[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Shocking: RF SCSI slower than HS SCSI???
Hi,
this got posted on GEnie. Now, admittedly, the guy who posted this has
a business interest in such benchmarks, but the method for testing is
straight: ReadSpeed (also on cco) is a program that does nothing more
than linear block read. No tricks, nothing fancy.
Here's a copy of the post:
=snip=
Ok, since there aren't any folks (until now) who dare to publish their
tests here, I will show some of mine.
BTW, if you have a RAMFast SCSI Card (rev D) and you are proud that
Disk/Verify (in the Finder) gives you a whopping calculated 700-900K
data transfer rate, you are being _cheated_. Why? Just ask the makers
of the RF card whether they can transfer data FROM HARD DISK TO THE
APPLE IIGS RAM with this speed or not. My answer doesn't count.
However, it seems that some people _want_ to be cheated.
Here are some speed test results with various interface cards and hard
drives (using the program ReadSpeed available in the A2 library).
--------------------------------------------
Hard Disk: Conner CP30104 120MB
Interface: Turbo IDE Card
Device number: 1
Device name: .TURBO.IDE.1
Total number of blocks: 65520
Request block size: 1024
Number of requests: 60
Number of blocks read: 61440
TWGS disabled:
Time for reading: 43 sec
Data transfer rate: 714.419 (+- 16.614) KBytes/sec
Single block access: 0.700 (+- 0.016) msec/block
TWGS enabled:
Time for reading: 41 sec
Data transfer rate: 749.268 (+- 18.275) KBytes/sec
Single block access: 0.667 (+- 0.016) msec/block
--------------------------------------------
Hard Disk: Conner CP3024 20MB
Interface: Turbo IDE Card
Device number: 1
Device name: .TURBO.IDE.1
Total number of blocks: 31632
Request block size: 1024
Number of requests: 30
Number of blocks read: 30720
TWGS disabled:
Time for reading: 33 sec
Data transfer rate: 465.455 (+- 14.105) KBytes/sec
Single block access: 1.074 (+- 0.033) msec/block
TWGS enabled:
Time for reading: 33 sec
Data transfer rate: 465.455 (+- 14.105) KBytes/sec
Single block access: 1.074 (+- 0.033) msec/block
Comment: Obviously the physical speed limit of this hard disk
--------------------------------------------
Hard Disk: Vulcan 20MB (Seagate ST125 AT)
Interface: Turbo IDE Card
Device number: 1
Device name: .TURBO.IDE.1
Total number of blocks: 32768
Request block size: 1024
Number of requests: 32
Number of blocks read: 32768
TWGS disabled:
Time for reading: 52 sec
Data transfer rate: 315.077 (+- 6.059) KBytes/sec
Single block access: 1.587 (+- 0.031) msec/block
TWGS enabled:
Time for reading: 36 sec
Data transfer rate: 455.111 (+- 12.642) KBytes/sec
Single block access: 1.099 (+- 0.031) msec/block
Comment: not bad, eh? This was one of the slowest mechanisms
ever built into Vulcan hard drives. Vulcan owners know
the speed of their HD.
--------------------------------------------
Hard Disk: Quantum ProDrive 40S 40MB
Interface: Apple SCSI Rev. C
Device number: 5
Device name: .APPLESCSI.HD01.00
Total number of blocks: 51200
Request block size: 1024
Number of requests: 50
Number of blocks read: 51200
TWGS disabled:
Time for reading: 309 sec
Data transfer rate: 82.848 (+- 0.268) KBytes/sec
Single block access: 6.035 (+- 0.020) msec/block
TWGS enabled:
Time for reading: 159 sec
Data transfer rate: 161.006 (+- 1.013) KBytes/sec
Single block access: 3.105 (+- 0.020) msec/block
Comment: in this case, for programmed I/O the TWGS doubles the speed
--------------------------------------------
Hard Disk: Quantum ProDrive 40S 40MB
Interface: Apple High Speed SCSI
Device number: 5
Device name: .APPLESCSI.HD01.00
Total number of blocks: 51200
Request block size: 1024
Number of requests: 50
Number of blocks read: 51200
TWGS disabled:
Time for reading: 40 sec
Data transfer rate: 640.000 (+- 16.000) KBytes/sec
Single block access: 0.781 (+- 0.020) msec/block
TWGS enabled:
Time for reading: 34 sec
Data transfer rate: 752.941 (+- 22.145) KBytes/sec
Single block access: 0.664 (+- 0.020) msec/block
Comment: the software overhead for DMA I/O still can be accelerated.
--------------------------------------------
Hard Disk: Quantum ProDrive 40S 40MB
Interface: RAMFast SCSI Rev. C
Device number: 5
Device name: .RAMFAST.SCSI.1
Total number of blocks: 51200
Request block size: 1024
Number of requests: 50
Number of blocks read: 51200
TWGS disabled:
Time for reading: 61 sec
Data transfer rate: 419.672 (+- 6.880) KBytes/sec
Single block access: 1.191 (+- 0.020) msec/block
TWGS enabled:
Time for reading: 61 sec
Data transfer rate: 419.672 (+- 6.880) KBytes/sec
Single block access: 1.191 (+- 0.020) msec/block
--------------------------------------------
Hard Disk: Quantum LPS270S 270MB
Interface: RAMFast SCSI Rev. D 3.01
Device number: 11
Device name: .CVTECH.S2.C
Total number of blocks: 65536
Request block size: 1024
Number of requests: 64
Number of blocks read: 65536
TWGS disabled:
Time for reading: 68 sec
Data transfer rate: 481.882 (+- 7.087) KBytes/sec
Single block access: 1.038 (+- 0.015) msec/block
TWGS enabled:
Time for reading: 67 sec
Data transfer rate: 489.075 (+- 7.300) KBytes/sec
Single block access: 1.022 (+- 0.015) msec/block
Comment: this is the maximum speed I ever could get out of
any configuration with a RF SCSI card. Also see comment
in introduction.
--------------------------------------------
Hard Disk: Quantum GoDrive 40AT 40MB
Interface: MicroDrive Card
Device number: 1
Device name: .MICRODRIVE.1
Total number of blocks: 41728
Request block size: 1024
Number of requests: 40
Number of blocks read: 40960
TWGS disabled:
Time for reading: 72 sec
Data transfer rate: 284.444 (+- 3.951) KBytes/sec
Single block access: 1.758 (+- 0.024) msec/block
TWGS enabled:
Time for reading: 63 sec
Data transfer rate: 325.079 (+- 5.160) KBytes/sec
Single block access: 1.538 (+- 0.024) msec/block
Comment: a fair competitor would be the Apple SCSI Rev. C.
--------------------------------------------
=snip=
Now, if this is accurate, I am truly crushed. It would make sense,
the RF SCSI has to to transfer twice ... first to buffer, then to
memory, which might be fine in P8 where you only have 1-block reads,
but with GS/OS, where multiple-block reads speed things up, this can be
devastating ... as shown.
And here's me, shelled out all the extra money for a RF for the speed,
and this is going to be the result? Argh.
Could some kind persons get the readspeed program and see what their
results are?
Or, even, to be absolutely sure this test is not being influenced
towards the "competitors" card, write our own version of that thing ...
shouldn't be hard, really.
I _really_ want to know what's what now. It's not as if HD speed is
ultra-important, but the difference between 490k/s and 750k/s might be
noticeable. Not great, granted, but noticeable. And speed is what the
RF is supposedly all about ... or so I thought up until now :(
Soenke
--
Standard disclaimer: My opinions are my own, not those of my employer