Path: utzoo!utgpu!watmath!att!tut.cis.ohio-state.edu!cica!gatech!bbn!kurz-ai!simcha From: simcha@kurz-ai.UUCP (Simcha Lerner) Newsgroups: comp.binaries.ibm.pc.d Subject: Re: DESQview 386 Question Message-ID: <301@kurz-ai.UUCP> Date: 7 Aug 89 16:47:52 GMT References: <956@swdev.Waterloo.NCR.COM> <8989@cs.Buffalo.EDU> <1681@garcon.cso.uiuc.edu> <1989Aug7.053410.12364@sli.com> Reply-To: simcha@kurz-ai.UUCP (Simcha Lerner) Organization: Kurzweil A.I. Waltham, Mass. Lines: 50 In article <1989Aug7.053410.12364@sli.com> rich%sli@uunet.uu.net (Rich K. Braun) writes: >...... >On disk I/O performance: apparently DESQview robs the foreground process of >enough CPU cycles to hurt disk I/O dramatically. I'm running an RLL disk >with 1:1 interleaving, and it appears that although I can read an entire >track full of sectors in a single revolution from PC-DOS, if I do the same >thing under DESQview w/ certain applications (XCOPY is the simplest example), >it's clear that the disk has to spin several times before a track is read in. >Any clues on how to prevent this from happening? > >-rich While I haven't used DESQview 2.2, under version 1.x you had the option to set how many ticks were spent in forground, and how many were spend on each background task. If you aren't getting enough tics __on_average__ to allow reading a full track uninterrupted, you can increase the number of ticks to allocate to the foreground task. If you wish to keep the same ratio of foreground to background ticks, you can increase both. What this will accomplish is to increase the "graininess" of the task swapping - which should give the forground enough time to completely read a full track. The problem of xcopy is that it will keep reading data until it is interrupted. Since it is impossible to guarantee that the ticks will always fall out between track reads, your choise is either: 1. Increase the clock ticks as above, and maximize the number of tracks read before an interrupt. 2. Buy a caching disk controller. Since the reading of a full track takes place independent of the program running, task swapping doesn't bother this type of controller so much. 3. Reduce your interleave to 2:1 (and cry :-) ) [or undocumented alternative 4: get a real operating system instead...] Good luck, -- Simcha Lerner Kurzweil Applied Intelligence PLEASE NOTE ADDRESS: NO RETURN MAIL VIA bbn PLEASE UUCP address: kurz-ai!simcha@talcott.harvard.edu or: ...{uunet,rutgers,ames}!harvard!talcott!kurz-ai!simcha