From: Digestifier To: Linux-Development@senator-bedfellow.mit.edu Reply-To: Linux-Development@senator-bedfellow.mit.edu Date: Tue, 11 Oct 94 13:13:14 EDT Subject: Linux-Development Digest #290 Linux-Development Digest #290, Volume #2 Tue, 11 Oct 94 13:13:14 EDT Contents: Re: Telnet & ftp freeze! (John Richardson) Re: Improving SLIP latency under Linux (Rob Janssen) replacements for dos _bios_keyboard calls (Paul Erkkila) Re: Could TCP/IP be implemented over SCSI? (Martin Blaschke) Re: DOSEMU Questions (386 mode?) (Charles E Meier) Re: Linux For Mac (Aaron 'Raz' Wrasman) Re: Practical experience with Gravis Ultrasound Max? (Patrick D. Ashmore) Re: Linux 1.1.52 (Lies, Damned Lies, and Benchmarks) (Jeff Kuehn) [Q] what is the story with linux.act.* newsgroups (Tim Bass (Network Systems Engineer)) Re: ext2fs vs. Berkeley FFS (Michael Bischoff) /proc/mtab 0.2 for 1.1.52 (Davor Jadrijevic) Linux 1.1.52 (Lies, Damned Lies, and Benchmarks) (Jeff Kuehn) ---------------------------------------------------------------------------- From: jrichard@cs.uml.edu (John Richardson) Crossposted-To: comp.os.linux.help,comp.os.linux.misc,comp.os.linux.admin Subject: Re: Telnet & ftp freeze! Date: 9 Oct 1994 01:05:30 GMT > >Right, but introducing a 3000ms latency in the PPP session is not >really acceptable. > Unfortuantely, I think the problem lies elsewhere if you have my patch. :( -- John Richardson ------------------------------ From: rob@pe1chl.ampr.org (Rob Janssen) Subject: Re: Improving SLIP latency under Linux Reply-To: pe1chl@rabo.nl Date: Sun, 9 Oct 1994 08:55:00 GMT In longyear@netcom.com (Al Longyear) writes: >If you are the only person using the link and you don't want the >telnet session to be degraded by ftp, then don't start the ftp >transfer. (Simple, isn't it?) If you are not the only user then you >don't have this option. However, consider that the person doing the >ftp transfer wants the data transferred as fast as possible as >well. You must consider all of the users, not just one. It is not like that... A real example of SLIP line use: I use SLIP links over 14k4 and 28k8 modems to remotly support networks. A Linux machine is on the client's network and has a dialin modem, and from a central site we dial these modems and can do all kinds of work (telnet, SNMP, remotely running dosemu which accesses Netware, etc) Often I want to transfer some upgrade using FTP, and at the same time look around with telnet. So I am the person that wants both activities and would really like the telnet to behave reasonable while the FTP continues in the background using the remainder of the bandwidth. >The best solution is to simply choose the priority when you do the >transmission and then send it in that order. That only helps when there are as little buffers as possible between the point where this decision is made and the line. Although the buffers in the modems are a problem, I don't think they are the full explanation of 6 second roundtriptimes when an FTP is running. Probably some frames get queued up in the kernel "waiting to be sent on the serial line". That is not good... I will try to install the new networking stuff that promises to solve this situation (the systems are now running plain slackware), but it only helps when the FTP is initiated from a Linux machine. It not always is. Peeking in the TCP headers inside the IP routing code (checking for port numbers, so that 20 and 25 can get a lower priority than 23 even when the TOS is the same) seems to be the best solution... (although dirty) >>How about setting the MTU based on the anticipated amount of interactive >>traffic, and adjusting it based on the actual amount? Or is the MTU set >>once for a session? >The maximum value is set once for the session. You could reduce it, >but increasing it is definitely not a good idea. The MTU should probably not be changed, but of course TCP can be asked to send fewer and smaller segments in those cases where the link has problems. In fact I think the "always use a big window because that performs so nicely on fast links" may be not optimal. One could argue that the optimum window size for a TCP is such that it can just keep sending (i.e. an ACK comes back just when the sending TCP runs out of send window), and not much larger than that. The sending TCP could adapt its send window accordingly (staying below the negotiated window size). Rob -- ========================================================================= | Rob Janssen | AMPRnet: rob@pe1chl.ampr.org | | e-mail: pe1chl@rabo.nl | AX.25 BBS: PE1CHL@PI8UTR.#UTR.NLD.EU | ========================================================================= ------------------------------ From: pee@cci.com (Paul Erkkila) Subject: replacements for dos _bios_keyboard calls Date: Tue, 11 Oct 1994 11:59:12 GMT I need a replacement for the MS/Zortech/Borland Dos _bios_keyboard functions to do a port of some dos code. The two functions that I need explicitly are: _bios_keyboard(KEY_READY) : which returns true if there are any more keystrokes to process. (chars in the keyboard buffer) and _bios_keyboard(READ) : which does a bios keyboard read. I fooled around with ncurses nocrmode; getch; crmode; But it gives me a SIGSEGV everytime I try it with these calls (do I need to set up a curses window for these? I'd rather not..) I am writing in g++ using the iostreams stuff, so if there are functional equivalents to these with cin i'd be a happy camper. Thanks in advance -pee -- ------------------------------ From: martin@mawurgl.franken.de (Martin Blaschke) Subject: Re: Could TCP/IP be implemented over SCSI? Date: Fri, 07 Oct 1994 16:57:52 GMT Reply-To: martin@mawurgl.franken.de In article , Ian McCloghrie writes: > Ummm... it's certainly be possible. But make sense? Last I checked, > fast/wide scsi 2 cards weren't exactly cheap. Granted, it's faster > than ethernet, but how many people really, seriously, need more > than 10Mbits on their LAN? Especially at home (the main Linux > presence). The cabling for it would be a nightmare (25 pairs, not > just 2 as in 10baseT, *and* you have to worry about termination). > It is not a question of what you need or what a scsi-controller would could! This "network-solution" would be for those who ALREADY HAVE these controllers in their PCs,... And it would be very nice to use them for cheap and fast networking (nearly as cheap as slip via nullmodem). > It's a cool idea, and perhaps useful in a Desk-Area-Network concept > (put an xterminal on your system via scsi and use that as your main > display, etc). But it's not well suited as a general LAN technology. Nobody is talking of GENERAL! Just for cheap and effective fun-networking... :-)))))) > > -- > Ian McCloghrie work: ianm@qualcomm.com home: ian@egbt.org > -- Martin Blaschke ------------------------------------------------------------------- > E-Mail: martin@mawurgl.franken.de < ------------------------------------------------------------------- ------------------------------ From: cemeier@magnus.acs.ohio-state.edu (Charles E Meier) Subject: Re: DOSEMU Questions (386 mode?) Date: 9 Oct 1994 14:01:12 GMT In article , Rob Janssen wrote: >In <36ugt2$id3@DGS.dgsys.com> hitman@dgs.dgsys.com (Douglas Rankin) writes: > >> I know this is probably a dumb question. does dosemu emulate 386 protected >>mode? If not are there plans to do so?? The reason I am wondering is that >>I have several dos programs that use Phar Lap dos extender and it won't >>run becasue it say the chip is in 8086 mode. Is there a way to have it run >>in 385 mode so I can run thse programs, If so how, any help would >>be apprecited!! Thanks > >It doesn't, and I don't think it will do it any time soon. > >However, most dos extenders (I don't know Phar Lap) are also willing to >live with one of the protected mode interfaces, like DPMI. This is >currently being worked on in dosemu. Phar Lap helped design/develop the DPMI standard. In fact, when I wanted a copy of the DPMI standards docs, I called Phar Lap and they sent them fer free. Nice folks there. cem ------------------------------ From: wrasman@duncan.cs.utk.edu (Aaron 'Raz' Wrasman) Subject: Re: Linux For Mac Date: 11 Oct 1994 13:38:08 GMT In article armb@setanta.demon.co.uk (Alan Braggins) writes: > In article , > Aaron 'Raz' Wrasman wrote: > >Actually could I get some info on Linux for the Mac also? > Whats to tell? The Linux FAQ lists no 68k mac ports, > > Amiga and Atari posrts exist though. > -- > Alan Braggins armb@setanta.demon.co.uk abraggins@cix.compulink.co.uk > "Any technology distinguishable from magic is insufficiently advanced" > But the FAQ is wrong, someone already mentioned that someone was porting it. THeir names were mentioned even. I think they were having problems with I/O because it was hard to get documentation on it. I'll try to go back and grab the message that said their names. I forgot to grab it. Raz -- @ Internet:wrasman@cs.utk.edu, as766@cleveland.freenet.edu (!=>--- Bitnet:Razron@utkvx, Kaspian@utkvx Phone# 615-584-8240 / ) Raz Usnail:602B Longview Rd Knoxville,TN 37919 ------------------------------ From: pda@procyon.com (Patrick D. Ashmore) Subject: Re: Practical experience with Gravis Ultrasound Max? Date: 10 Oct 1994 09:40:02 GMT Jon Lasser (jon.lasser%goucher@wb3ffv.ampr.org) wrote: : Does anyone have practical experience with the Gravis Ultrasound Max : under Linux? I'm currently running Linux 1.1.52 with a Gravis Ultrasound MAX, and it works great. : I have three primary questions: : (1) Does the CD-ROM interface for the Panasonic drive work : properly under Linux (I have a Panasonic 563b drive, which currently : runs with my Sound Blaster 16). I have a Mitsumi 2X, which is also a supported drive for the GUS MAX interface. It works great under Linux and even the Mitsumi's own DOS driver ran fine with the GUS MAX. : (2) Are there any general incompatibilities between the board : and Linux? Any specific known problems? As of 1.1.35 or so, the MAX was included in Hannu's voxware sound driver in the kernel. The first few versions had problems (FTPs would hang while I listened to sounds or mods, and then resume once the music stopped). But that was quickly fixed, and now I sit back and enjoy music from Purple Motion, Skaven, and CC Catch while I hack away. :^) : (3) I'm currently using a Sound Blaster 16. IsGravis : Ultrasound Max considered an upgrade from the Sound Blaster? I mean, I : know that the digital audio is about the same, but is the WaveTable that : much better? Well, I'm biased... I had a SB Pro, but after I played with a friend's GUS I decided that was for me and went ahead and got the GUS MAX. I don't know the specifics on the SB 16, but the fact that the GUS/GUS MAX have on-board memory makes it great for a multitasking OS like Linux. : Frankly, I'm lusting over the board, but I want to make sure that it : works before I buy it, because I don't know enough to hack the kernel. : (yet). I consider it a great investment myself. Patrick -- Patrick D. Ashmore Network Administrator 1322 W. Cleveland pda@procyon.com Procyon Networking Fayetteville, AR 72701 Finger for PGP key URL: http://procyon.com/~pda/ +1 501 443 4393 pda@texas.net pda@sibylline.com publius enigma ------------------------------ From: kuehn@citadel.scd.ucar.edu (Jeff Kuehn) Subject: Re: Linux 1.1.52 (Lies, Damned Lies, and Benchmarks) Date: 11 Oct 1994 14:53:42 GMT Rob Janssen (rob@pe1chl.ampr.org) wrote: : In <37c8hk$ekl@ncar.ucar.edu> kuehn@citadel.scd.ucar.edu (Jeff Kuehn) writes: : >In his last kernel patch announcement, Linus requested information on : >performance of the recent kernels. I took it upon myself to gather a few : >hard numbers in as much as this is possible. (Lies, Damned Lies, and : >Benchmarks!) While I consider myself far too skeptical to call Byte : >Magazine's UNIX Benchmark a "good" benchmark, it seemed like a reasonable : >place to start for several reasons. : I have noticed ever since 0.98pl4 that I started with, that performance : has been dropping. The only exception is the disk performance, which : increased because of the clustering. : In the early days, starting something in X windows was snappy. Now I : get a noticable delay. I still have a copy of Byte-3.1 results of : 0.98pl4 and indeed it shows a decrease since then. At that time, my : system yielded 3.0 and on 1.1.33 it had 2.3 : (not as much decrease as it seems to be. however, I should also note : that my fist reaction when running Linux was "boy this is fast", as my : only frame of reference at that time was SCO XENIX and SCO UNIX running : on comparable hardware. over time, one gets used to the speed and wonders : how it could ever be experienced as "fast" :-) True. Linux looks *damn* good compared to most everything else out there in most regards with two areas excepted: the scheduler and the system call overhead. The reason I chose 1.1.0 as the baseline for these benchmark numbers was to emphasize how performance of linux has changed over several releases... The original baseline was a VAX 11/780, which as you point out, a decent 486 will beat by a factor of 2.5 or 3.0. Again, it should be noted that because the filesystem tests ran out of cache, they are more indicative of ext2fs/cache performance than the device driver performance. The big question is: Can we reclaim the losses? --Jeff Kuehn, NCAR/SCD ------------------------------ From: bass@cais.cais.com (Tim Bass (Network Systems Engineer)) Subject: [Q] what is the story with linux.act.* newsgroups Date: 11 Oct 1994 13:34:44 GMT Twice, I posted some relevant information to the linux.act.* groups. The lastest post was directly related to the linux.act.net group, where I posted the bug fix for lmail.c in the sendmail distribution. None of my posts ever made it to the 'big net in the sky'. As a very active linux person, I'm getting a sinking feeling about the way the linux.act.* groups are being moderated or managed. Before I react, can somewhat send me some info on these groups? Do I need a club membership card to get in the clubhouse? ------------------------------ From: mbi@math.nat.tu-bs.de (Michael Bischoff) Subject: Re: ext2fs vs. Berkeley FFS Date: 10 Oct 1994 09:30:37 GMT Reply-To: mbi@mo.math.nat.tu-bs.de In article adc@bach.coe.neu.edu (Albert D. Cahalan) writes: I think 3 or 4 bits could be spared for extended attributes like this: 0 flat file (default) 1 NEW record file 2 Mac file 3 NT file 4 OS/2 file 5 DOS executable 6 Windows executable 7 etc. Hi, this would be against the philosophy of UNIX: less is more! The success and flexibility of UNIX is due to the fact that all I/O is done through files: ordinary files, hard disks, printers are all accessable through the same open() and write() calls. The mess of DOS is that you need special programs if you want to read A: as entire image. Michael -- EOI ============================================================================ Michael Bischoff e-mail: mbi@mo.math.nat.tu-bs.de Abt. Mathematische Optimierung or m.bischoff@tu-bs.de Inst. Angewandte Mathematik or on.bischoff@zib-berlin.de Pockelsstrasse 14 Tel. +49-531-391-7555, Fax: +49-531-391-4577 38106 Braunschweig Germany ------------------------------ From: davj@ds5000.irb.hr (Davor Jadrijevic) Subject: /proc/mtab 0.2 for 1.1.52 Date: 9 Oct 1994 14:10:39 GMT HI This is a up-to-date /proc/mtab release for kernel 1.1.52 with a minor bug fix and more cleanup. I didn't have tested if it would compile with every kernel config, so please inform me or better send patches. the stuff is uploaded at sunsite under the name procfs-0.2-1.1.52.tgz d. -- , ================ Davor Jadrijevic ==================== ------------------------------ From: kuehn@citadel.scd.ucar.edu (Jeff Kuehn) Subject: Linux 1.1.52 (Lies, Damned Lies, and Benchmarks) Date: 10 Oct 1994 20:35:00 GMT Hi All! In his last kernel patch announcement, Linus requested information on performance of the recent kernels. I took it upon myself to gather a few hard numbers in as much as this is possible. (Lies, Damned Lies, and Benchmarks!) While I consider myself far too skeptical to call Byte Magazine's UNIX Benchmark a "good" benchmark, it seemed like a reasonable place to start for several reasons. 1. it contains tests which are intended to measure the performance of several critical operating system services (although you're welcome to dabate this with the folks at Byte Magazine -- not me). 2. it is widely used to compare performance of small UNIX systems and should a magazine such as Byte ever choose to do a comprehensive review of Linux we can expect that it will include such a benchmark. 3. the benchmark contains several convenient awk scripts for post processing the timings. Which allowed me to select a single patch level as my "base", and report all the number as a ratio to that base performance. The results aren't particularly good. Version 1.1pl0 returned the best overall performance of all the patch levels, so I used it as the base against which I did all of the comparisions. The current patch level (1.1.52) was second to worst of all those I tested, having slipped by over 10% off the best value. For grins, I also ran the same thing on the SPARCstation I have at work for comparison... This pointed out a problem endemic to all of the patch levels I tested (1.0.9 to 1.1.52). Specifically, my Linux box (486dx2/66-256Kb cache, 16MB ram, AHA1742scsi, ext2 filesystems) was consisently faster than the SPARCstation in every area except one: Concurrent users (simulated by running multiple concurrent shell scripts. On SunOS, doubling the number of concurrent scripts, halved the number of loop iterations which could complete within the given period of time. For Linux on the other hand, running one user allowed about 80 iterations to occur within the time limit, whereas when running two users, Linux could not complete a single iteration. Note the SunOS line in the table below... for SunOS Sh1 == 0.2 and Sh2 == 11.3, this is not an indicator of how good SunOS is, but rather how weak Linux is in this area. As several have mentioned previously, this is probably the fault of the process table handling and scheduling algorithms. This needs to be fixed BADLY. Also worthy of note are the following: 1. Process creation performance has dropped 30%. 2. System call overhead has doubled. 3. Filesystem read and write performance has dropped between 20%-40%. Note that since the I/O benchmarks ran almost entirely in cache, I consider this to reflect performance of the ext2 filesystem/buffer cache more than the lower level device drivers, Which (Eric's?) benchmarks have shown to have improved significantly. 4. Pipe overhead performance has dropped 50%. 5. Execl is faster in 1.0.9 6. And of course, the scheduler sucks mud. and on the brighter side: 7. File copy performance is up over 1.1.0 having reached a peak during 1.1.10-1.1.40 and dropped off again after that. 8. Scheduling of single users has improved over 1.1.0, having reached a peak somewhere around 1.1.40 9. While the size of the kernel source has ballooned over the last several months, I'm happy to report that the size of a kernel has changed very little over that period. I think this reflects all of the new drivers rather than a bloating of existing drivers. xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx The headers in the table below refer to the various tests in the Byte Magazine UNIX Benchmark as follows. Versn linux kernel version Ari Arithmetic Test (type=double) Dry Dhrystone 2 w/o register vars Exc Execl Throughput Test FC1 File Copy (10 seconds) FC3 File Copy (30 seconds) FR1 File Read (10 seconds) FR3 File Read (30 seconds) FW1 File Write (10 seconds) FW3 File Write (30 seconds) Pip Pipe-based Context Switch Test PCT Process Creation Test Sh1 Shell scripts (1 concurrent) Sh2 Shell scripts (2 concurrent) Sh4 Shell scripts (4 concurrent) Sh8 Shell scripts (8 concurrent) Sys System Call Overhead Test SUM Sum of all of the previous The numbers in the table, represent the ratio of test(pl)/test(1.1.0), Bigger numbers indicate better performance. xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx Versn |Ari|Dry|Exc|FC1|FC3|FR1|FR3|FW1|FW3|Pip|PCT|Sh1| Sh2|Sh4|Sh8|Sys|SUM 1.0.09|1. |1. |1.1| .9|1. |1. |1. | .9|1. | .5| .7|1. | 1. |1. |1. | .5|14.6 1.1.00|1. |1. |1. |1. |1. |1. |1. |1. |1. |1. |1. |1. | 1. |1. |1. |1. |16. 1.1.01|1. |1. |1. |1. |1. |1. |1. |1. |1. |1. |1. |1. | 1. |1. |1. |1. |15.9 1.1.02|1. |1. |1. | .9|1. |1. |1. |1. |1. | .6| .7|1. | 1. |1. |1. | .5|14.7 1.1.03|1. |1. |1. | .8|1. |1. |1. | .6| .9| .6| .7|1. | 1. |1. |1. | .5|14. 1.1.04|1. |1. | .9| .8|1. |1. |1. | .6| .9| .9|1. | .9| 1. |1. |1. |1. |15.1 1.1.05|1. |1. | .9| .8|1. |1. |1. | .6| .9| .8|1. |1. | 1. |1. |1. |1. |15.1 1.1.10|1. |1. | .9|1.1|1.3|1. |1. | .6| .9| .9|1. |1. | 1. |1. |1. |1. |15.6 1.1.15|1. |1. | .9|1.1|1.3|1. |1. | .6| .9| .9|1. |1. | 1. |1. |1. |1. |15.6 1.1.20|1. |1. | .9|1.1|1.3| .8| .9| .6| .8| .8|1. |1. | 1. |1. |1. |1. |15.1 1.1.25|1. |1. | .9|1.1|1.3| .9| .9| .6| .9| .9|1. |1. | 1. |1. |1. |1. |15.4 1.1.30|1. |1. | .9|1.1|1.3| .9| .9| .6| .9| .7|1. |1. | 1. |1. |1. |1. |15.3 1.1.35|1. |1. | .9|1.1|1.3| .9| .9| .6| .9| .9|1. |1. | 1. |1. |1. |1. |15.4 1.1.40|1. |1. |1. |1.1|1.3| .8| .8| .6| .8| .7|1. |1.2| 1. |1. |1. | .9|15.1 1.1.45|1. |1. |1. |1.1|1.3| .9| .9| .6| .9| .8|1. |1.1| 1. |1. |1. |1. |15.4 1.1.50|1. |1. |1. |1.1|1.2| .8| .9| .6| .9| .7|1. |1.1| 1. |1. |1. | .9|15.3 1.1.51|1. |1. |1. |1.1|1.2| .8| .8| .6| .9| .7|1. |1.1| 1. |1. |1. | .9|15.2 1.1.52|1. |1. |1. |1.1|1.2| .8| .8| .6| .8| .5| .7|1.1| 1. |1. |1. | .5|14.1 SunOS | .5| .4| .2| .8|1.1| .1| .1| .1| .1| .2| .4| .2|11.3|5.9|3. | .2|24.4 Here's the run for SunOS (result and index column) as compared to Linux 1.1.0 (the baseline column) (note, lps==loops/sec, lpm==loops/min) in all cases bigger numbers indicate better performance). xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx BYTE UNIX Benchmarks (Version 3.11) System -- citadel (SunOS) Start Benchmark Run: Mon Oct 10 10:58:32 MDT 1994 1 interactive users. Arithmetic Test (type = double) 2508.2 lps (10 secs, 6 samples) Dhrystone 2 without register variables 23116.3 lps (10 secs, 6 samples) Execl Throughput Test 9.2 lps (9 secs, 6 samples) File Read (10 seconds) 7472.0 KBps (10 secs, 6 samples) File Write (10 seconds) 1200.0 KBps (10 secs, 6 samples) File Copy (10 seconds) 1003.0 KBps (10 secs, 6 samples) File Read (30 seconds) 7649.0 KBps (30 secs, 6 samples) File Write (30 seconds) 1133.0 KBps (30 secs, 6 samples) File Copy (30 seconds) 997.0 KBps (30 secs, 6 samples) Shell scripts (1 concurrent) 19.6 lpm (60 secs, 3 samples) Shell scripts (2 concurrent) 11.3 lpm (60 secs, 3 samples) Shell scripts (4 concurrent) 5.9 lpm (60 secs, 3 samples) Shell scripts (8 concurrent) 3.0 lpm (60 secs, 3 samples) System Call Overhead Test 5219.9 lps (10 secs, 6 samples) Pipe-based Context Switching Test 1334.7 lps (10 secs, 6 samples) Process Creation Test 48.3 lps (10 secs, 6 samples) INDEX VALUES TEST BASELINE RESULT INDEX Arithmetic Test (type = double) 5069.5 2508.2 0.5 Dhrystone 2 without register variables 56103.3 23116.3 0.4 Execl Throughput Test 60.8 9.2 0.2 File Copy (10 seconds) 1310.0 1003.0 0.8 File Copy (30 seconds) 919.0 997.0 1.1 File Read (10 seconds) 117181.0 7472.0 0.1 File Read (30 seconds) 117335.0 7649.0 0.1 File Write (10 seconds) 13856.0 1200.0 0.1 File Write (30 seconds) 13643.0 1133.0 0.1 Pipe-based Context Switching Test 8197.6 1334.7 0.2 Process Creation Test 112.1 48.3 0.4 Shell scripts (1 concurrent) 81.1 19.6 0.2 Shell scripts (2 concurrent) 1.0 11.3 11.3 Shell scripts (4 concurrent) 1.0 5.9 5.9 Shell scripts (8 concurrent) 1.0 3.0 3.0 System Call Overhead Test 31574.7 5219.9 0.2 ========= SUM of 16 items 24.4 AVERAGE 1.5 xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx Final Notes: The Arithmetic and Dhrystone Benchmarks were included both as a sanity check and to provide a basis for speed comparison with the SPARCstation IPC running SunOS 4.1.3U1 shown in the last line. The remaining tests were included as the member-tests of the Byte benchmark which focused on critical system services, with the intention of providing an initial guess of how kernel performance has varied over several levels of revision. You'll note that I haven't run the Benchmark for every patch level... Each run takes about an hour for good results. If you ask nicely (and I don't get flamed too heavily for posting "Lies, Damned Lies, and Benchmarks") I'll run additional patch levels on request. If there are other benchmarks you want me to run, go ahead and send them... comparing results will be easier if we run everything on one machine whenever possible. If you want to try the Byte benchmark suite yourself, you can pick it up off ftp.uu.net. Jeff Kuehn Scientific Computing Division National Center For Atmospheric Research Boulder, Colorado USA ------------------------------ ** FOR YOUR REFERENCE ** The service address, to which questions about the list itself and requests to be added to or deleted from it should be directed, is: Internet: Linux-Development-Request@NEWS-DIGESTS.MIT.EDU You can send mail to the entire list (and comp.os.linux.development) via: Internet: Linux-Development@NEWS-DIGESTS.MIT.EDU Linux may be obtained via one of these FTP sites: nic.funet.fi pub/OS/Linux tsx-11.mit.edu pub/linux sunsite.unc.edu pub/Linux End of Linux-Development Digest ****************************** .