Path: utzoo!utgpu!news-server.csri.toronto.edu!cs.utexas.edu!uunet!ncrlnk!ncr-mpd!Chuck.Phillips From: Chuck.Phillips@FtCollins.NCR.COM (Chuck.Phillips) Newsgroups: comp.sys.amiga.hardware Subject: UNIX memory consumption (was: Amiga-Sun unix question) Message-ID: Date: 14 Jun 90 13:02:00 GMT References: <2029@mindlink.UUCP> <12477@cbmvax.commodore.com> <5777@hub.ucsb.edu> Sender: uucp@ncr-mpd.FtCollins Organization: NCR Microelectronics, Ft. Collins, CO Lines: 87 In-reply-to: dz@lemon.ucsb.edu's message of 12 Jun 90 01:08:49 GMT Daniel> On 12 Jun 90 01:08:49 GMT, dz@lemon.ucsb.edu (Daniel James Zerkle) said: Randell> In article <12477@cbmvax.commodore.com> jesup@cbmvax (Randell Jesup) writes: Randell> Our SysVR4 Unix will support both X and OpenLook (on top of X I think), Randell> so it should work. (No Suntools, of course.) Daniel> Wonderful! Is this official? >I think X works in 4MB, but >OpenLook may require more than 4MB ram to work well. THe unix guys could >answer more precisely. Daniel> Being a "Unix Guy", I pity the fool who tries to run ANY windowing Daniel> environment, and expecially X, in less than 8 megabytes on UNIX. For the MC680x0 architecture, I agree. Unsurprisingly most RISC processors want _lots_ more. (12-16MB for Sparc) There are mitigating factors on the Amiga, however. One is shared libraries (which UNIX didn't have until SVr4 ;-). The other is the fact that many of the graphics routines are in ROM or hardware and needn't be compiled into the server. (Compare this to the Mac and PC situations. ;-) Daniel> I don't know about OpenLook, but as it is a huge application that runs Daniel> on top of X, I would add memory appropriately. "olwm" Open Look Window Manager) isn't the largest window manager for X (mwm and gwm are larger), but it isn't the smallest, either. "twm" by comparison is smaller and faster, just not as fancy. Who knows, C= could write "wbwm" (WorkBench Window Manager) that would provide a Workbench like look and feel and release a compatibility library to allow old Workbench programs to be recompiled to run under X. This is essentially what Sun did when they wrote the Open Look toolkit. Sun also made the source code for the Open Look libraries and window manager freely redistributable. (Hint, hint. :-) Daniel> Also, remember that if you get Unix, you will need high performance disk Daniel> storage. EVERYTHING that Unix does requires an access to the disk. Very true. Question: Does the C= (or any other) Amiga SCSI controller suport "synchronous" SCSI? I seem to recall the difference in maximum throughput being more than a factor of 3. (BTW, the Quantum 110MB drives support this.) Daniel> What's more, the virtual memory system swaps and pages memory that has Daniel> not been used lately to the disk drive. Under SVr4 UNIX (and SunOS >= 4.0), the paging algorithm is a bit different than most versions of UNIX. Memory is only paged out when more RAM is needed than is available. In a machine with lots of memory, unused pages could hang around for a long time. Daniel> However, I cringe at the thought of Unix Daniel> running on a slow disk with not enough RAM. Actually, I cringe at the Daniel> memory of doing this. To be fair, UNIX isn't much worse about memory consumption than OS/2. The CPU overhead of UNIX is also _much_ less than that of OS/2. (Based on the results of benchmarks run on an NCR 486 PC published recently in "Personal Workstation") Daniel> Anyway- Here's a summary of performance effects from most to least Daniel> important: Daniel> ... Other than to comment that the RAM estimates might be a bit high, I'd like to add: 6. Programs that read and write files sequentially (rather than random access) may see a dramatic improvement in execution time. In UNIX, disk writes are asynchronous and parallel to the the process requesting the write. (This can be overridden in SVr4 and Xenix.) Many versions of UNIX will also read ahead, so disk reads may see an improvement, also. In UNIX, you can actually have a program complete execution in less wall clock time than the sum of disk seek times required for the program's I/O. Daniel> | Dan Zerkle home:(805) 968-4683 work:687-0110 | Daniel> | dz@cornu.ucsb.edu dz%cornu@ucsbuxa.bitnet ...ucbvax!hub!cornu!dz | Daniel> | Snailmail: 6681 Berkshire Terrace #5, Isla Vista, CA 93117 | Daniel> | Disclaimer: My fish are stupid. | Disclaimer: Not a spokesman, etc., etc. And his fish ARE stupid. (Just kidding! See the smiley? -> :-) -- Chuck Phillips MS440 NCR Microelectronics Chuck.Phillips%FtCollins.NCR.com Ft. Collins, CO. 80525 uunet!ncrlnk!ncr-mpd!bach!chuckp