From quozl at us.netrek.org Wed Apr 4 23:17:21 2018 From: quozl at us.netrek.org (James Cameron) Date: Thu, 5 Apr 2018 14:17:21 +1000 Subject: [netrek-dev] Time based packet transmission In-Reply-To: <0dd62a8f-dc80-837c-756f-26f66a23f764@sven.de> References: <20180309043302.GB2818@us.netrek.org> <0dd62a8f-dc80-837c-756f-26f66a23f764@sven.de> Message-ID: <20180405041720.GI24965@us.netrek.org> On Thu, Mar 15, 2018 at 09:56:55AM +0100, Sven Neuhaus wrote: > Am 09.03.2018 um 05:33 schrieb James Cameron: > > New feature for Linux kernel may be used to more accurately pace > > packets sent from Netrek server to client. > > > > https://lwn.net/Articles/748744/ > > > > At the moment packet timing has jitter because it is implemented in > > user space. > > Interesting. > However, do you think it's worth it? Agreed, it is interesting. But no, I don't think it is worth it. > Have you measured the jitter? Yes. It wasn't huge on a LAN, from memory about 2ms at 10 Hz. On a wide area network other effects become significant. > My guess is that it's less than 1ms unless the server and/or its network > interfaces is/are under considerable load. > > Also delaying the packet just to reduce the jitter would make them > stale, wouldn't it? Yes, it would make them stale, but making all packets stale by a small amount would be acceptable. > Dropping a packet when it couldn't be sent in time doesn't sound very > appealing either. Indeed. Thanks for the response. -- James Cameron http://quozl.netrek.org/