From owner-freebsd-hackers Mon Dec 26 17:34:02 1994 Return-Path: hackers-owner Received: (from root@localhost) by freefall.cdrom.com (8.6.9/8.6.6) id RAA03228 for hackers-outgoing; Mon, 26 Dec 1994 17:34:02 -0800 Received: from n8ino.mainstream.com (n8ino.mainstream.com [192.231.143.2]) by freefall.cdrom.com (8.6.9/8.6.6) with ESMTP id BAA03222 for ; Tue, 27 Dec 1994 01:33:58 GMT Received: from localhost (uucp@localhost) by n8ino.mainstream.com (8.6.5/8.6.5) id UAA21844 for FreeBSD-hackers@freefall.cdrom.com; Mon, 26 Dec 1994 20:33:26 -0500 Received: by dg1020 (5.4.2/5.40/1.0) id AA08221; Mon, 26 Dec 1994 19:55:23 -0500 Date: Mon, 26 Dec 1994 19:55:23 -0500 From: dg1020!w_taylor@n8ino.Mainstream.com (Bill Taylor) Message-Id: <9412270055.AA08221@dg1020> To: n8ino!FreeBSD-hackers@freefall.cdrom.com Cc: n8ino!dufault@hda.com, n8ino!smcarey@mailbox.syr.edu Subject: Real-time and parallel extensions to FreeBSD Sender: hackers-owner@freebsd.org Precedence: bulk I recently heard from dufault@hda.com that there is interest in real-time extensions to FreeBSD. Since my firm is interested in both real-time and parallelism, he suggested that I ask this mailing list about both. My firm sells an extremely powerful software package designed for real-time process control. Details are availble by e-mail on request, but to oversimplify, our product is a knowledge processing system which handles more than 300,000 "rules" per second on a single 68030. In the very short term, a PowerPC is probably enough, but most of our systems require a rather powerful MMI which usually needs its own processor. In our design, the MMI task and the control task or tasks communicate via shared memory. Our customers are sufficiently interested in what we are doing to ask us for proposals for systems which are too big to run on a single power PC. We need more MIPs; the simplest approach we know is to spread our application across multiple processors in a single VME rack. We do not know of a successful parallelism setup for the IBM PC architecture, if there is one, we would consider it, too. As to real-time requirements, our real need is for consistent response. We do NOT process a high volume of interrupts, we tend to offload interrupt-intense parts of the application to auxiliary processors, but we need consistent response. Normal Unix systems seem to snatch a significant chunk of resources for housekeeping on a basis which we do not understand. One approach would be to put the housekeeping on its own processor. dufault@hda.com recently distributed a message speaking of reducing the cost of task switching and making several other kernel adjustments in the interest of making FreeBSD better for real-time. I would appreciate knowing the level of interest in both real-time and parallel processing in the FreeBSD community. We use FreeBSD to run a single-task portion of our application and are impressed with it. We would like to standardize on one operating system; as far as we know, FreeBSD is the most robust around, but we nee to know whether there are plans to take it parallel. If there is interest but implementing the interest requires nonexistent resources, we would also be interested in knowing the resource requirements, there might be something we could do. Thank you very much. bill@mainstream.com