From owner-freebsd-hackers Tue Oct 24 12:53:19 1995 Return-Path: owner-hackers Received: (from root@localhost) by freefall.freebsd.org (8.6.12/8.6.6) id MAA28787 for hackers-outgoing; Tue, 24 Oct 1995 12:53:19 -0700 Received: from godzilla.zeta.org.au (godzilla.zeta.org.au [203.2.228.19]) by freefall.freebsd.org (8.6.12/8.6.6) with ESMTP id MAA28777 for <hackers@freebsd.org>; Tue, 24 Oct 1995 12:53:13 -0700 Received: (from bde@localhost) by godzilla.zeta.org.au (8.6.9/8.6.9) id FAA30444; Wed, 25 Oct 1995 05:51:42 +1000 Date: Wed, 25 Oct 1995 05:51:42 +1000 From: Bruce Evans <bde@zeta.org.au> Message-Id: <199510241951.FAA30444@godzilla.zeta.org.au> To: bde@zeta.org.au, dennis@etinc.com Subject: Re: Async utilization..... Cc: hackers@freebsd.org Sender: owner-hackers@freebsd.org Precedence: bulk > Its not a real measurement, so you can't use it. Period. It is a real measurement. >Figure out the processing requirement for handling one average frame size of > bytes with a 16450 with 8-bit I/O cycles and loads of interrupts, add 20% >and compare I do that routinely, but it's not as accurate as an actual measurement. E.g., typical 16550 input on a 486DX2/66 8MHz ISA bus (non-multiport case), for a typical receiver interrupt: interrupt overhead 5-10 us i/o's for line status 14 i/o's for data 14 i/o's for modem status 1 i/o's for interrupt id 1 total i/o's 30 @ 1.25 us each 37.5 us ------- h/w dominated i/o overhead 3-3.5 us/byte ----------- measured total overhead 6.3 us/byte For a typical 16550 transmitter interrupt: interrupt overhead 5-10 us i/o's for line status 1 i/o's for data 16 i/o's for modem status 1 i/o's for interrupt id 1 total i/o's 19 @ 1.25 us each 24 us ------- h/w dominated i/o overhead 1.8-2.1 us/byte ----------- measured total overhead 2.9 us/byte For typical 16450 input: interrupt overhead 5-10 us i/o's for line status 1 i/o's for data 1 i/o's for modem status 1 i/o's for interrupt id 1 total i/o's 4 @ 1.25 us each 5 us ------- h/w dominated i/o overhead 10-15 us/byte ----------- measured total overhead 16 us/byte For WB8013EBT input, per 1500 byte packet: interrupt overhead low i/o's for interrupt 3 low memory mapped i/o's for data 750 @ 0.625 us each other things I forgot low? ------------- more than h/w dominated i/o overhead 0.3125 us/byte ------------- measured total overhead 0.7 us/byte >it to a single interrupt and one 16-bit bus transfer per frame. It doesn't >take a rocket scientist >to know there's a signficant difference in processing requirements. The factor tends to get reduced a lot by non-data i/o's and protocol protocol processing. Bruce