Re: ftp and mail much slower into fbsd 4.4 vs and old BSDi
In article [EMAIL PROTECTED], Len Conrad [EMAIL PROTECTED] wrote: Sorry, hackers, I posted this twice in -questions and got no response. [...] FreeBSD 4.4-RELEASE #0 FreeBSD-4.4 had terrible performance bugs in the TCP stack, with or without newreno. Matt Dillon fixed them -- I believe before 4.5 was released. There have been recent reports that there are still problems when newreno is enabled. So your best bet is to update at least to 4.5-RELEASE and turn newreno off. John -- John Polstra John D. Polstra Co., Inc.Seattle, Washington USA Disappointment is a good sign of basic intelligence. -- Chögyam Trungpa To Unsubscribe: send mail to [EMAIL PROTECTED] with unsubscribe freebsd-hackers in the body of the message
Re: ftp and mail much slower into fbsd 4.4 vs and old BSDi
In article [EMAIL PROTECTED], [EMAIL PROTECTED] writes: Len Conrad wrote: Sorry, hackers, I posted this twice in -questions and got no response. If the problem is newreno, can somebody say how to up just that piece for 4.4 so as to be as non-disruptive, non-dice-rolling as possible on this otherwise solid machine? You can turn off newreno with the sysctl. I put this in /etc/sysctl.conf on my machines: net.inet.tcp.newreno=0 Could you explain what newreno is, in a nutshell, the upshots of using it, and what the ramifications of turning it off are? I'm running FreeBSD 4.5-RELEASE-p7, and it's on by default. Thanks, Dave -- Windows: Where do you want to go today? Linux: Where do you want to go tomorrow? FreeBSD: Are you guys coming, or what? To Unsubscribe: send mail to [EMAIL PROTECTED] with unsubscribe freebsd-hackers in the body of the message
Re: ftp and mail much slower into fbsd 4.4 vs and old BSDi
On Monday, July 1, 2002, at 12:00 , D J Hawkey Jr wrote: Could you explain what newreno is, in a nutshell, the upshots of using it, and what the ramifications of turning it off are? http://www.google.com/search?q=tcp+new+reno To Unsubscribe: send mail to [EMAIL PROTECTED] with unsubscribe freebsd-hackers in the body of the message
Re: ftp and mail much slower into fbsd 4.4 vs and old BSDi
D J Hawkey Jr wrote: You can turn off newreno with the sysctl. I put this in /etc/sysctl.conf on my machines: net.inet.tcp.newreno=0 Could you explain what newreno is, in a nutshell, the upshots of using it, and what the ramifications of turning it off are? I'm running FreeBSD 4.5-RELEASE-p7, and it's on by default. If you turn it off, you rat all your firends out in the Prisoner's Dilemma. Congestion control protocols only work if everyone participates. All Reno TCP implementations include TCP Fast Retransmit and Fast Recovery algorithms [RFC2581]. Fast retransmit relies on three duplicate acknowledgements to trigger the retransmission of a single lost segment. Once the Fast Retransmit has occurred, TCP then waits for enough additional duplicate ACKs to arrive, indicating that half of the data in flight has left the network. Only when this has occurred will TCP send additional new data. The consequence of this delay is that the entire new window of data is transmitted in one half of one Round Trip Time (RTT). This burst can cause repeated bursts in successive RTTs following the recovery, which can result in overall additional burstiness on the network. Hoe [Hoe95] suggested that during Fast Recovery the TCP data sender space out retransmissions and new data on alternate acknowledgements across the entire recovery RTT. (Note that this eliminates the half RTT lull in sending which occurs in Reno TCP.) ...THat's the PSC nutshell description of NewReno. -- Terry To Unsubscribe: send mail to [EMAIL PROTECTED] with unsubscribe freebsd-hackers in the body of the message
Re: ftp and mail much slower into fbsd 4.4 vs and old BSDi
On Jul 01, at 09:56 AM, Terry Lambert wrote: D J Hawkey Jr wrote: You can turn off newreno with the sysctl. I put this in /etc/sysctl.conf on my machines: net.inet.tcp.newreno=0 Could you explain what newreno is, in a nutshell, the upshots of using it, and what the ramifications of turning it off are? I'm running FreeBSD 4.5-RELEASE-p7, and it's on by default. If you turn it off, you rat all your firends out in the Prisoner's Dilemma. Congestion control protocols only work if everyone participates. All Reno TCP implementations include TCP Fast Retransmit and Fast Recovery algorithms [RFC2581]... [SNIP] ...THat's the PSC nutshell description of NewReno. OK. Thanks, Terry. I'm guessing that the suggestion to turn it off (the original thread) is valid enough, as all machines referred to were on the same network, else why wouldn't the above still hold true? Or is it a matter of FTP and/or SMTP (specifically, sendmail) not playing nice with newreno? -- Terry Dave -- __ __ \__ \D. J. HAWKEY JR. / __/ \/\ [EMAIL PROTECTED]/\/ http://www.visi.com/~hawkeyd/ To Unsubscribe: send mail to [EMAIL PROTECTED] with unsubscribe freebsd-hackers in the body of the message
Re: ftp and mail much slower into fbsd 4.4 vs and old BSDi
D J Hawkey Jr wrote: If you turn it off, you rat all your firends out in the Prisoner's Dilemma. Congestion control protocols only work if everyone participates. All Reno TCP implementations include TCP Fast Retransmit and Fast Recovery algorithms [RFC2581]... [SNIP] ...THat's the PSC nutshell description of NewReno. OK. Thanks, Terry. I'm guessing that the suggestion to turn it off (the original thread) is valid enough, as all machines referred to were on the same network, else why wouldn't the above still hold true? Or is it a matter of FTP and/or SMTP (specifically, sendmail) not playing nice with newreno? No. Disabling congestion control is a bad thing. There are some known bugs in the FreeBSD implementation; see the Matt Dillon postings on the subject; he posted a couple of patches that would work around (but not fix) the problems with the FreeBSD code. I thought they had been checked in, so as long as you are running a recent release, it should not be an issue. If you are hitting the problem, then you are hitting congestion, or it would not be triggering. Butting in front of other people in line at the movie theater seems like a good idea: you get your tickets and popcorn faster, and you always get a nice seat. Until everyone else starts to do the same thing. Then it becomes a fist-fight. I personally like TCP Rate Halving (Hoe, et. al.) as a congestion solution: half-rate ACK-based self-pacing during congestion. The patches for NetBSD are up on the CMU PSC site, and are easily portable to FreeBSD. In any case, you need to read Dillon's postings and the other references (particularly, RFC-2581) so that you understand what it is that you are turning off. Search for Reno in the FreeBSD-current archives to find the thread in which the comments were posted. -- Terry To Unsubscribe: send mail to [EMAIL PROTECTED] with unsubscribe freebsd-hackers in the body of the message
Re: ftp and mail much slower into fbsd 4.4 vs and old BSDi
D J Hawkey Jr wrote: I'm guessing that the suggestion to turn it off (the original thread) is valid enough, as all machines referred to were on the same network, else why wouldn't the above still hold true? Or is it a matter of FTP and/or SMTP (specifically, sendmail) not playing nice with newreno? The problem is that Terry has described the theory, whereas many of us who have observed the situation in the real world have noticed that even on a homogenous network (all with newreno enabled) performance is still worse than with newreno disabled. I don't have enough network stack fu to debug or improve this, but I have enough experience with it to know that off is better. For you, I'd say turning it off is the first thing you should test, and see what happens. Doug To Unsubscribe: send mail to [EMAIL PROTECTED] with unsubscribe freebsd-hackers in the body of the message
Re: ftp and mail much slower into fbsd 4.4 vs and old BSDi
On Mon, 1 Jul 2002, Doug Barton wrote: D J Hawkey Jr wrote: I'm guessing that the suggestion to turn it off (the original thread) is valid enough, as all machines referred to were on the same network, else why wouldn't the above still hold true? Or is it a matter of FTP and/or SMTP (specifically, sendmail) not playing nice with newreno? The problem is that Terry has described the theory, whereas many of us who have observed the situation in the real world have noticed that even on a homogenous network (all with newreno enabled) performance is still worse than with newreno disabled. I don't have enough network stack fu to debug or improve this, but I have enough experience with it to know that off is better. For you, I'd say turning it off is the first thing you should test, and see what happens. Doug I've been meaning to investigate, but I keep getting sidetracked... But yeah, I agree with you. Going back to plain reno isn't the end of the world; it's still behaves decently and won't cause problems. Mike Silby Silbersack To Unsubscribe: send mail to [EMAIL PROTECTED] with unsubscribe freebsd-hackers in the body of the message
Re: ftp and mail much slower into fbsd 4.4 vs and old BSDi
Mike Silbersack wrote: On Mon, 1 Jul 2002, Doug Barton wrote: The problem is that Terry has described the theory, whereas many of us who have observed the situation in the real world have noticed that even on a homogenous network (all with newreno enabled) performance is still worse than with newreno disabled. I guess you missed the part where I said that FreeBSD had bugs, and Matt Dillon posted patches? You guys are acting like this is something new that was recently discovered, like the recent SlashDot story on initial sequence numbers, which was first published a year and a quarter ago, but is somehow now magically, once again, news. I don't have enough network stack fu to debug or improve this, but I have enough experience with it to know that off is better. For you, I'd say turning it off is the first thing you should test, and see what happens. I've been meaning to investigate, but I keep getting sidetracked... But yeah, I agree with you. Going back to plain reno isn't the end of the world; it's still behaves decently and won't cause problems. Off has higher performance in the absence of congestion. If you can guarantee no congestion, I'd like to purchase network connectivity from you. I guess I will have to repeat Matt Dillon's postings verbatim to make the points that Matt, Poul, and others made with regard to NewReno and the FreeBSD failure case, as well as the workaround? Off is not the answer; fix it is the answer. If you can't do that, then use Matt's workaround is the answer. FWIW, Matt's workaround turns off NewReno for specific instances where it's known to fail because of trigerring on an incorrect one packet boundary, but it doesn't turn it off completely. This whole problem comes up every month or two, and has been known since NewReno was integrated, and has been characterized since Matt first looked at it in depth and provided patches. If you are running an old FreeBSD, then manually apply Matt's patches from the previous threads where this issue was discussed to death. -- Terry To Unsubscribe: send mail to [EMAIL PROTECTED] with unsubscribe freebsd-hackers in the body of the message
Re: ftp and mail much slower into fbsd 4.4 vs and old BSDi
On Mon, 1 Jul 2002, Terry Lambert wrote: Mike Silbersack wrote: On Mon, 1 Jul 2002, Doug Barton wrote: The problem is that Terry has described the theory, whereas many of us who have observed the situation in the real world have noticed that even on a homogenous network (all with newreno enabled) performance is still worse than with newreno disabled. I guess you missed the part where I said that FreeBSD had bugs, and Matt Dillon posted patches? Nope. I think you missed the part where I said I was talking about reality, not theory. :) The reality is, it's broken now, and in my experience, turning it off makes the system work better. I agree that it should be fixed in the long term, but it ain't gonna be me that fixes it. Doug -- We have known freedom's price. We have shown freedom's power. And in this great conflict, ... we will see freedom's victory. - George W. Bush, President of the United States State of the Union, January 28, 2002 Do YOU Yahoo!? To Unsubscribe: send mail to [EMAIL PROTECTED] with unsubscribe freebsd-hackers in the body of the message
Re: ftp and mail much slower into fbsd 4.4 vs and old BSDi
In message [EMAIL PROTECTED], Doug Barton writ es: On Mon, 1 Jul 2002, Terry Lambert wrote: Mike Silbersack wrote: On Mon, 1 Jul 2002, Doug Barton wrote: The problem is that Terry has described the theory, whereas many of us who have observed the situation in the real world have noticed that even on a homogenous network (all with newreno enabled) performance is still worse than with newreno disabled. I guess you missed the part where I said that FreeBSD had bugs, and Matt Dillon posted patches? Nope. I think you missed the part where I said I was talking about reality, not theory. :) The reality is, it's broken now, and in my experience, turning it off makes the system work better. Yes, I can attest to this an I belive it is actually the case on both -current and -releng4 that disabling newreno improves TCP performance. I belive running an X11 application or scp(1) over a wavelan is a very good test-bed for this issue. -- 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-hackers in the body of the message
Re: ftp and mail much slower into fbsd 4.4 vs and old BSDi
Poul-Henning Kamp wrote: Yes, I can attest to this an I belive it is actually the case on both -current and -releng4 that disabling newreno improves TCP performance. I belive running an X11 application or scp(1) over a wavelan is a very good test-bed for this issue. Wireless breaks a lot of optimizations, doesn't it? Congestion control assumes that packet loss is due to congestion, and less than 1% of loss is due to damage -- quite the opposite of 802.11(b) in an urban environment -- cordless phones, microwaves, etc. To Unsubscribe: send mail to [EMAIL PROTECTED] with unsubscribe freebsd-hackers in the body of the message
Re: ftp and mail much slower into fbsd 4.4 vs and old BSDi
On Jul 01, at 01:50 PM, Doug Barton wrote: On Mon, 1 Jul 2002, Terry Lambert wrote: Mike Silbersack wrote: On Mon, 1 Jul 2002, Doug Barton wrote: The problem is that Terry has described the theory, whereas many of us who have observed the situation in the real world have noticed that even on a homogenous network (all with newreno enabled) performance is still worse than with newreno disabled. I guess you missed the part where I said that FreeBSD had bugs, and Matt Dillon posted patches? Nope. I think you missed the part where I said I was talking about reality, not theory. :) The reality is, it's broken now, and in my experience, turning it off makes the system work better. I agree that it should be fixed in the long term, but it ain't gonna be me that fixes it. Guys, guys... Take 'er easy. All I asked is what it is, what it does, and the ramifications. Thank you both for enlightening me. Really. No sarcasm. If none of the three of us are going to fix it, but we all have opinions on how to deal with with, let's just be tolerant of those opinions, and move on, OK? I have yet to determine if 4.5-REL-p7 has the patches Terry speaks of, but I'll look into it, and chart my own course, just as the two of you have. Doug Thanks, Dave -- __ __ \__ \D. J. HAWKEY JR. / __/ \/\ [EMAIL PROTECTED]/\/ http://www.visi.com/~hawkeyd/ To Unsubscribe: send mail to [EMAIL PROTECTED] with unsubscribe freebsd-hackers in the body of the message
Re: ftp and mail much slower into fbsd 4.4 vs and old BSDi
In message [EMAIL PROTECTED], Michael Sierchio writes: Poul-Henning Kamp wrote: Yes, I can attest to this an I belive it is actually the case on both -current and -releng4 that disabling newreno improves TCP performance. I belive running an X11 application or scp(1) over a wavelan is a very good test-bed for this issue. Wireless breaks a lot of optimizations, doesn't it? Congestion control assumes that packet loss is due to congestion, and less than 1% of loss is due to damage -- quite the opposite of 802.11(b) in an urban environment -- cordless phones, microwaves, etc. newreno is not sold as being a significant pessimization in some cases. -- 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-hackers in the body of the message
Re: ftp and mail much slower into fbsd 4.4 vs and old BSDi
On Mon, 1 Jul 2002, D J Hawkey Jr wrote: Guys, guys... Take 'er easy. All I asked is what it is, what it does, and the ramifications. Thank you both for enlightening me. Really. No sarcasm. I actually wasn't being sarcastic... You've wandered into a briar patch of long standing though, so you might get scratched... such is life. :) If none of the three of us are going to fix it, but we all have opinions on how to deal with with, let's just be tolerant of those opinions, and move on, OK? I have yet to determine if 4.5-REL-p7 has the patches Terry speaks of, but I'll look into it, and chart my own course, just as the two of you have. I don't think anyone is suggesting otherwise. I know I'm certainly not. I have been very careful to say, In my experience, life is better without newreno. Doug -- We have known freedom's price. We have shown freedom's power. And in this great conflict, ... we will see freedom's victory. - George W. Bush, President of the United States State of the Union, January 28, 2002 Do YOU Yahoo!? To Unsubscribe: send mail to [EMAIL PROTECTED] with unsubscribe freebsd-hackers in the body of the message
Re: ftp and mail much slower into fbsd 4.4 vs and old BSDi
Yes, I can attest to this an I belive it is actually the case on both
-current and -releng4 that disabling newreno improves TCP performance.
I belive running an X11 application or scp(1) over a wavelan is a very
good test-bed for this issue.
--
Poul-Henning Kamp | UNIX since Zilog Zeus 3.20
I find this patch also improves performance of SSH on most network
media. Why so many applications go to the trouble to disable
Nagle's algorithm is beyond me. I suspect the socket option has
too seductive of a name.
louie
Index: packet.c
===
RCS file: /a/cvs/src/crypto/openssh/packet.c,v
retrieving revision 1.1.1.1.2.4
diff -u -r1.1.1.1.2.4 packet.c
--- packet.c28 Sep 2001 01:33:34 - 1.1.1.1.2.4
+++ packet.c2 Apr 2002 18:26:41 -
@@ -1281,9 +1281,11 @@
error(setsockopt IPTOS_LOWDELAY: %.100s,
strerror(errno));
}
+#if 0
if (setsockopt(connection_in, IPPROTO_TCP, TCP_NODELAY, (void *) on,
sizeof(on)) 0)
error(setsockopt TCP_NODELAY: %.100s, strerror(errno));
+#endif
} else if (packet_connection_is_ipv4()) {
/*
* Set IP options for a non-interactive connection. Use
To Unsubscribe: send mail to [EMAIL PROTECTED]
with unsubscribe freebsd-hackers in the body of the message
Re: ftp and mail much slower into fbsd 4.4 vs and old BSDi
Doug Barton wrote: I guess you missed the part where I said that FreeBSD had bugs, and Matt Dillon posted patches? Nope. I think you missed the part where I said I was talking about reality, not theory. :) The reality is, it's broken now, and in my experience, turning it off makes the system work better. I agree that it should be fixed in the long term, but it ain't gonna be me that fixes it. Then use Dillon's patches, instead of just turning it off. Your anecdotal experience with works better is just that -- anecdotal. It doesn't hold for the general case. I guess the problem is that the patches are not committed to the version of FreeBSD you are using, and you are unwilling to upgrade, and unwilling to go look where I tell you to go look? This was only 7 months ago, people! For those people who are unable to search list archives themselves, because they don't know how to ask search engines questions, or whatever, here is a reference to Dillon's patch: http://docs.freebsd.org/cgi/getmsg.cgi?fetch=881527+0+archive/2001/freebsd-hackers/20011202.freebsd-hackers Now *QUIT* turning off NewReno, it's *not* doing what you *think* it's doing, and doing something without understanding why it *appears* to work is as bad as waving a dead chicken over the server to keep away denial of service attacks! Sheesh! -- Terry To Unsubscribe: send mail to [EMAIL PROTECTED] with unsubscribe freebsd-hackers in the body of the message
Re: ftp and mail much slower into fbsd 4.4 vs and old BSDi
On Mon, 1 Jul 2002, Terry Lambert wrote: Doug Barton wrote: I guess you missed the part where I said that FreeBSD had bugs, and Matt Dillon posted patches? Nope. I think you missed the part where I said I was talking about reality, not theory. :) The reality is, it's broken now, and in my experience, turning it off makes the system work better. I agree that it should be fixed in the long term, but it ain't gonna be me that fixes it. Then use Dillon's patches, instead of just turning it off. Dillon's patches were committed, by Dillon. :) My systems still work better with newreno off than with it on. Your anecdotal experience with works better is just that -- anecdotal. It doesn't hold for the general case. I never claimed otherwise. In fact, I specifically suggested that this would be _one_ place to start testing. -- We have known freedom's price. We have shown freedom's power. And in this great conflict, ... we will see freedom's victory. - George W. Bush, President of the United States State of the Union, January 28, 2002 Do YOU Yahoo!? To Unsubscribe: send mail to [EMAIL PROTECTED] with unsubscribe freebsd-hackers in the body of the message
Re: ftp and mail much slower into fbsd 4.4 vs and old BSDi
Doug Barton wrote: Then use Dillon's patches, instead of just turning it off. Dillon's patches were committed, by Dillon. :) My systems still work better with newreno off than with it on. Your anecdotal experience with works better is just that -- anecdotal. It doesn't hold for the general case. I never claimed otherwise. In fact, I specifically suggested that this would be _one_ place to start testing. The whole thread from which the quoted message was taken is relevent. I give up. Turn of NewReno. At the very least, it means that under congestion in any segment, I won't have to worry about your packets being able to reach my segment. 8^p. -- Terry To Unsubscribe: send mail to [EMAIL PROTECTED] with unsubscribe freebsd-hackers in the body of the message
Re: ftp and mail much slower into fbsd 4.4 vs and old BSDi
On Jul 01, at 03:34 PM, Terry Lambert wrote: Then use Dillon's patches, instead of just turning it off. Your anecdotal experience with works better is just that -- anecdotal. It doesn't hold for the general case. I guess the problem is that the patches are not committed to the version of FreeBSD you are using, and you are unwilling to upgrade, and unwilling to go look where I tell you to go look? That was back in 4.3, I believe. I patched my 4.2-REL box with Matt's diffs from that thread, then backported it as far back as 4.1. That was the FreeBSD performing worse than Linux? thread, in November last. The 'tbench', DELACK, and newreno brouhaha. I think it was committed by 4.4-REL, but I know 4.5-REL has that code. This was only 7 months ago, people! For those people who are unable to search list archives themselves, because they don't know how to ask search engines questions, or whatever, here is a reference to Dillon's patch: http://docs.freebsd.org/cgi/getmsg.cgi?fetch=881527+0+archive/2001/freebsd-hackers/20011202.freebsd-hackers That was but a part of Matt's patches. There were a total of three from him about this, in rapid succession, I believe. -- Terry I was right behind you in looking this stuff up. :-) Dave -- __ __ \__ \D. J. HAWKEY JR. / __/ \/\ [EMAIL PROTECTED]/\/ http://www.visi.com/~hawkeyd/ To Unsubscribe: send mail to [EMAIL PROTECTED] with unsubscribe freebsd-hackers in the body of the message
Re: ftp and mail much slower into fbsd 4.4 vs and old BSDi
Len Conrad wrote: Sorry, hackers, I posted this twice in -questions and got no response. If the problem is newreno, can somebody say how to up just that piece for 4.4 so as to be as non-disruptive, non-dice-rolling as possible on this otherwise solid machine? You can turn off newreno with the sysctl. I put this in /etc/sysctl.conf on my machines: net.inet.tcp.newreno=0 To Unsubscribe: send mail to [EMAIL PROTECTED] with unsubscribe freebsd-hackers in the body of the message
