This is a freeware benchmark for OS/2. Like all such benchmark programs, its results can be read in many ways. Beware of making too many decisions based on its output - by their very nature, benchmark programs are *not* real world applications. By tuning your system with the aid of benchmarks, you may end up with a system that runs benchmark programs wonderfully well but runs applications badly! Just so you're aware....
This program is distributed complete with all source needed to compile it. The source was originally written by Henrik Harmsen and was donated to the OS/2 community by him when he moved off to use Linux. The source has undergone many revisions, bugfixes and has a number of new functions since that time. Please do not report problems to him as he hasn't touched the code for a long time. The version of the source you now have was coded by me, Trevor Hemsley, and you should report any and all problems to me (Trevor-Hemsley@dial.pipex.com).
Disclaimer
Standard disclaimers apply. This program is supplied in good faith but I am not responsible for any loss or damage that may result from use of it. All use of this program is strictly at your own risk.
Changes in Sysbench 0.9.1c
1) fixed problems with disks larger than 2Gb whose size didn't display correctly when the number hit 2Gb and then went negative. When I acquired a 4.3Gb disk, I rewrote this fix again as it still didn't work correctly. It should now handle disks up to the terrabyte range so it should last a couple of years at least :-)
2) Added a new suite of disk tests. The existing ones perform i/o directly to the disk bypassing any and all operating system caching. I've now added a longish set of tests that read and write both sequentially and randomly at varying blocksizes with caching enabled and disabled. The idea of these new tests is so that you can gain some idea about the performance of your machine using different cache sizes and parameters. Compare and contrast the results obtained after adjusting things and see if they've made it better or worse. The new tests each run for 15 seconds and there are 40 of them so they add quite a bit to the elapsed time of the benchmark. This benchmark isn't really designed to be used to compare different systems, more so you can see the effect that tuning the file system cache has on your own.
3) If you are on a network and have remote drives mapped the startup of Sysbench may take some time. It queries all attached drives and checks them to see if the filesystem is eligible for the new file tests. If your LAN connections have timed out it has to wait for OS/2 to re-make them before it can continue. If the host machine is down the connection attempt needs to time out. I don't know if I can make this wait-time any shorter since, as far as I am aware, the only way to find out if it's a LAN drive is to query it (catch-22). What I've done instead is to put up a nice little dialog window with a flashing LED on it to keep you all amused while it does its stuff ;-). You'll also see this if you have a drive without a CD in it.
4) Added a bunch of new code that monitors swapper.dat during the benchmark. It also looks at various settings in CONFIG.SYS and recommends changes that you might want to make. It looks at speed of processor and MAXWAIT, size of RAM and diskcaches and how much space in each file systtem you have. If you quit the application using File/Quit rather than Alt+F4 or the Warp 4 Xit box you'll see these recommendations if any are to be made. If you use Alt+F4 or the exit button then you will not see these recommendations (such as they are) at all. I've deliberately coded it this way so that there's a fast exit way out of the app if you want it.
5) Added an option to the "File" menu to allow you to specify various data elements that will appear in the saved results. This includes machine name, motherboard, chipset, manufacturer, cache, processor, graphics card and disk controller. You can use up to 100 characters in each of these fields, freeform text. This data is written to a SYSBENCH.INI file and remembered for the next time you run the program. If the file doesn't exist, Sysbench will display the dialog window automatically when you start it. Subsequent runs should be as normal.
6) Added a couple of command line options for the program so you can run the suite of tests without intervention. These are
/ALL - run all tests and then quit saving the results to the current
directory as RESULT.TXT
/R:filename - specify the filename to which to save results. It must specify
a pre-existing directory if a directory name is given.
Running
SYSBENCH /all /r:y:\sysbench\res091c.686
will run all tests and save the results to the file \sysbench\res091c.686 on the Y: drive.
7) Added "mnemonic" keys to all menu options so that you can now access the entire range using the keyboard only.
8) Corrected the amount of installed RAM which was under reported by 384Kb.
Results from this release should be comparable to those from any release back to 0.9.1.
Sysbench 0.9.1b
V0.9.1b: fixed a display problem when scrolling through the window using the arrow keys on some machines. Symptoms were repeated lines and shadow data.
Fixed various divide by zero problems that hit the Linpack CPU benchmark on fast machines.
Fixed DIVE problem at 800x600 on video cards that use bank switching. This caused the display to be garbage on first and third tests and it to disappear with a SYS3175 during the second DIVE test.
Disabled DIVE tests on releases of OS/2 prior to Warp as reports indicate it fails or crashes the machine on 2.11.
Added information to printouts to identify more about the machine being tested. This adds swapfile current and initial sizes and video resolution/colour depths to the current RAM size, priority, maxwait and co-processor listings. Also listed some data about video characteristics that will aid debugging any future video problems.
Changed coding in all files to eliminate compilation warnings and informational messages. The whole suite now compiles cleanly using IBM's VisualAge C++ 3.0 product.
Sysbench 0.9.1a
Ooops. Recompiled with the GA version of VisualAge C++ with all the latest fixes on and finally found the switch to turn off debug mode! (Duh!). Cuts the size of the EXE in half and then some and pushes the CPU and memory benchmarks up by 40%.
Corrected bug that causes a SYS3175 on machines with video memory that don't have enough space to fit the same number of vertical lines as horizontal lines. Machines at 800x600x64K with only 1Mb would hit this - maybe others too.
Sysbench 0.9.1
This is a revision of the Sysbench 0.9.0 program, originally written by Henrik Harmsen. His introductory readme file has been renamed to README.HCH in this package and needs to be read in conjunction with _this_ file.
Warning - Warning - Warning.
You should NOT compare results produced using this revision of the program with results produced from versions prior to 0.9.1. Too many things have changed - timing bugs have been eliminated, methods of calculating test averages have changed and a different compiler has been used.
I would also suggest (very strongly) that this benchmark program NOT be run on any machine that has critical applications running on it. It sets itself to be the highest priority thing on your system at points in its execution and this can adversely affect other processes. It will probably stop any communications apps you have running and connections may be lost.
Installation
No special installation steps are required to install this program.
I have included the latest DIVE.DLL from The Developer Connection for OS/2 volume 10. It may not be necessary as the program appears to work with the version that's on my system dated 17-02-95 4:51p. I have included the latest version anyway as that's the one I've done all my testing with. If you want to use this, copy it to your \MMOS2\DLL directory _after_ saving a copy of the existing version. This copy is *older* than that shipped in Warp 4 so don't use it for this release of OS/2. It's also possible that IBM have shipped a newer copy with a fixpack so check the date and time of your file before replacing it.
Bug fixes from original code
I have revised various routines to fix some bugs, notably some of the CPU benchmarks returned impossible values on very fast machines as some of the passes through the code completed before the timer could register a value.... dividing by zero gives some very strange results! The memory tests also gave invalid values on some of the smaller values.
Now uses DISKIO 1.9 code
In addition, I have incorporated the latest DiskIO benchmark code (1.9) into this release and have modified the display to include some of the new values reported by this. I have doubts about my implementation of the concurrent disk i/o and CPU test as it returns results sometimes that are wildly out of kilter with the norm. Treat this one with suspicion for the present ;-). Thanks go to Kai Uwe Rommel for this code - bugs are mine<g>.
DIVE Tests now enabled
I have enabled the DIVE tests and checked them out on both my machines. I have written a lot of code for bank-switched graphics cards and it seems to work fine.
In case it's not immediately obvious what these tests look like in use, the first should move the entire screen image downwards with the image that disappears from the bottom of the screen re-appearing from the top. The movement should be quite smooth. The second test squashes the screen image from top and bottom and eventually flips it over completely upside-down, then back up the right way and finally stretches and compresses various parts of the image. If you see something radically different from this, I either haven't described it very well or you have a bug.
I am not sure what Henrik's intent was when he wrote the original DIVE programs but I have eliminated two of the tests from the program as they seemed to be identical to each other. As it is, two of the tests actually do identical things (the first and third) but report their values in different ways, one as MB/second and one as frames/second. I've left them both there as this seemed quite useful. I didn't think it useful to run the same test four times...
To run the DIVE tests you must have a graphics card that's enabled for it. My ATI GPT does not support DIVE when running at 1024x768x16.7M colours - it's necessary to reset to a lower colour depth in order to run the DIVE tests. If you don't have a card that supports DIVE, the tests will just blink the screen momentarily and the test will be skipped.
Sample result files
I have included sample result files from my two machines in this file. All results on the smaller machine were generated using OS/2 Warp Connect V3 as I do not have MMOS/2 installed on the copy of V4 that is on this machine. MMOS/2 is required for the DIVE tests.
Res091.486 is generated using Sysbench 0.9.1 on my Dell 486DX2/66, 24MB RAM, with Diamond Stealth32 2MB VLB graphics card and Promise DC4030-VL2 4MB cache IDE controller, Western Digital AC31200 1.2Gb hard disk.
Res091b.486 is generated using Sysbench 0.9.1b on my Dell 486DX2/66, 24MB RAM, with Diamond Stealth32 2MB VLB graphics card and Promise DC4030-VL2 4MB cache IDE controller, Western Digital AC31200 1.2Gb hard disk.
Res091c.686 is generated using Sysbench 0.9.1c on my Dell machine which now consists only of the original case with completely new innards! This is a Freetech 586F62 motherboard based on the Intel 430HX chipset, Cyrix P120+ running at 100MHz, 40MB RAM, with no-name S3Trio64V+ 2MB PCI graphics card, Western Digital AC31200 1.2Gb hard disk attached to the onboard PCI EIDE controller.
Res091c1.686 - same system as above, running Warp 3 with an ATI Graphics Xpression 2Mb card.
Res091c2.686 - same system as above, running Warp 4 with an ATI Graphics Xpression 2Mb card.
Res091.PRO is generated using Sysbench 0.9.1 on a Micron Pentium Pro 200MHz, 64MB RAM, with ATI Graphics Pro Turbo 4MB VRAM, Adaptec AHA2940UW PCI SCSI controller, Seagate Barracuda 2.1Gb fast/wide hard disk (ST32550W). This set of results uses the Conner 1Gb narrow SCSI disk that's also installed.
Res091b.PRO is generated using Sysbench 0.9.1b on a Micron Pentium Pro 200MHz, 128MB RAM, with ATI Graphics Pro Turbo 4MB VRAM, Adaptec AHA2940UW PCI SCSI controller, Seagate Barracuda 2.1Gb fast/wide hard disk (ST32550W).
Res091c.PRO is generated using Sysbench 0.9.1c on a Micron Pentium Pro 200MHz, 128MB RAM, with ATI Graphics Pro Turbo 4MB VRAM, Adaptec AHA2940UW PCI SCSI controller, Seagate Barracuda 2.1Gb fast/wide hard disk (ST32550W) and IBM DCHS-04W Scorpion, fast/wide 4.3Gb hard disk (the subject of this test).
Bugs.
None that I know of (but there are bound to be some). If you find one, report it to me - Trevor-Hemsley@dial.pipex.com - and include as much information as you can. If it falls over and displays an error box with registers, quote them (a HELP SUPPRESSPOPUPS on Warp will tell you an easy way of collecting this information and save you some tedious typing). If the program generates a log file called SB_ERROR.LOG, include that too. Tell me about your machine in as much detail as you can, even if it seems irrelevant
As of 0.9.1b there's a section in the results file that reports on video data. If you suffer a video problem please save a set of results (even if they're blank) and include the video data section in your bug report. It'll tell me more about your video card and might help to solve the problem.
