Category : Utilities for DOS and Windows Machines
Archive   : NJRAMD14.ZIP
Filename : BENCH

Big apologies to everyone who wanted this information, but had to
wait. I've been too busy, lately.

I benchmarked NJRAMD, Ray Duncan's EMS DISK, IBM DOS 3.10's VDISK,
and my Seagate ST-238/Adaptek 2070A RLL hard drive subsystem with
eachother.

The first important point is how much DOS memory each of the drives
take:

Seagate 0 bytes
NJRAMD 705 bytes
EMSDISK 836 bytes
VDISK.SYS 1048 bytes

Of course, the Seagate doesn't require any DOS memory, as all of it's
control routines are stored in the ROM on the controller card.

NJRAMD takes slightly less memory than EMSDISK because I've used some
extremely efficient programming methods. Just the bare essentials
are left in memory when NJRAMD loads. It seems that EMSDISK leaves
around a lot of "work space" and "temporary" variables, that don't
even need to be there. Much of NJRAMD's code is shorter, also.

VDISK.SYS takes a great deal more memory than NJRAMD because VDISK is
a much more powerful program, in a way. It provides code for a RAM
Disk for either regular DOS RAM, or for the Extended memory that AT
systems have. NJRAMD only provides a RAM Disk for the EMS and EEMS
memory systems.

The other important consideration is speed. NJRAMD proved faster
than either of the RAM Drives, and certainly faster than my hard disk
drive. (The times for the hard disk drive are included only for a
reference.)

My system runs a 4.77 MHz NEC V-20 CPU. I wrote a simple program
using SYMDEB, Microsoft's Symbolic Debugger, to test the drives. I
ran the program three times, and took an average of the three runs.

The first version of the program is simply a loop that executes 10000
times. In each iteration, the program reads one 512-byte sector from
the drive. The program always reads the same sector. All in all,
5,120,000 bytes of data were transferred during the test. These are
the results:

Program Try 1 Try 2 Try 3 Average

NJRAMD 36.01 36.03 36.02 (36.02)
NJRAMD, /S option 35.60 35.71 35.63 (35.65)
EMSDISK 36.53 36.73 36.45 (36.55)
VDISK.SYS 1:01.46 1:01.38 1:01.42 (1:01.42)
Hard Drive 2:47.30 2:47.45 2:47.53 (2:47.43)


NJRAMD and EMSDISK ran neck-and-neck for this first test; NJRAMD was
only negligably faster. My hard drive turned out a reasonable time,
considering that it's not buffered.

It seems strange to me that VDISK.SYS was so much slower than the
other programs, however. I checked my tests several times, and they
seem to be correct. VDISK.SYS seems to execute with very much
overhead. I would have thought that VDISK.SYS would've been *faster*
than the EMS drives, since VDISK doesn't have to deal with the
overhead of the EMS manager, as the information is right in standard
DOS memory, and doesn't need to be bank-switched into place.

The second test is also a loop that executes 10000 times. Each
iteration, however, read eight sectors, starting at the seventh
logical sector of the drive. In total, 40.960 million bytes of data
were transferred. These are the results:


Program Try 1 Try 2 Try 3 Average

NJRAMD 1:46.75 1:46.64 1:46.68 (1:46.69)
NJRAMD /S 1:45.80 1:45.73 1:45.89 (1:45.80)
EMSDISK 2:11.07 2:11.30 2:11.12 (2:11.19)
VDISK.SYS 1:17.78 1:17.70 1:17.60 (1:17.69)
Hard Drive 5:34.10 5:33.96 5:34.12 (5:34.06)


In this test, NJRAMD's special optimizations are shown. Since DOS,
and application programs under DOS, often read or write several
sectors in a row, I thought it would be beneficial if NJRAMD had
logic to account for this. As the results show, the idea payed off.
The program became significantly faster than EMSDISK.

VDISK.SYS had only slightly slower times, which, again, suggests that
the program has some sort of large speed overhead, wasting time in
its execution.


My Hard Drive again performed as expected. The mechanical device
read the data at approximately 122000 bytes per second, which is
rather good, considering the limit placed on DMA transfers in my
relatively slow system.

NJRAMD is avaialable in the file NJRAMD.ARC, here in the IBM Software
Forum. If you'd like a copy of the program on a special
pre-registered disk, send $10 to


Mike Blaszczak
112 Verlinden Drive
Monroeville, PA 15146-2041


13 May 1987