Re: ftp vs. nfs install times
In message [EMAIL PROTECTED], John Hay write s: I've tested last nights make release built install via both ftp and nfs and am seeing some rather strange results timeing wise: A full install (ie: select ALL) w/ ports. NFS: about 18 minutes. (ave. about 1000KB/sec) FTP: about 70 minutes. (ave. about 45KB/sec) Maybe just as a datapoint. My -current snap building machine is running a kernel of Oct 24 and I noticed this morning that it is taking a very long time to scp the snap to internat. Try disabling newreno in both ends: sysctl -w net.inet.tcp.newreno=0 On my laptop with Wavelan cards this increases TCP throughput by a factor of 5. -- Poul-Henning Kamp | UNIX since Zilog Zeus 3.20 [EMAIL PROTECTED] | TCP/IP since RFC 956 FreeBSD committer | BSD since 4.3-tahoe Never attribute to malice what can adequately be explained by incompetence. To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-current" in the body of the message
Re: ftp vs. nfs install times
I've tested last nights make release built install via both ftp and nfs and am seeing some rather strange results timeing wise: A full install (ie: select ALL) w/ ports. NFS: about 18 minutes. (ave. about 1000KB/sec) FTP: about 70 minutes. (ave. about 45KB/sec) Maybe just as a datapoint. My -current snap building machine is running a kernel of Oct 24 and I noticed this morning that it is taking a very long time to scp the snap to internat. Try disabling newreno in both ends: sysctl -w net.inet.tcp.newreno=0 On my laptop with Wavelan cards this increases TCP throughput by a factor of 5. Yes, that makes a HUGE difference. I only did it on the current box. Internat is still running 4.x and don't have it. I think it made more than a factor of 5 difference here. :-) John -- John Hay -- [EMAIL PROTECTED] To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-current" in the body of the message
Re: ftp vs. nfs install times
In message [EMAIL PROTECTED], John Hay write s: Try disabling newreno in both ends: sysctl -w net.inet.tcp.newreno=0 On my laptop with Wavelan cards this increases TCP throughput by a factor of 5. Yes, that makes a HUGE difference. I only did it on the current box. Internat is still running 4.x and don't have it. I think it made more than a factor of 5 difference here. :-) I sent packet traces to yan some time back, but have not heard from him since. I wonder if we should disable newrene for now... -- Poul-Henning Kamp | UNIX since Zilog Zeus 3.20 [EMAIL PROTECTED] | TCP/IP since RFC 956 FreeBSD committer | BSD since 4.3-tahoe Never attribute to malice what can adequately be explained by incompetence. To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-current" in the body of the message
Re: ftp vs. nfs install times
Poul-Henning Kamp writes: In message [EMAIL PROTECTED], John Hay write s: Try disabling newreno in both ends: sysctl -w net.inet.tcp.newreno=0 On my laptop with Wavelan cards this increases TCP throughput by a factor of 5. Yes, that makes a HUGE difference. I only did it on the current box. Internat is still running 4.x and don't have it. I think it made more than a factor of 5 difference here. :-) I sent packet traces to yan some time back, but have not heard from him since. I wonder if we should disable newrene for now... That would be a shame. It makes a huge difference when shovling bits across my DSL line to my -stable laptop. Running netperf from a -current alpha with newreno, I consistantly see 650-700Kb/s. From a -stable i386 without newreno, I see variable results from 300-500Kb/sec with large stutters, so to speak, where there is no traffic at all for one second (or at least I don't see any packets in systat -tcp 1). I hope this problem can be fixed -- I was actually hoping that newreno could be MFC'ed. Drew -- Andrew Gallatin, Sr Systems Programmer http://www.cs.duke.edu/~gallatin Duke University Email: [EMAIL PROTECTED] Department of Computer Science Phone: (919) 660-6590 To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-current" in the body of the message
ftp vs. nfs install times
Hi, I've tested last nights make release built install via both ftp and nfs and am seeing some rather strange results timeing wise: A full install (ie: select ALL) w/ ports. NFS: about 18 minutes. (ave. about 1000KB/sec) FTP: about 70 minutes. (ave. about 45KB/sec) on the same box after the install, I can ftp to the server and mget all the files in just a few moments. ie: The snap server I'm using isn't the problem. Any ideas of what the best way to debug this from the holographic shell might be? Some tools are available, others are available but don't work (like top). While this is happenning, the idle process is accumulating time almost lockstep with walltime. Ideas welcome. John To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-current" in the body of the message
Re: ftp vs. nfs install times
Hi, I've tested last nights make release built install via both ftp and nfs and am seeing some rather strange results timeing wise: A full install (ie: select ALL) w/ ports. NFS: about 18 minutes. (ave. about 1000KB/sec) FTP: about 70 minutes. (ave. about 45KB/sec) on the same box after the install, I can ftp to the server and mget all the files in just a few moments. ie: The snap server I'm using isn't the problem. Maybe just as a datapoint. My -current snap building machine is running a kernel of Oct 24 and I noticed this morning that it is taking a very long time to scp the snap to internat. Normally it takes a few minutes but it is now more than a hour and it isn't halfway yet. The machine is almost totally idle. The machine is running an SMP kernel if it matters. John -- John Hay -- [EMAIL PROTECTED] To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-current" in the body of the message