IDE (and adaptor-ed) devices benchmarks / remarks

[Rybags]

I've started this so we can put our benchmarks and comments relating to real HDDs and "adaptor-ed" devices like CF and SD cards interfaced via IDE Plus or other IDE interfaces.

 

Use the RWTEST.COM program to test your speeds. Note that 512 byte/sec partitions should be quicker than 256 or 128 byte ones.

Also, you'll generally get a speedup with the screen turned off.

 

To assist the benchmarking with the screen turned off, making up a quick Batch file can help. I just called mine "RWT.BAT". You can create it using the "ED" command that shouuld be resident within SpartaDos 4.44

 

POKE 559,0
RWTEST %1
POKE 559,34

 

Just enter that into the window that ED brings up, remember press RETURN after the second POKE command. Then CTRL-S to Save and call it "RWT.BAT" (or whatever).

 

Then, to run a benchmark of each type:

-RWT D2:

RWTEST D2:

 

Note the "-" prefix to tell DOS to run the batch file.

 

I've just started on a second SD card, and have another to try after that. Strange results on this one. This is with IDE Plus 2, 0.7 BIOS, SDX 4.44. Currently the only media I have available is SD and Micro SD cards, used with a cheap IDE to SD adaptor.

 

Digitech 2 GB SD Card (generic yumcha thing, just bought for $8 )

 

Partition 1 (256 B sectors)

Screen on

DOS write: 13,904.8

DOS Read: 37,995.6

DOS average: 25,950.2

 

Screen off

DOS write: 13,729.5

DOS read: 48,053.3

DOS average: 30,891.4

 

Partition 2 (512 B sectors)

Screen on

DOS write: 21,784.2

DOS read: 58,350.4

DOS avg: 40,067.3

 

Screen off

DOS write: 24,204.6

DOS read: 75,991.3

DOS average: 50,098

 

Comment: I'm sure this is substantially slower than the first SD card I tried. Will get back shortly with it's results. Weird how on the 256 byte partition that it actually wrote slower with screen DMA off... I ran it several times and got much the same results.

Edited July 5, 2011 by Rybags

[Rybags]

Verbatim 2 Gig Micro SD housed in provided full-size SD adaptor

 

Partition 1 (256 B / sect)

Screen on

DOS write: 35,135.8

DOS read: 41,892.6

DOS avg: 38,514.2

 

Screen off

DOS write: 69,523.9

DOS read: 81,690.6

DOS avg: 75, 607.3

 

Partition 2 (512 B / sect)

Screen on

DOS write: 52,703.6

DOS read: 61,653

DOS avg: 57,178.5

 

Screen off

DOS write: 68,075.5

DOS read: 81,690.6

DOS avg: 74, 883.1

[GoodByteXL]

Toshiba HD MK8032GAX

 

SpartaDOS File System 2.1

 

Partition 1 (256 B sectors)

 

Screen on

DOS write: 37995.639

DOS Read: 45383.6799

DOS average: 41689.6594

 

Screen off

DOS write: 51867.0628

DOS read: 61653.301

DOS average: 56760.1819

 

Partition 2 (512 B sectors)

 

Screen on

DOS write: 55383.4738

DOS read: 66686.2236

DOS avg: 61034.8487

 

Screen off

DOS write: 74264.2036

DOS read: 90767.3599

DOS average: 82515.7817

 

Dramatically slow on MyDOS, of course

 

Besides these practically usable performance data you might get more excited when testing with Draco's SysInfo

Edited July 5, 2011 by GoodByteXL

[Rybags]

Sandisk 2 Gig SD (fullsize)

 

Part. 1 (256 B sec)

Screen on

write: 34,762 read: 42,995 avg: 38,879

 

Screen off

write: 44,762 read: 57,327 avg: 51,044

 

Part. 2 (512 B sec)

Screen on

write: 50,271 read: 62,839 avg: 56,555

 

Screen off

write: 66,686 read: 83,785 avg: 75,236

[flashjazzcat]

SIDE cart, SDX 4.44, test driver, Sandisk II Ultra 1GB CF card:

 

Partition 1 (256bps):

 

Screen on

 

DOS write: 40845.312

DOS Read: 55383.4738

DOS average: 48114.3929

 

Screen off

 

DOS write: 55383.4738

DOS read: 74264.2036

DOS average: 64823.8387

 

Partition 2 (512bps)

 

Screen on

 

DOS write: 59411.3628

DOS read: 72613.8879

DOS avg: 66012.6253

 

Screen off

 

DOS write: 79698.1697

DOS read: 96106.6164

DOS average: 87902.393

 

[drac030]

That's a great pity I hadn't this idea earlier ... I can make transfers ~10% faster, but unfortunately only breaking the compatibility with IDEa and the current IDE Plus BIOS-es. So it is probably not worth it.

Edited July 5, 2011 by drac030

[+orpheuswaking]

10% isn't worth it, I'll just drink a cup of coffee while I wait

[flashjazzcat]

That's a great pity I hadn't this idea earlier ... I can make transfers ~10% faster, but unfortunately only breaking the compatibility with IDEa and the current IDE Plus BIOS-es. So it is probably not worth it.

Not worth it. Just accept second-place gracefully.

 

Storing pairs of even/odd bytes from the sector buffer would really save some time, but I haven't done that.

Edited July 5, 2011 by flashjazzcat

[Frankie]

Fujitsu 40 GB 2.5" HDD

 

512 Bytes/sector

 

DOS writing: 74092 B/sek

DOS reading: 91323 B/sek

DOS average: 82708 B/sek

 

-- just for fun, Transcend 1 GB 40-pin Flash module using external MyIDE card + SDX 4.44

 

DOS writing: 72720 B/sek

DOS reading: 93498 B/sek

DOS average: 83109 B/sek

[UNIXcoffee928]

I wouldn't bother messing around with low-end SD cards, if I was looking for sheer performance numbers... Get a Sandisk Extreme SDHC, to ensure that you have the lowest on-device latency, then perform your tests.

 

I have read that your block-size and partitioning scheme can severely effect both your device's performance and your device's reliability. There is an excellent overview here, where electrical engineers are discussing card-based filesystem tuning with other technical users. You will certainly learn all of the tricks, if you read that whole thread.

[+Stephen]

Transcend 2GB CF card, not sure of the adapter I am using (got it from Mech). With DMA off and 512GB sectors via SDX 4.44, I get 76998.375 Write, 93498.03 Read, 85248.2 Average. DMA on, 53066.45 Write, 65448.62 Read, 59257.53 Average.

[drac030]

Not worth it. Just accept second-place gracefully.

 

Well, I took a look at this and there indeed were places where my I/O routines could be improved a bit. So the next IDE Plus BIOS will be faster But in fact I am now focusing my attention on other functionalities.

[flashjazzcat]

Well, I took a look at this and there indeed were places where my I/O routines could be improved a bit. So the next IDE Plus BIOS will be faster But in fact I am now focusing my attention on other functionalities.

Your software is of such high quality and so consistently innovative that I don't think 3-4KB/s in either direction is a subject for concern. Of course it's always nice when we have time to optimize and have a bit of fun trying to score high benchmarks. However, efficiency has to be (certainly in the case of SIDE) balanced against the code footprint. Although I'm still using the 8 in-line reads/writes that the MyIDE driver used, I'm currently using some techniques which are the preserve of a RAM-based driver, such as self-modifying code (although there's no guarantee that this will make the cut for the final driver). In addition, I no longer have to worry about the slave device. And on top of all this, I can run the 256 byte sectors in 16 bit mode with the 8-bit data register, which removes the need for any sector padding.

[+Stephen]

I want to see who will be the first to break the 100K/sec barrier (DMA off of course)

[+bf2k+]

Western Digital WD600VE Hard Disk on IDE + 2.0

60gb 2.5" 0.5 amps

 

Partition 1 - 256b

-------------------------

Screen ON

Writ: 37399.2106

Read: 45661.8269

Avg: 41530.5187

 

Screen OFF

Writ: 53793.3851

Read: 65448.6186

Avg: 59621.0018

 

Partition 2 - 512b

-------------------------

Screen ON

Writ: 53066.4475

Read: 64375.6904

Avg: 58721.0689

 

Screen OFF

Writ: 78538.3424

Read: 93498.0266

Avg: 86018.1845

[GoodByteXL]

I want to see who will be the first to break the 100K/sec barrier (DMA off of course)

if it's just for the figures - I did -> 112.

 

Use Draco's SysInfo Tool. http://drac030.krap.pl/en-si-info.php

[GoodByteXL]

Something to compare with ...

 

MSC IDE Controller on PBI with

 

no name SD-to-IDE-Adaptor (master only)

 

2GB SanDisk SD-Card

 

CHS-mode only

 

SpartaDOS File System 2.1

 

Partition 1 (256 B sectors)

 

Screen on

DOS write: 14785.6332

DOS Read: 28169.1806

DOS average: 21477.4069

 

Screen off

DOS write: 19684.4876

DOS read: 37558.9075

DOS average: 28621.6975

 

Partition 2 (512 B sectors)

no 512BPS mode possible

 

240 (master & slave with hd or CF card) partitions of 16MB each at max.

[Rybags]

I was thinking earlier maybe it'd have been an idea to do PC-based tests on the media before using it on the Atari, e.g. HD Tach.

 

I was expecting maybe a little less margin among the SD types and more margin between real HDD and flash.

 

Still waiting for my CF adaptor to arrive, once it lands I'll be able to get the new cheapie CF card out and try it.

[+bob1200xl]

Most any hardware storage device will probably have a cache in it, so you may not be measuring the native speed of the unit. You want to see a CF card go nuts, format it on an old PC where it writes to every sector. I did that once... never again.

 

Bob

 

 

I was thinking earlier maybe it'd have been an idea to do PC-based tests on the media before using it on the Atari, e.g. HD Tach.

 

I was expecting maybe a little less margin among the SD types and more margin between real HDD and flash.

 

Still waiting for my CF adaptor to arrive, once it lands I'll be able to get the new cheapie CF card out and try it.

[GoodByteXL]

Something to compare with ...

 

More to compare with ...

 

MSC IDE Controller on PBI with

 

no name CF-to-IDE-Adaptor (master & slave)

 

CHS-mode only

 

SpartaDOS File System 2.1

 

4GB Noname CF (256 B sectors)

 

Screen on

DOS write: 13447.0162

DOS Read: 29438.0626

DOS average: 21442.5394

 

Screen off

DOS write: 17289.0209

DOS read: 39849.0848

DOS average: 28569.0528

 

1GB Noname CF (256 B sectors)

 

Screen on

DOS write: 14207.065

DOS Read: 29175.2228

DOS average: 21691.1439

 

Screen off

DOS write: 18566.0508

DOS read: 39368.9753

DOS average: 28967.513

 

no 512BPS mode possible

[GoodByteXL]

POKE 559,0
RWTEST %1
POKE 559,34

I noticed that a program called from within a batch performs somewhat slower. Any explanations ...