Date: 9 Dec 1995 16:37:11 +0800 From: peter@haywire.dialix.com (Peter Wemm) To: freebsd-hackers@freebsd.org Subject: Re: User space PPP & leased line Message-ID: <4abhrn$hv8$1@haywire.DIALix.COM> References: <28835.818443536@time.cdrom.com>, <199512081834.LAA00699@rocky.sri.MT.net>
next in thread | previous in thread | raw e-mail | index | archive | help
nate@rocky.sri.MT.net (Nate Williams) writes: >> > One thing we DID notice was that the transfer speeds using the user mode >> > code were SIGNIFICANTLY higher than those using the kernel mode code - >> > for example, transferring a medium size binary file (one that could be >> > fully transferred in < 2 minutes :-) using pppd takes an average of >> > 2.3KB/s whereas with ijppp it's more like 3.3KB/s (using a V.fc modem >> > connection at 26.4Kb/s). I'm not sure why this should be, but given the >> >> That's also worth investigating. I can't see any inefficiency that >> should result in *that* much slowdown, and my own tests showed >> virtually no difference between the two >In a word, 'Predictor-1 compression'. It's available on ijppp, and not >with kernel-mode ppp and makes a significant difference. Kernel-mode PPP has LZ/LZW compression (based on the /usr/bin/compress code). It works, but I'm not happy with it. There is a subtle bug in kernel-mode ppp that I've been unable to find, but it makes the compression go to hell if it's active. For example, if I sent a compressible text file uncompressed (over 11 bit LZW v42bus modems), I was getting about 5.6KB/sec (on 57.6K DTE speed). If I turned on software compression, and off on the modem (28.8K), I was getting about than 2K/sec throughput. The machine was practically idle. Somehow packets were getting excessively lost and TCP was loosing the plot. Most likely, I botched something with the splxxx() locking. I'd love to hear any suggestions.. I dont have the time to look at it at the moment... -Peter >Nate
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?4abhrn$hv8$1>