOS iSCSI initiator, Linux iSCSI targets). I don't believe we did any specific kernel tuning (and in fact some of our attempts to fiddle ixgbe driver parameters blew up in our face). We did tune iSCSI connection parameters to increase various buffer sizes so that ZFS could do even large single operations in single iSCSI transactions. (More details available if people are interested.) 10: At the wire level, the speed problems are clearly due to pauses in response time by omnios. At 9000 byte frame sizes, I see a good number of duplicate ACKs and fast retransmits during read operations (when omnios is transmitting). But below about a 4100-byte MTU on omnios (which seems to correlate to 4096-byte iSCSI block transfers), the transmission errors fade away and we only see the transmission pause problem. This is what really attracted my attention. In our OmniOS setup, our specific Intel hardware had ixgbe driver issues that could cause activity stalls during once-a-second link heartbeat checks. This obviously had an effect at the TCP and iSCSI layers. My initial message to illumos-developer sparked a potentially interesting discussion: http://www.listbox.com/member/archive/182179/2014/10/sort/time_rev/page/16/entry/6:405/20141003125035:6357079A-4B1D-11E4-A39C-D534381BA44D/ If you think this is a possibility in your setup, I've put the DTrace script I used to hunt for this up on the web: http://www.cs.toronto.edu/~cks/src/omnios-ixgbe/ixgbe_delay.d This isn't the only potential source of driver stalls by any means, it's just the one I found. You may also want to look at lockstat in