Re: Sysctl knob(s) to set TCP 'nagle' time-out?
On 24/06/2008, at 2:42 AM, Matthew Dillon wrote: It should be noted that Nagle can cause high latencies even when delayed acks are turned off. Nagle's delay is not timed... in its simplest description it prevents packets from being transmitted for new data coming from userland if the data already in the sockbuf (and presumably already transmitted) has not yet been acknowledged. Assuming that a full data packet can't be constructed in the time it takes for the acknowledgement. If you CAN construct a whole packet in that time then Nagle is either doing a good job or you're sending large amounts of data.. Perhaps nagle a) needs a time out, though I don't really think that would help, or b) uses a dynamic 'in-flight' count where it tries to maintain x packets in-flight and only holds packets up when that value is reached.. The idea being that you get the ack on your first packet at the same time as the host should be getting your second packet.. That way you still get to concatenate lots of small packets being generated in a short space of time, but don't hold up sending data because of the ack latency. It should also be possible to detect if the remote host is using delayed acks and compensate for that? Though I'v not considered it in much detail.. ___ freebsd-stable@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-stable To unsubscribe, send any mail to [EMAIL PROTECTED]
Re: Sysctl knob(s) to set TCP 'nagle' time-out?
:Hi, : :I'm wondering if anything exists to set this.. When you create an INET :socket :without the 'TCP_NODELAY' flag the network layer does 'naggling' on your :transmitted data. Sometimes with hosts that use Delayed_ACK :(net.inet.tcp. :delayed_ack) it creates a dead-lock where the host will not ACK until :it gets :another packet and the client will not send another packet until it :gets an ACK.. : :The dead-lock gets broken by a time-out, which I think is around 200ms? : :But I would like to change that time-out if possible to something :lower, yet :I can't really see any sysctl knobs that have a name that suggests :they do :that.. : :So does anyone know IF this can be tuned and if so by what? : :Cheers, :Jerahmy. : :(And yes you could solve it by setting the TCP_NODELAY flag on the :socket, :but not everything has programmed in options to set it and you don't :always :have access to the source, besides setting a sysctl value would be much :simpler than recompiling stuff) There is a sysctl which adjusts the delayed-ack timing, its called net.inet.tcp.delacktime. The default is 1/10 of a second (100 == 100 ms = 1/10 of a second). BUT, it shouldn't be possible for nagle to deadlock against delayed acks unless the TCP implementation is broken somehow. A delayed ack is simply that... the ack is delayed 100 ms in order to improve its chances of being piggy-backed on return data. The ack is not blocked completely, just delayed, and certain events (such as the receiving end turning around and sending data back, which is typical for an interactive connection)... certain events will cause the delayed ack to be aborted and for the ack to be immediately sent with the return data. Can it break down and cause excessive lag? Yes, it can. Interactive games almost universally have to disable Nagle because the lag is actually due to the data relay from client 1 - server then relaying the interactive event to client 2. Without an immediate interactive response to client 1 the ack gets delayed and the next event from client 1 hits Nagle and stops dead in the water until the first event reaches client 2 and client 2 reacts to it (then client 2 - server - (abort delayed ack and send) - client 1 (client 1's nagle now allows the second event to be transmitted). That isn't a deadlock, just really poor interactive performance in that particular situation. Delayed acks also have a safety valve. The spec says that an ack cannot be delayed more then two packets. In a batch link when the second (unacked) packet is received, the delayed ack is aborted and an ack is immediately returned to the sender. This is to prevent congestion control (which is based on acks) from getting completely out of whack and also to prevent the TCP window from getting exhausted. In anycase, the usual solution is to disable Nagle rather then mess with delayed acks. What we need is a new Nagle that understands the new reality for interactive connections... something that doesn't break performance in the 'server in the middle' data relaying case. -Matt ___ freebsd-stable@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-stable To unsubscribe, send any mail to [EMAIL PROTECTED]
Re: Sysctl knob(s) to set TCP 'nagle' time-out?
On Mon, Jun 23, 2008 at 05:25:49PM +1000, Jerahmy Pocott wrote: So does anyone know IF this can be tuned and if so by what? You can tune it with net.inet.tcp.delacktime - it should be is ms. David. ___ freebsd-stable@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-stable To unsubscribe, send any mail to [EMAIL PROTECTED]
Re: Sysctl knob(s) to set TCP 'nagle' time-out?
On 23/06/2008, at 6:27 PM, Matthew Dillon wrote: Can it break down and cause excessive lag? Yes, it can. Interactive games almost universally have to disable Nagle because the lag is actually due to the data relay from client 1 - server then relaying the interactive event to client 2. Without an immediate interactive response to client 1 the ack gets delayed and the next event from client 1 hits Nagle and stops dead in the water until the first event reaches client 2 and client 2 reacts to it (then client 2 - server - (abort delayed ack and send) - client 1 (client 1's nagle now allows the second event to be transmitted). That isn't a deadlock, just really poor interactive performance in that particular situation. Yeah, that's what I'm talking about. True, it's not really a dead-lock, but it's terribly slow! The interaction can cause a 200ms delay on a LAN, as can be seen with samba if you disable tcp_nodelay.. In anycase, the usual solution is to disable Nagle rather then mess with delayed acks. What we need is a new Nagle that understands the new reality for interactive connections... something that doesn't break performance in the 'server in the middle' data relaying case. Exactly, there is nothing really wrong with delayed acks.. But with sysctl I CAN disable and mess with the delayed acks, but I can't seem to do anything to Nagle. That's why I was thinking if I could change the Nagle time-out to 0ms it would effectively disable it.. Cheers. J. ___ freebsd-stable@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-stable To unsubscribe, send any mail to [EMAIL PROTECTED]
Re: Sysctl knob(s) to set TCP 'nagle' time-out?
On 23/06/2008, at 7:00 PM, David Malone wrote: On Mon, Jun 23, 2008 at 05:25:49PM +1000, Jerahmy Pocott wrote: So does anyone know IF this can be tuned and if so by what? You can tune it with net.inet.tcp.delacktime - it should be is ms. Yeah I saw that one. But that only changes the delayed ack... The default value of 100ms seems fairly reasonable unless you're talking about a LAN.. I guess what I really want to do is disable Nagle in the tcp stack, but since you do that with the sockopts call on a per socket basis I'm guessing there isn't any system wide tunable for it.. Thanks, Jerahmy. ___ freebsd-stable@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-stable To unsubscribe, send any mail to [EMAIL PROTECTED]
Re: Sysctl knob(s) to set TCP 'nagle' time-out?
Matthew Dillon wrote: In anycase, the usual solution is to disable Nagle rather then mess with delayed acks. What we need is a new Nagle that understands the new reality for interactive connections... something that doesn't break performance in the 'server in the middle' data relaying case. One possibility I see is a statistic about DelACKs per TCP connection, counting those that were rightfully delayed (with hindsight). I.e., if an ACK is delayed, but there was no chance to piggy-back it or to combine it with another ACK, it could have been sent without delay. Only those delayed ACKs that reduce load are good, all others cause additional state to be maintained and may increase latencies for no good reason. Therefore, I thought about starting with Nagle enabled, but give up on delaying ACKs, when doing so is found to be ineffective. The only problem with this approach is that once TCP_NODELAY is implicitly set due to measured behavior of the communication, a situation that would benefit from delayed ACKs can no longer be detected. (Well, you could measure the delay between an ACK and the next data sent to the same destination; disable TCP_NODELAY if ACKs could have been piggy-backed on data packets without too much delay. May be we could really have TCP auto-tune with respect to use of delayed ACKs ... I had suggested this years back, when the issue was discussed, but consensus was, that you should just set TCP_NODELAY. But automatic adjustment could also (implicitly) take RTT, window size into consideration. And to me, automatic setting of TCP_NODELAY seems more useful than automatic clearing (after delayed ACKs had been found to be of no use for a window of say 8 or 16 ACKs). The implementation would be quite simple: Whenever a delayed ACK is sent, check whether it is sent on its own (bad) or whether it could be piggy-backed (good). If, say, 7 of 8 delayed ACKs had to be sent as ACK-only packets, anyway, set TCP_NODELAY and do not bother to keep on deciding whether delayed ACKs had become useful in a different phase of the communication. If you want to be able to automatically disable TCP_NODELAY, then just set a time-stamp whenever an ACK is sent and when the next data is sent through this same socket, check whether delaying the ACK had allowed to send it with that data packet (i.e. the delay was less than the maximum hold time of the delayed ACK). If it had been beneficial to delay ACKs (say 3 out of a window of 4) then clear TCP_NODELAY. I have no idea, whether SMP locking would be problematic, but I guess the checks and counter updates could be put in sections that are appropriately locked, anyway. Regards, STefan ___ freebsd-stable@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-stable To unsubscribe, send any mail to [EMAIL PROTECTED]
Re: Sysctl knob(s) to set TCP 'nagle' time-out?
:One possibility I see is a statistic about DelACKs per TCP connection, :counting those that were rightfully delayed (with hindsight). I.e., :if an ACK is delayed, but there was no chance to piggy-back it or to :combine it with another ACK, it could have been sent without delay. :Only those delayed ACKs that reduce load are good, all others cause :additional state to be maintained and may increase latencies for no :good reason. : :... :consideration. And to me, automatic setting of TCP_NODELAY seems :more useful than automatic clearing (after delayed ACKs had been :found to be of no use for a window of say 8 or 16 ACKs). : :The implementation would be quite simple: Whenever a delayed ACK :is sent, check whether it is sent on its own (bad) or whether it :could be piggy-backed (good). If, say, 7 of 8 delayed ACKs had to :be sent as ACK-only packets, anyway, set TCP_NODELAY and do not :bother to keep on deciding whether delayed ACKs had become useful :in a different phase of the communication. If you want to be able :to automatically disable TCP_NODELAY, then just set a time-stamp :... :Regards, STefan That's an interesting approach. I think it would catch some of the cases, but not enough of them. If the round-trip in the server-relaying case is less then the delayed-ack, the acks will still wind up piggy-backed on return traffic but the latency will also still remain horrible. It should be noted that Nagle can cause high latencies even when delayed acks are turned off. Nagle's delay is not timed... in its simplest description it prevents packets from being transmitted for new data coming from userland if the data already in the sockbuf (and presumably already transmitted) has not yet been acknowledged. For interactive traffic this means that Nagle is putting the screws on the packet stream even if the acks aren't delayed, simply from the ack latency. With delayed acks turned off the latency is lower, but not 0, so interactive traffic is still being held up by Nagle. The effect is noticeable even on a LAN. Jerahmy brought up Samba... that is an excellent example. NFS-over-TCP would be another good example. Any protocol which multiplexes multiple commands from different sources over the same connection gets really messed up (slowed down) by Nagle. On the flip side, Nagle can't just be turned off by default because it would cause streaming connections from user programs which do tiny writes to generate a lot of unnecessarily tiny packets. This can become apparent when using SSH over a slow link. Numerous programs run from a shell generate fairly ineffcient packets which could have easily been batched when operating over SSH. The result can be sludgy performance for output which ought be batched up by TCP but isn't because SSH turns off Nagle unconditionally. -Matt Matthew Dillon [EMAIL PROTECTED] ___ freebsd-stable@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-stable To unsubscribe, send any mail to [EMAIL PROTECTED]