Date: Wed, 12 Jan 2022 11:35:19 -0800 From: Gleb Smirnoff <glebius@freebsd.org> To: net@freebsd.org Subject: Re: compressed TIME-WAIT to be decomissioned Message-ID: <Yd8td6GjihtXtV7f@FreeBSD.org> In-Reply-To: <Yd8im/VkTU1zdvOi@FreeBSD.org> References: <Yd8im/VkTU1zdvOi@FreeBSD.org>
next in thread | previous in thread | raw e-mail | index | archive | help
On Wed, Jan 12, 2022 at 10:48:59AM -0800, Gleb Smirnoff wrote: T> * Many of HTTP connections were made by older browsers, which were not able T> to use persistent HTTP connections. Those browsers that could, would T> recycle connections more often, then today. Default timeouts in Apache T> for persistent connections were short. So, the ratio of connections T> in TIME-WAIT compared to live connections was much bigger than today. T> Here is sample data from 2008 provided to me by Igor Sysoev: T> T> ITEM SIZE LIMIT USED FREE REQUESTS FAILURES T> tcpcb: 728, 163840, 22938, 72722, 13029632, 0 T> tcptw: 88, 163842, 10253, 72949, 2447928, 0 T> T> We see that TIME-WAITs are ~ 50% of live connections. T> T> Today I see that TIME-WAITs are ~ 1% of connections. My data is biased T> here, since I'm looking at servers that do mostly video streaming. I'd T> be grateful if anybody replies to this email with some other modern data T> on ratio between tcpcb and tcptw allocations. Yes, my data was quite biased. I found evidence that for smaller assets served by HTTP the ratio of tcptw/tcpcb can be up to 35% these days. -- Gleb Smirnoff
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?Yd8td6GjihtXtV7f>