Out of curiosity, I performed a few tests to see how AOS 4.1 compares to AOS 4.0 (on the same machine with the same settings, of course). The conclusion is that AOS 4.1 is (slightly) slower almost always, although an important aspect like system memory to video memory transfers seem to have improved noticeably. Note that these tests do not prove AOS 4.1 is slower than AOS 4.0 in absolute: they cover only some aspects in a idle enviroment, so it could well be that the overall performance during normal usage is better (f.ex. thanks to the composition engine moving some of the load from the CPU to the GPU). Moreover, let's not forget that AOS 4.1 has some added functionality that may come at a cost
****************************************************************************** * DiskSpeed DRIVE=HD3: ALL LONG * (SFS partition)
* 4.0
DiskSpeed 4.3, OS4 version Copyright ? 1989-92 MKSoft Development Copyright ? 2003-04 Daniel J. Andrea II & St?phane Guillard ------------------------------------------------------------ CPU: 68020 AmigaOS Version: 52.8 Normal Video DMA Device: HD3: Buffers: 1000
Testing with a 512 byte, MEMF_FAST, LONG-aligned buffer. Create file: 13205184 bytes/sec Write to file: 54920384 bytes/sec Read from file: 79091264 bytes/sec
Testing with a 4096 byte, MEMF_FAST, LONG-aligned buffer. Create file: 24080896 bytes/sec Write to file: 374712832 bytes/sec Read from file: 278949888 bytes/sec
Testing with a 32768 byte, MEMF_FAST, LONG-aligned buffer. Create file: 28024832 bytes/sec Write to file: 194387968 bytes/sec Read from file: 233078784 bytes/sec
Testing with a 262144 byte, MEMF_FAST, LONG-aligned buffer. Create file: 27590656 bytes/sec Write to file: 42991616 bytes/sec Read from file: 230522880 bytes/sec
* 4.1
DiskSpeed 4.3, OS4 version Copyright ? 1989-92 MKSoft Development Copyright ? 2003-04 Daniel J. Andrea II & St?phane Guillard ------------------------------------------------------------ CPU: 68020 AmigaOS Version: 53.5 Normal Video DMA Device: HD3: Buffers: 1000
Testing with a 512 byte, MEMF_FAST, LONG-aligned buffer. Create file: 12938304 bytes/sec Write to file: 43209233 bytes/sec Read from file: 70056576 bytes/sec
Testing with a 4096 byte, MEMF_FAST, LONG-aligned buffer. Create file: 23139328 bytes/sec Write to file: 250321408 bytes/sec Read from file: 238501888 bytes/sec
Testing with a 32768 byte, MEMF_FAST, LONG-aligned buffer. Create file: 26972160 bytes/sec Write to file: 103411712 bytes/sec Read from file: 174030848 bytes/sec
Testing with a 262144 byte, MEMF_FAST, LONG-aligned buffer. Create file: 5570560 bytes/sec Write to file: 5832704 bytes/sec Read from file: 5734400 bytes/sec
(side note: I don't have the results anymore nor a partition to repeat the test, but JXFS turned out to be slightly slower).
******************************************************************************* * Custom program that renders stuff in system memory and then transfers it to * video memory for showing in a window on Workbench.
* 4.0
tiles rendering speed: 461.095101 fps, 2.168750 ms/loop bobs rendering speed: 288.548242 fps, 3.465625 ms/loop video data transfer speed: 107.346528 fps, 9.315625 ms/loop total speed: 66.861680 fps
* 4.1 - composition ON
tiles rendering speed: 468.864469 fps, 2.132812 ms/loop bobs rendering speed: 259.109312 fps, 3.859375 ms/loop video data transfer speed: 131.497843 fps, 7.604688 ms/loop total speed: 73.520965 fps
* 4.1 - composition OFF
tiles rendering speed: 468.864469 fps, 2.132812 ms/loop bobs rendering speed: 277.898393 fps, 3.598438 ms/loop video data transfer speed: 116.257947 fps, 8.601562 ms/loop total speed: 69.754768 fps
****************************************************************************** * A script opens and closes 2 windows 1000 times.
* 4.0 / 4.1 - composition ON / 4.1 - composition OFF
Seeing how much OWB seems to have benefitted from today's AOS 4.1 QuickFix, I thought I'd have a go at some of the tests in the above post. The results are as follows:
******************************************************************************
* DiskSpeed DRIVE=HD3: ALL LONG
* (SFS partition)
DiskSpeed 4.3, OS4 version
Copyright ? 1989-92 MKSoft Development
Copyright ? 2003-04 Daniel J. Andrea II & St?phane Guillard
------------------------------------------------------------
CPU: 68020 AmigaOS Version: 53.8 Normal Video DMA
Device: hd3: Buffers: 1000
****************************************************************************** * DiskSpeed DRIVE=HD3: ALL LONG * (SFS partition)
DiskSpeed 4.3, OS4 version Copyright ? 1989-92 MKSoft Development Copyright ? 2003-04 Daniel J. Andrea II & St?phane Guillard ------------------------------------------------------------ CPU: 68020 AmigaOS Version: 53.12 Normal Video DMA Device: HD3: Buffers: 1000
Testing with a 512 byte, MEMF_FAST, LONG-aligned buffer. Create file: 12270080 bytes/sec Write to file: 26468736 bytes/sec Read from file: 66065856 bytes/sec
Testing with a 4096 byte, MEMF_FAST, LONG-aligned buffer. Create file: 23381504 bytes/sec Write to file: 240470016 bytes/sec Read from file: 239370752 bytes/sec
Testing with a 32768 byte, MEMF_FAST, LONG-aligned buffer. Create file: 23715840 bytes/sec Write to file: 156913664 bytes/sec Read from file: 187850752 bytes/sec
Testing with a 262144 byte, MEMF_FAST, LONG-aligned buffer. Create file: 30900224 bytes/sec Write to file: 39616512 bytes/sec Read from file: 31358976 bytes/sec
Not much changed. The system seems to be snappier, though, so maybe memory allocation/deallocation has improved.
EDIT
I gave a spin to SFS2 and JFXS, too.
Quote:
****************************************************************************** * DiskSpeed DRIVE=HD3: ALL LONG * (SFS2 partition)
DiskSpeed 4.3, OS4 version Copyright ? 1989-92 MKSoft Development Copyright ? 2003-04 Daniel J. Andrea II & St?phane Guillard ------------------------------------------------------------ CPU: 68020 AmigaOS Version: 53.12 Normal Video DMA Device: dh0: Buffers: 2000
Testing with a 512 byte, MEMF_FAST, LONG-aligned buffer. Create file: 12184512 bytes/sec Write to file: 28714364 bytes/sec Read from file: 71491392 bytes/sec
Testing with a 4096 byte, MEMF_FAST, LONG-aligned buffer. Create file: 22918656 bytes/sec Write to file: 221022720 bytes/sec Read from file: 239467520 bytes/sec
Testing with a 32768 byte, MEMF_FAST, LONG-aligned buffer. Create file: 24588288 bytes/sec Write to file: 61997056 bytes/sec Read from file: 189485056 bytes/sec
Testing with a 262144 byte, MEMF_FAST, LONG-aligned buffer. Create file: 23363584 bytes/sec Write to file: 32309248 bytes/sec Read from file: 23035904 bytes/sec
****************************************************************************** * DiskSpeed DRIVE=HD3: ALL LONG * (JXFS partition)
DiskSpeed 4.3, OS4 version Copyright ? 1989-92 MKSoft Development Copyright ? 2003-04 Daniel J. Andrea II & St?phane Guillard ------------------------------------------------------------ CPU: 68020 AmigaOS Version: 53.12 Normal Video DMA Device: dh0: Buffers: 2000
Testing with a 512 byte, MEMF_FAST, LONG-aligned buffer. Create file: 7174848 bytes/sec Write to file: 11622753 bytes/sec Read from file: 13663552 bytes/sec
Testing with a 4096 byte, MEMF_FAST, LONG-aligned buffer. Create file: 20091392 bytes/sec Write to file: 76486144 bytes/sec Read from file: 87129088 bytes/sec
Testing with a 32768 byte, MEMF_FAST, LONG-aligned buffer. Create file: 23171072 bytes/sec Write to file: 371011584 bytes/sec Read from file: 165412864 bytes/sec
Testing with a 262144 byte, MEMF_FAST, LONG-aligned buffer. Create file: 30375936 bytes/sec Write to file: 36274176 bytes/sec Read from file: 28606464 bytes/sec
Verdict: while memory access is only slightly slower, the disk access performance has decreased sensibly. Since the filesystem module has remained the same, I guess the performance loss is in the new dos.library (and, maybe, sii0680ide.device).
Well, since this is my first (and probably only) report about Update 3, I'll add here that the system seems to be very stable (never crashed until now).
saimo wrote: Returned home after more than 6 months. Switched the A1 on. Booted perfectly. Installed Update 3.]
Verdict: while memory access is only slightly slower, the disk access performance has decreased sensibly. Since the filesystem module has remained the same, I guess the performance loss is in the new dos.library (and, maybe, sii0680ide.device).
Well, since this is my first (and probably only) report about Update 3, I'll add here that the system seems to be very stable (never crashed until now).
thanks for the update Simone. I'm hoping to try out update 3 soon. btw, how's the new job & new town?
It would be interesting to see your tests for SFS2/JXFS on 4.1.3 for comparison.
OK, I made a test with JXFS. And, unfortunately, the figures tell that the performance (with respect to the previous version of the same filesystem, of course) has dropped dramatically...
Quote:
****************************************************************************** * DiskSpeed DRIVE=HD3: ALL LONG * (JXFS partition)
It's quite a while since I made the last test (under AOS 4.1 u3!), but here are now the results under AOS 4.1 FE. After the last test I have installed an SSD, so the drive results can't be compared, but it was interesting to discover that, contrary to the official suggestion, a block size of 512 b gives a better performance than a block size of 4096 b with SFS0 (on the very same partition). Memory speed seems to have decreased a little. I don't have figures relative to AOS 4.1 u6, but the feeling is that the system is smoother and snappier.
Quote:
****************************************************************************** * DiskSpeed DRIVE=SSD0: ALL LONG (4096 b block size)