From owner-freebsd-hackers Thu Oct 19 17:04:53 1995 Return-Path: owner-hackers Received: (from root@localhost) by freefall.freebsd.org (8.6.12/8.6.6) id RAA06657 for hackers-outgoing; Thu, 19 Oct 1995 17:04:53 -0700 Received: from brasil.moneng.mei.com (brasil.moneng.mei.com [151.186.20.4]) by freefall.freebsd.org (8.6.12/8.6.6) with ESMTP id RAA06652 for ; Thu, 19 Oct 1995 17:04:50 -0700 Received: (from jgreco@localhost) by brasil.moneng.mei.com (8.7.Beta.1/8.7.Beta.1) id TAA28849; Thu, 19 Oct 1995 19:03:19 -0500 From: Joe Greco Message-Id: <199510200003.TAA28849@brasil.moneng.mei.com> Subject: Re: Bragging rights.. To: root@spiffy.cybernet.com Date: Thu, 19 Oct 1995 19:03:19 -0500 (CDT) Cc: hackers@freebsd.org In-Reply-To: from "Mark J. Taylor" at Oct 19, 95 07:53:59 pm X-Mailer: ELM [version 2.4 PL24] Content-Type: text Sender: owner-hackers@freebsd.org Precedence: bulk > On 10/19/95 20:25:56 David Greenman wrote: > [clip] > > > > Let me add a bit of sanity to this part of the discussion. 115200 baud async > > >will give you about 11.52Kbytes/second if you have no packet overhead. 115200 > >baud sync will give you 14.40Kbytes/second if you have no packet overhead. > > Why? Because we're talking bits - async is 8 data bits plus 1 start and 1 > >stop bit...10 bits. With synchronous serial, it's just 8 data bits. So sync > >always has the potential to give you 25% more bytes throughput at the same bit > >rate compared to async. > > Now with sync you'll also be running at a faster bit rate (128000bits/sec). > >This is 16Kbytes/second. This is 38.9% faster. > > > >-DG Granted; but you need sync hardware on both ends of the connection, then. My original point is that that is typically more expensive... > As a slightly interested party, I'd like to ask: > > As mentioned recently on -hackers, isn't it possilbe to up the rate of the serial > chip simply by doubling (or quadding) the rate of the xtal driving the chip? Generally. Check the chip specs. NS parts can be. 3.6864 MHz = 2x (so "57600" stty value yields a 115200 real bit rate) 7.3728 MHz = 4x (max supported by many chips) > Many (most?) 16550 chips should be able to handle a Fmax higher than they are being > driven, and with 16 byte FIFOS (set to trigger at 14 bytes), the interrupt overhead > would not necessarily be increased. No, because you're receiving (or sending) at a higher rate of speed. The per-call interrupt overhead may not be higher, but there will be twice (or four times) as many interrupts as you would normally expect. > Is the same xtal trick applicable to sync serial, to get 32 KBytes/second @256000 > bits/sec (as opposed to 28.8 KBytes/sec async serial @230400 bits/sec)? Dennis started this thread on the assertion that his sync serial cards can do very high speeds quite easily :-) I have a 386DX/40 routing between a T1 (1.544Mb/s) and an Ethernet and it handles average mixes of traffic without any trouble (it runs into problems if you start saturating the T1 with small packets however). It is QUITE impressive :-) I don't think you'd need to do any crystal switches to do what you describe, since the board is designed to be quite flexible. However, you are probably limited to packet modes such as PPP (Dennis? I am extrapolating here, any solid info? I was never able to access raw data streams, although I only looked briefly) ... Joe ------------------------------------------------------------------------------- Joe Greco - Systems Administrator jgreco@ns.sol.net Solaria Public Access UNIX - Milwaukee, WI 414/342-4847