Date: Sat, 8 Dec 2012 14:17:16 +0000 From: David Chisnall <theraven@theravensnest.org> To: Aleksandr Rybalko <ray@FreeBSD.org> Cc: svn-src-projects@FreeBSD.org, Roman Divacky <rdivacky@FreeBSD.org>, src-committers@FreeBSD.org, Jung-uk Kim <jkim@FreeBSD.org> Subject: Re: svn commit: r243914 - projects/bpfjit Message-ID: <06193EEB-C26B-4110-980B-F04A815C9871@theravensnest.org> In-Reply-To: <20121208152447.5b2958d2.ray@freebsd.org> References: <201212052312.qB5NC2Hn056351@svn.freebsd.org> <20121206084936.GA58940@freebsd.org> <50C0DFB0.6030007@FreeBSD.org> <20121208152447.5b2958d2.ray@freebsd.org>
next in thread | previous in thread | raw e-mail | index | archive | help
On 8 Dec 2012, at 13:24, Aleksandr Rybalko wrote: >> LLVM compilation took too much time to be useful: >>=20 >> engine filter cycles compile cycles >> - ---------------+---------------+---------------- >> jit-linux 106468 33126+72796 >> jit-freebsd 113958 48292+72796 >> llvm 157394 380843640+72796 >> pcap 276910 72796 >> linux 351391 9245+72796 As a further note: in the small print for this benchmark, it was done on = 1,000 packets. On a typical network where you might want to use BPF, = that's, what, 20-100ms of network traffic? If you're changing BPF rules = over ten times per second, then you are probably in quite an unusual = usecase. Alternatively, if you're on a network where 1,000 packets take = so long to arrive that it's significant, then your packet filtering = startup time is almost certainly not an issue - no one will notice if it = takes even an extra few seconds before the first pigeon takes off... David=
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?06193EEB-C26B-4110-980B-F04A815C9871>