Date: Wed, 11 Jul 2001 12:55:47 -0400 From: Leo Bicknell <bicknell@ufp.org> To: Jim Pirzyk <Jim.Pirzyk@disney.com> Cc: Leo Bicknell <bicknell@ufp.org>, freebsd-bugs@FreeBSD.ORG Subject: Re: conf/28882: Network defaults are absurdly low. Message-ID: <20010711125547.A67113@ussenterprise.ufp.org> In-Reply-To: <01071109140103.89158@snoopy>; from Jim.Pirzyk@disney.com on Wed, Jul 11, 2001 at 09:14:01AM -0700 References: <200107111550.f6BFo2g12771@freefall.freebsd.org> <01071109140103.89158@snoopy>
next in thread | previous in thread | raw e-mail | index | archive | help
On Wed, Jul 11, 2001 at 09:14:01AM -0700, Jim Pirzyk wrote: > I find in our enviroment that 16K is an optimal size for 10M Flat > enet and 32K is the best for 100Mb switched and 512k works for OC3 > atm. This is though with a maxsockbuf set to 768k and tcp_extentions > turned on. Two FreeBSD 4.2 systems, 100 Meg FE to an ISP network that can support a full 100 Mbps end to end. round-trip min/avg/max/stddev = 66.893/66.952/67.021/0.050 ms (San Jose CA, to Washington DC) System defaults, 16k socket buffers: 36985815 bytes received in 172.74 seconds (209.10 KB/s) It can fill a T1. It's only 20% of a 10 Meg cable modem though, or a 10 Meg ethernet connected host off a high speed line. I don't know about you, but everyone I've talked to finds defaults that produce this sort of rate limit absurd. Customers routinely install boxes on the network, do a transfer between them, and then spend the next 3 days bitching to us about how much the network stinks. However, if you try some other values: 32k socket buffers: 36985815 bytes received in 81.30 seconds (444.29 KB/s) Note, we still can't saturate even a 10 Meg ethernet (almost 20 year old technology), or a 10 Meg cable modem. Getting better though. Can the network support more, sure! 256k socket buffers (FreeBSD's default maximum): 36985815 bytes received in 10.26 seconds (3.44 MB/s) Now we're cooking with gas. I might consider this good enough for the average user on a single connection. Those who really need better performance can tune their systems. Let's do one last try: 1M socket buffers: 36985815 bytes received in 5.39 seconds (6.55 MB/s) But you know what, that 36 Meg file is now too large to get us past slow start and other goodies, and get a good transfer rate! So, let's try a bigger file: 1M socket buffers: 184929075 bytes received in 23.77 seconds (7.42 MB/s) So, FreeBSD's defaults are limiting the install to 209 K / 7.42 MB or 2.8% OF MAXIMUM THROUGHPUT! Even if you consider that my world of 100 meg cross country bandwidth is rare, a fair number of people these days have the ability to get 10Mbps/sec, and they are being liminted to 20% of their available bandwidth. Notably, most Linux installations use 32k defaults, which at least doubles the FreeBSD performance. If nothing else comes out of this I would hope FreeBSD can step up to the same level of performance as Linux and Solaris. I can only assume that if you find 16k sufficient, that your hosts are only running across a LAN segment on the same campus. -- Leo Bicknell - bicknell@ufp.org Systems Engineer - Internetworking Engineer - CCIE 3440 Read TMBG List - tmbg-list-request@tmbg.org, www.tmbg.org To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-bugs" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20010711125547.A67113>