*********************************************************************
Note: this version is fixed to work with the PPS Zeus '040 card for
the 2000. The fix is simply to update and return the proper io_Error
code (0 if no error exists). Previously, this code failed to do this
(mode cache codes fail to do this!) which results in problems for the
Zeus. This code should have no problems working with any other
controller cards as well.
This code has been recompiled with SASC 6.3.
Any questions about this fix can be directed to me. Questions about
the original cache software must be directed to to the original
author of this code.
David Walthour
walthour dartmouth.edu
*********************************************************************
This is SmartDisk1.3.1, written by David Le Blanc.
* removed some accidental 2.0 dependent code.
* added command line parameters
It is an '8-Way set associative' cache, with 8 sets, lines of 4 sectors with
prefetch, and 32 lines. If you under stand cache jargon, then you'd know that
adds up to:
32*4*8*512 = 524288 or 512K and sectors are 'prefetched' four at a time.
The frefetch strips time of program loading and directories. It may
work against you on a heavily fragmented disk, but there is a general
noticable gain in performance EVEN when the data you want is not in
the cache.
So, this cache uses 512k, which gets allocated as the cache expands.
There are a few limitations, but thats only because all parameters are
hard coded into the source.
| Since you cannot specify the type of drive you have, there are three versions
| in the archive.
| SmartDisk.xt : Any xt.device user (A590 xt/ide drive)
| SmartDisk.scsi : Any scsi.device user, (a590 scsi, A2091 etc)
| SmartDisk.gvpscsi : Any gvpscsi.device user, namely all GVP controllers.
^ Obsolete.
Arguments:
SmartDisk <devicename> <unit>
******** ******** ******** ******** ******** ******** ******** ********
I Must put a note in here to warn users with multiple drives. THIS
PROGRAM WILL FAIL on a system which uses the same device (eg scsi.device)
for many physical units. The reason is that it patches the device 'BeginIO'
vector, effectively catching calls to read ANY unit, but it does not
internally differentiate amongst different devices. The reason is simple,
there is not indication of the 'unit' in the IOStdReq structure.
(There is a 'Unit' pointer but that is device private and I'd be stupid
to depend on it :-)
The effect is that if a read from unit 1 gets cached, the same read to
unit 0 may grab an entry from the cache which actually came from unit
1! (can you spell BOOM!!)
BE WARNED!
******** ******** ******** ******** ******** ******** ******** ********
* SmartDisk 1.3.1
*
* Added code to grab device and unit from the command line.
* Removed 2.0 dependent code which had crept in. (CreateMsgPort() instead of
* CreatePort() for example)
*
* SmartDisk 1.3.
*
* Created by David Le Blanc 29/10/91 Absolutely no copywrite. But if you
improve
* it, please send me a new version (with source!)
*
* Some performance quotes: (Doesn't everyone make these??)
* Background: I have a directory called MAN: which has 355 manuals.
*
* WARNING: These are bad examples, since a 'dir' reads the disk, then sorts the
* contents, then writes the data to the screen. These times include the sorting
* and output of the directory. This sorting and output time is in the order
* of 1.5 to 2 seconds.
*
* Normal DIR MAN: 12 seconds
* Cache enabled but empty 9 seconds (prefetch does work!)
* Cache primed 5 seconds.
*
* With a slower drive and/or faster machine these times can only improve.
* I have a drive capable of 800k/sec on my unaccelerated A500. Those with
* a 150K/sec A590 would notice a greater performance boost. Same for those
* with 'bloody fast machines' (grumble :) If I was REALLY worried about making
* the statistic look good, then I'd test it on an A590.
*
|
