CompuServe Thread

#DSB/ASDG

10 messages in this thread
#56402From: W. Craig MorrowJul 9, 1989 10:35 AM
I recently purchased a USRobotics HST 14.4k modem. I immediately began experiencing transfer errors while trying to download at 14.4k. I tried all the usual remedies, a one bit plane screen, no heavy multi-tasking, CTS/RTS flow control and a multitude of term programs. These all helped to varying degrees, but the errors persisted. I proceded to do some extensive testing, first eliminating the modem as the source of the problem, then the phone line and finally RF interference. This left me with my Hardware(2000, 2090a,2052) or software as the cause of my transfer errors. I finally discovered that unmounting all my HD partitions, except the autoboot partion, eliminated the errors. Needless to say this left me with a somewhat lifeless system and what I felt was a unacceptable soloution. After reading a review of the DSB from ASDG, I thought it was the answer to my problems. I purchased the DSB and installed it with little difficulty. The DSB did eliminate my transfer errors, but I also lost 200 – 300 cps in transfer speed. I found this somewhat puzzling, as the DSB allowed me to use a DTE rate of 38400, as opposed to 19200 with the Amiga serial port. I did some more tesing and I discovered that the level of system activity had little effect on the transfer speeds achieved by the DSB, which is encouraging. I also spoke to Perry K. of ASDG who had a few suggestions, but these had little effect. I would like to have changed the size of my serial buffer, but JRComm is currently hardwired for 8K. It seems that the DSB adds an extra layer of overhead to the transfer process, although I'm not sure why. I would like some feedback from anyone who is using a similar setup, please send me your suggestions or comments. WCM
#56411From: ASDGJul 9, 1989 1:31 PM
Some additional information such as: 1. Which protocol is in use. 2. What are the speeds in question? You mentioned a delta in speed but did not mention any figure to base the delta upon. 3. Have you tried the DSB at 19.2KB. You mentioned that you were running it at 38.4KB. 4. Has it occurred to you that since you were getting transmission errors you weren't actually writing to your hard disk as much. IE: Every block which you received that contained an error was tossed out without actually being written to the hard disk. Now that you are getting perfect reception (thanks to the DSB) you are writting every block received to the hard disk. Surely you don't suggest that writting to the hard disk takes no time and imposes no penalty. 5. It might be nice to try downloading to RAM: and reporting the results. 6. How many hard disk partitions DO you have? Each mounted partition (idle or active), will slow your machine down once each second. The device driver for hard disk controllers generally run at high priorities as well so these slow downs can be quite dramatic. Six mounted partitions <continued>
#56412From: ASDGJul 9, 1989 1:31 PM
<continuation> will slow you down more than 6 times one partition, by the way. As the author of the driver software which controls the DSB, I can assure you that it's impact upon the system is dramatically less than the impact of the serial.device. As evidence, try receiving a YModem transfer over the standard serial.device at 115,200 baud. See how far you get.
#56416From: W. Craig MorrowJul 9, 1989 2:31 PM
First of all, I would like to point out that my original message was not intended as a critique of the DSB. The DSB seems to be well manufactured and works as advertised. Now to answer your questions: I always use Z-Modem for downloading to either RAD (FFS) or Ram. The speeds I achieve on the Amiga serial port, with no HD partitions mounted, can be in excess of 1650 cps, with NO errors. The equivalent transfer with the DSB comes in at less than 1450 cps. A fully loaded system gives me transfer speeds of less than 1550 on the Amiga serial port, usually with errors of course. The equivalent DSB rates are approximately 1300 cps. Errors should not factor into the Amiga serial calculations, as JRComm uses actual data transferred in its calculations. I have also verified these numbers on the host system. I also thought the CTS/RTS handshaking was causing a major delay, as the RTS line drops frequently during transfers with the DSB, so I lowered the DTE rate to 19,200. This reduced the RTS delays marginally and it also caused a minor drop in transfer speeds. I don't quite have six HD partitions, only three FFS partitions plus my autoboot partition. Jack Radigan claims that other people using the 2090a are having similar problems with high speed transfers. I'll be switching to a HardFrame when it gets back from the shop. Please note that the above transfer speeds quoted for the Amiga serial port under "loaded conditions" are highly variable, depending on the error count and the phase of the moon. However, I can always achieve better transfer speeds through the Amiga serial port running a "dead system". WCM
#56421From: ASDGJul 9, 1989 3:43 PM
Given your answers to my questions, I don't see anything obvious which could account for the speed differential. I do know that Jack Radigan said that some of his users were getting in excess of 1700 cps using JR-Comm and the ASDG DSB. Maybe he or someone else can compare system configurations with you. I can say from personal experience that the 2090 does incur a substantial (but short) performance hit every second which gets worse for each disk partition mounted. Perhaps the first question to be answered by the folks getting better performance is: Who'se hard disk controller to you have?
#56431From: W. Craig MorrowJul 9, 1989 4:40 PM
Thanks Perry. BTW is there any difference in the way the DSB and the Amiga Serial port access the bus, ie. DMA vs. CPU? WCM
#56463From: ASDGJul 9, 1989 7:22 PM
The Amiga's serial port is not accessed over the “bus'' but rather is a direct access to two registers on Paula (if I remember). As such they (accesses) do not go over the Zorro bus. This means, of course, that accesses to the Amiga's serial port suffer the same restrictions on bandwidth as Chip memory. Another difference might be noted. The Amiga's serial port has a receive interrupt priority of 11 (using the soft notation) and a transmit interrupt priority of 0 (again using the soft notation). Bus based hardware can have interrupt priorities of either 3 or 13 (using the soft notation). The DSB can only generate an interrupt at one priority (ie: cannot separate receive priority from transmit priority). Priority level 3 is way too low. So it interrupts at 13 which is the same as most disk controllers. There really is no choice about this. Nor is there any overwhelming drawbacks. But it would have been nice if the designers of the Amiga had allowed more flexibility than the too low or quite high settings. pk
#56488From: jack radiganJul 9, 1989 9:28 PM
You need to bump the priority of JR-Comm up to about 10 when you're using the DSB and a 14.4k HST at 38.4k. That's what's proved to work the best for it. Also, turn off data compression on the HST, it can slow things down on binary data. -Jack-
#56616From: W. Craig MorrowJul 10, 1989 6:58 PM
I had JRcomm set to a priority of ten and I froze most of the other tasks in my system when I tested the DSB. YModem-G gave me a top speed of 1564 with the DSB, while the Amiga serial port (19200) gave me a top speed of 1718. I have compression turned off BTW. WCM
#56661From: jack radiganJul 10, 1989 10:53 PM
Two more questions, did you try 19.2k on the DSB and was it the same system you called both times? On, one more, if it was the same system, is it networked? Did you notice the receive light go out? Ok, more than two… sorry. Will contact one of the people using it and get back to you with his results tommorow. -Jack-