daily CVS update output
Updating src tree: P src/distrib/sets/lists/tests/mi P src/share/misc/acronyms.comp P src/sys/arch/arm/nvidia/tegra210_xusbpad.c P src/sys/arch/macppc/stand/fixcoff/elf32_powerpc_merge.x P src/sys/dev/i2c/adm1021.c P src/tests/net/route/Makefile U src/tests/net/route/t_rtcache.sh P src/usr.bin/stat/stat.1 P src/usr.bin/stat/stat.c P src/usr.bin/sys_info/sys_info.1 P src/usr.bin/sys_info/sys_info.sh Updating xsrc tree: Killing core files: Updating file list: -rw-rw-r-- 1 srcmastr netbsd 49145655 Sep 21 03:05 ls-lRA.gz
Re: Automated report: NetBSD-current/i386 test failure
Date:Wed, 20 Sep 2017 21:06:51 + (UTC) From:NetBSD Test FixtureMessage-ID: <150594161124.25268.1441500354190307...@babylon5.netbsd.org> | This is an automatically generated notice of new failures of the | NetBSD test suite. | | The newly failing test cases are: | | bin/dd/t_dd:seek | fs/tmpfs/t_create:attrs These should be fixed (fallout from stat breakage.) kre
Re: ssh, HPN extension and TCP auto-tuning
On Wed, 20 Sep 2017, Havard Eidnes wrote: the OpenSSH in NetBSD has for quite a while had the "high- performance networking" patches applied. I wasn't aware of this. That's good to know. I've often wondered what patch sets we apply to NetBSD's SSH implementation. However, when you copy "in the other direction", i.e. when the remote sshd is the one which is pushing the file across the network, we get an average of 8.4MB/s when copying a 143MB large file, What is on the other side? NetBSD or something else? From the context, I'm assuming "something else" and that the something-else has the HPN patches also. and a tcpdump + tcptrace reveals that in this case the system's automatic tuning of the TCP window is indeed kicking into action. This could be the result of some kind of router issue. Perhaps the side that's not working is the victim of some kind of misconfiguration. Some possibilities are: * One of the routers/firewalls is doing some reassembly or "cleansing" the TCP frames and preventing something like window scaling. * Path MTU discovery is failing because it's blocking the ICMP probes. So, your MTU is artificially smaller in one direction. * The connection on one side is being proxied in such a way that doesn't allow the optimizations to kick in. I'm sure there are a few I'm not thinking of, too. send space, performance improves, but again, TCP auto-tuning does not appear to be kicking in. Am I alone in seeing this? I have to say that I've had a lot of weird issues with auto-tuning on NetBSD but I've never been had the right case to submit a PR over it. It's just... squirrly. I can't give enough details to be helpful, but no, you aren't alone in seeing it. I've noticed similar issues, but never narrowed it down enough to blame NetBSD. -Swift
Automated report: NetBSD-current/i386 test failure
This is an automatically generated notice of new failures of the NetBSD test suite. The newly failing test cases are: bin/dd/t_dd:seek fs/tmpfs/t_create:attrs fs/tmpfs/t_devices:basic fs/tmpfs/t_link:basic fs/tmpfs/t_link:subdirs fs/tmpfs/t_mkdir:attrs fs/tmpfs/t_mkdir:many fs/tmpfs/t_mkdir:single fs/tmpfs/t_mknod:block fs/tmpfs/t_mknod:char fs/tmpfs/t_mknod:pipe fs/tmpfs/t_mount:attrs fs/tmpfs/t_rename:dotdot fs/tmpfs/t_rmdir:links fs/tmpfs/t_rmdir:single fs/tmpfs/t_setattr:chgrp fs/tmpfs/t_setattr:chown fs/tmpfs/t_setattr:chowngrp fs/tmpfs/t_setattr:chtimes fs/tmpfs/t_sizes:overwrite fs/tmpfs/t_times:empty fs/tmpfs/t_times:link fs/tmpfs/t_times:non_empty fs/tmpfs/t_times:rename The above tests failed in each of the last 3 test runs, and passed in at least 27 consecutive runs before that. The following commits were made between the last successful test and the failed test: 2017.09.19.20.45.09 jmcneill src/sys/arch/arm/nvidia/tegra210_car.c,v 1.2 2017.09.19.20.46.12 jmcneill src/sys/arch/arm/nvidia/Attic/tegra_xusbpadreg.h,v 1.3 2017.09.19.20.46.12 jmcneill src/sys/arch/arm/nvidia/files.tegra,v 1.42 2017.09.19.20.46.12 jmcneill src/sys/arch/arm/nvidia/tegra124_xusbpad.c,v 1.1 2017.09.19.20.46.12 jmcneill src/sys/arch/arm/nvidia/tegra124_xusbpadreg.h,v 1.1 2017.09.19.20.46.12 jmcneill src/sys/arch/arm/nvidia/tegra_ahcisata.c,v 1.11 2017.09.19.20.46.12 jmcneill src/sys/arch/arm/nvidia/tegra_var.h,v 1.40 2017.09.19.20.46.12 jmcneill src/sys/arch/arm/nvidia/tegra_xusb-fw.mk,v 1.2 2017.09.19.20.46.12 jmcneill src/sys/arch/arm/nvidia/tegra_xusb.c,v 1.7 2017.09.19.20.46.12 jmcneill src/sys/arch/arm/nvidia/tegra_xusbpad.c,v 1.6 2017.09.19.20.46.12 jmcneill src/sys/arch/arm/nvidia/tegra_xusbpad.h,v 1.1 2017.09.19.20.46.12 jmcneill src/sys/arch/evbarm/conf/TEGRA,v 1.28 2017.09.19.21.45.28 christos src/usr.bin/stat/stat.1,v 1.39 2017.09.19.21.45.28 christos src/usr.bin/stat/stat.c,v 1.39 Log files can be found at: http://releng.NetBSD.org/b5reports/i386/commits-2017.09.html#2017.09.19.21.45.28
Re: ssh, HPN extension and TCP auto-tuning
Hello. I spent quite a bit of time looking at this under NetBSD-5 and discovered that the default ssh settings, along with the default tcp network settings precluded the adaptive network performance from working. As a result, I've added the following lines to the ssh configs as well as the sysctl.conf files on the NetBSD-5 hosts we manage. As faras I've been able to tell, we've been able to realize good performance gains as a result of these changes. Unless there have been a lot of regressions, I have no reason to believe that these settings won't yield similar performance improvements under NetBSD-7 and NetBSD-8. Here are the fixes I came up with. # Improves TCP performance significantly with ssh. net.inet.tcp.recvbuf_auto=1 net.inet.tcp.sendbuf_auto=1 net.inet.tcp.sendbuf_max=16777216 net.inet.tcp.recvbuf_max=16777216 # Put the following lines in both /etc/ssh/ssh_config # and /etc/ssh/sshd_config #Enable High Performance Networking options (BB 12/27/2010) #Turn on HPN features HPNDisabled no #Allow 5MB of ssh window buffer HPNBufferSize 5000 #Enable dynamic window sizing of SSH buffers #You must have tcp autotuning turned on in the kernel for this to work TcpRcvBufPoll yes
ssh, HPN extension and TCP auto-tuning
Hi, the OpenSSH in NetBSD has for quite a while had the "high- performance networking" patches applied. However, despite this, we are observing rather low performance when copying files over a distance, e.g. we have a pair of hosts running netbsd-7 code, placed some 14-15ms apart, where scp'ing a file only manages to give around 2.6MB/s. Doing a tcpdump and an analysis using tcptrace + looking at the result with xplot reveals that the TCP window never climbs above the default 32KB size. This is when the scp client is pushing the file to the remote server. However, when you copy "in the other direction", i.e. when the remote sshd is the one which is pushing the file across the network, we get an average of 8.4MB/s when copying a 143MB large file, and a tcpdump + tcptrace reveals that in this case the system's automatic tuning of the TCP window is indeed kicking into action. The same behaviour can be observed with the scp client from 8.0_BETA: pushing with scp is slow, pulling with scp from the remote server is quite a bit faster. I'm going to guess that "pushing with scp" is the most often used mode, as you may get file name completion in that case... If, on the other hand, I bump the recvspace and sendspace on the two involved hosts, so that the scp client has a larger default send space, performance improves, but again, TCP auto-tuning does not appear to be kicking in. Am I alone in seeing this? I must say I'm puzzled by the result. The configuration on both systems are pretty much "stock", and the network is not the bottleneck in my case. Admittedly, the OpenSSH in netbsd-7 is quite old, and the HPN patches are probably of the same vintage, and I've not checked if a newer combination on that front will improve matters; I may do that next. Regards, - HÃ¥vard