[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Recent MINITS changes
- To: bug-minits@MIT-MC, bug-twenex@MIT-XX, info-lispm@MIT-OZ, network@SCRC-POINTER
- Subject: Recent MINITS changes
- From: dcp@SCRC-POINTER
- Date: Sat ,9 Apr 83 17:17:00 EDT
It used to be the case that MINITS would read packets out of the chaos
board at process level instead of interrupt level. It did this so it
would not spend excessive time at priority 5 frobbing the chaos board
and not being able to service 9600 baud interrupts (whether they be
DL-type or DUP-11). Unfortunately, waiting for a process to schedule
would often cause many packets to get lost. Much of this is no longer
true.
It now drops the priority to that before the interrupt happened and
services the interface immediately (with appropriate checks and locks).
This allows 9600 baud interrupts to still happen, but does not have the
scheduling latency. For a LispM blasting packets to a MINITS as fast as
it can (e.g., using the MINITS SINK contact name), about 34% of the
packets used to get lost. Now this number is about 2%. This also
improves output traffic because the output interrupt will also get
serviced nearly immediately (providing the receiving end does not
delay).
When the new software propigates (I'll make an attempt to do the 4
PDP-20s tomorrow [Sunday]) the effect users will see include the
following:
* Faster writing of files to TWENEX (e.g., writing a LispM editor
buffer or a CFTP transfer)
* Better bridging between subnets (especially from a chaos subnet to an
ethernet subnet).
* Perhaps better terminal response on concentrators due to less lost
packets on both the concentrator and other MINITS bridges on the
route to/from the mainframe.