80-Bus News |
July–August 1983 · Volume 2 · Issue 4 |
Page 39 of 55 |
---|
5″ Pertec GEMDDDS format | 8 mins 6 secs | (15 tracks) |
5″ Microp. GEMQDSS format | 8 mins 40 secs | (40 tracks) |
8″ Standard Single density | 6 mins 20 secs | (43 tracks) |
R0201 Winchester | 1 min 42 secs | (85 tracks) |
GM833 RAM-DISK | 17 secs |
The RAM-DISK figure is not a mis-print, it is just 17 seconds. This test is perhaps a little artificial, so the next benchmark is the time taken to PIP a source file from the Winchester to the destination drive, load M80 (from the drive under test), assemble it, then link and load it using L80.
5″ Pertec GEMDDDS format | 2 mins 34 secs |
5″ Microp. GEMQDSS format | 2 mins 35 secs |
8″ Standard Single density | 2 mins 39 secs |
RO201 Winchester | 2 min 00 secs |
GM833 RAM-DISK | 1 min 30 secs |
Here the performance difference is not so marked, but it is still significant. (This shows that in this instance the majority of the total time taken was in actual processing time and not disk access time.) The floppy disks had an edge on the Winchester as they only started off with M80 and L80 on them. By comparison the Winchester was already holding 3.9Mbytes of data and programs, and the files created during the test filled in odd holes here and there on the disk, and weren’t stored in successive blocks. Also the head was on average about 200 tracks away from the directory track for most of the time!
Using the RAM-DISK requires a careful approach to work to ensure that at the end of the day all changed and new programs end up back on permanent storage (disk). Copying a file to disk as soon as it is changed rather defeats one of the benefits of the RAM-DISK, but it has to be remembered that the RAM-DISK is volatile. Though the Gemini BIOS does not obliterate files if you are forced to press the reset switch, it cannot protect you against accidental (or deliberate!), powering down.
This is the time when the programmable function keys of the Gemini keyboard come into their own. At the start of a session one (or more) can be programmed up to provide a backup command, (e.g. PIP A:=M:*.MAC^M), and then this key can be pressed at idle moments when you pause for thought (or answer the phone), and, finally, at the end of the session.
[Ed. – In one application that I know a RAM-DISK is being used, its volatility does not matter. The application requires as rapid as possible access to any record in a very large database. When it is started the program copies a number of index files to the RAM-DISK, and uses these to control access to the database. If the power fails nothing of importance is lost.]
One area open to debate is whether, with so much memory on the board, GM833 should incorporate some form of error detection/protection? The soft error rate of RAMS is very low, but here we have 64 of them in an array (plus an additional 8 in the system memory).
Error detection and correction would be an overkill in this environment, but a possibility lies in a simple parity check on each byte. (I note a number of the large memory expansion boards for the IBM PC now offer parity protected memory). This would add eight more dynamic RAMs to the board, along with the parity generation/check logic (assuming it could all be fitted on). It also raises the question of what to do in the event of a parity error being detected. Light a LED? Generate an interrupt? Halt?
Page 39 of 55 |
---|