Measuring hard drive speeds

[SimonV]

I am bit confused about measuring drive speeds.


I have an A2000 with a Fusion Forty and 2 options for the HD.


- First there's the GVP Impact HC+8 scsi controller with a 9gb screamer outperforming the controller and the zorro2 bus. Drive is on synchronous, gururom 6.14 . When I do a rawscsi test I get 3.3mb/s and since it's a DMA controller cpu doesn't have to do a lot of work. But with sysinfo I am getting like 1.4mb/s . Where's the difference ?


- Secondly I have a Buddha Flash, the old one, with a 200gb Maxtor drive and with sysinfo I get 1.7 to 2mb/s , forgot the rawspeed but there's like zero cpu cycles left.


So what should I consider ? The GVP has another 6mb for the system although I probably don't need more than the 32mb on the fusion. I would of course swap the test drive in both cases for a flash card.


Sysinfo says the Buddha's faster but is it ?

[daxb]

You could do a manual test via script for example:

date >ram:time.txt
copy stuff from to
date >>ram:time.txt

Copy a large file from ram to hd, hd to ram and hd to hd. The same for several files like SYS:. The time is only +-1 sec accuracy with date. For 1/50 sec accuracy use ARexx time(r) for reset the timer and time(e) for elapsed time.

Edit: Add ">" sign to the second date command.

[Jope]

>> for the second command so it will not overwrite the start date :-)

[SimonV]

Originally Posted by Jope

>> for the second command so it will not overwrite the start date :-)

Yes ... What ? :-)

[daxb]

Ups... of course. Thanks Jope for fixing!

[SpeedGeek]

Originally Posted by SimonV

I am bit confused about measuring drive speeds.


I have an A2000 with a Fusion Forty and 2 options for the HD.


- First there's the GVP Impact HC+8 scsi controller with a 9gb screamer outperforming the controller and the zorro2 bus. Drive is on synchronous, gururom 6.14 . When I do a rawscsi test I get 3.3mb/s and since it's a DMA controller cpu doesn't have to do a lot of work. But with sysinfo I am getting like 1.4mb/s . Where's the difference ?


- Secondly I have a Buddha Flash, the old one, with a 200gb Maxtor drive and with sysinfo I get 1.7 to 2mb/s , forgot the rawspeed but there's like zero cpu cycles left.


So what should I consider ? The GVP has another 6mb for the system although I probably don't need more than the 32mb on the fusion. I would of course swap the test drive in both cases for a flash card.


Sysinfo says the Buddha's faster but is it ?

I think I will have to begin ignoring MisInfo discussion threads from now on. I must apologize for the sarcasm, but they are just too annoying to me.

https://eab.abime.net/showthread.php?t=106329

[SimonV]

I tried rscp, it's on my gururom install disk. There's the regular test where the Buddha beats the GVP speedwise and then there's the 24bitdma test ( or something like that ) where the GVP more or less destroys the buddha. I know the Buddha doesn't do DMA but which of the two tests is relevant for the GVP.

All too confusing.

And the sarcasm is a bit lost on me to be honest.

[SpeedGeek]

Because:

1) Your question essentially asks why do different measurement tools give different results? So why ask a question which answers itself?

2) Your question assumes SysInfo is a good benchmarking tool. Is there any hint with this name or it is not called SysBench or BenchInfo for a reason?

3) rawscsispeed attempts to measure SCSI target device to SCSI controller speed but how is this fundamentally different than DMA to RAM transfer speed?

4) Buffer size typically influences DMA to RAM transfer speed more than PIO transfers

5) No matter how many times I explain this information, these threads will continue to appear.

P.S. This is the last time.

[Damion]

Originally Posted by SimonV

I tried rscp, it's on my gururom install disk. There's the regular test where the Buddha beats the GVP speedwise and then there's the 24bitdma test ( or something like that ) where the GVP more or less destroys the buddha. I know the Buddha doesn't do DMA but which of the two tests is relevant for the GVP.

I think the slower test is the more relevant one. Here's why: the GVP's controller chip can only DMA transfer to its own onboard memory, not to your (much better performing) Fusion Forty's memory (which the GVP can't "see"). When you run the rawscsi or the RSCP DMA test, you're measuring a test case where the GVP's SCSI controller transfers data to it's own onboard memory via DMA (CPU use is low, since it only has to arrange the parameters of the transfer, while the GVP's controller chip does the "heavy lifting").

However, since your accelerator RAM would normally be highest priority, transfers involving either device will be handled solely by the CPU to the accelerator memory (PIO), hence the higher CPU use and lower numbers in the other tests.

If you have Ralph's book for the Guru-ROM, it gives a *significantly* better explanation.

I like the idea of measuring these things to see what typically works best in your scenario, but it looks like the Buddha might be the winner. If you were stuck with the 68000 (or an accelerator without RAM) the
GVP would be the far better option.

Caveat: the above is according to my admittedly far from perfect understanding.

[SimonV]

Originally Posted by SpeedGeek

Because:

1) Your question essentially asks why do different measurement tools give different results? So why ask a question which answers itself?

2) Your question assumes SysInfo is a good benchmarking tool. Is there any hint with this name or it is not called SysBench or BenchInfo for a reason?

3) rawscsispeed attempts to measure SCSI target device to SCSI controller speed but how is this fundamentally different than DMA to RAM transfer speed?

4) Buffer size typically influences DMA to RAM transfer speed more than PIO transfers

5) No matter how many times I explain this information, these threads will continue to appear.

P.S. This is the last time.

Last time for you maybe.


1)2) Sorry for being dumb.


3)4) If I had all the answers I would understand what you were saying but I don't. Same as 1)2) I guess.


5) I can see why.






Damion's

[SimonV]

Originally Posted by Damion

I think the slower test is the more relevant one. Here's why: the GVP's controller chip can only DMA transfer to its own onboard memory, not to your (much better performing) Fusion Forty's memory (which the GVP can't "see"). When you run the rawscsi or the RSCP DMA test, you're measuring a test case where the GVP's SCSI controller transfers data to it's own onboard memory via DMA (CPU use is low, since it only has to arrange the parameters of the transfer, while the GVP's controller chip does the "heavy lifting").

However, since your accelerator RAM would normally be highest priority, transfers involving either device will be handled solely by the CPU to the accelerator memory (PIO), hence the higher CPU use and lower numbers in the other tests.

If you have Ralph's book for the Guru-ROM, it gives a *significantly* better explanation.

I like the idea of measuring these things to see what typically works best in your scenario, but it looks like the Buddha might be the winner. If you were stuck with the 68000 (or an accelerator without RAM) the
GVP would be the far better option.

Caveat: the above is according to my admittedly far from perfect understanding.

That surrely points me in the good direction. So basically DMA on a zorro2 card is useless when you have fast accelerator. Even my Oktagon and DKB spitfire beat the GVP in speed but I guess they wont in 68000 mode. My Microbotics hardframe, does dma, goes as fast on 68000 as the GVP on 040.


I guess it's best to stick with the Buddha. More convenient also as I am using it's clockport or a Subway USB and no need to buy a scsi2sd that probably brings it's own frustrations.