-*- Mode:Text -*-

(hardcopy-file "tish:keith;illops-memo.text"
	:page-headings nil)
		

				 MEMORANDUM
  
 TO:     Dan Thomas
 
 FROM:   Keith Corbett
 
 DATE:   August 11, 1986
 
 SUBJ:   MCC System Crashes 
 
 COPY:   John Salis, Ron Rando, Joe Dreussi, Scott Whalen
	 Wendy Rowe, Dave Goodine, Rick Greenblatt
 
 
 From various reports, my assessment of the current situation at MCC in
 the CAD group is that the users are very unhappy.  They continue to see
 many system crashes, with several different symptoms (different ILLOPs).
 
 Although LMI has made some progress on a few of the specific ILLOP
 reports, they perceive an overall lack of responsiveness.  Some
 customers are turning off the Garbage Collector in the hope of preventing
 some of the ILLOPs; others have expressed a desire to be put on the
 list for the next purchase of Symbolics systems.  Concern has been
 expressed by John Haney.
 
 These problems are mounting, and I don't think anything is going to
 solve the situation except for Engineering to provide some on-site
 assistance.  Someone is needed at least to investigate the problems that
 are not reproducible (most of them).  I doubt if they are willing to
 wait until after AAAI for LMI to get back to them with further bug
 investigation activity; they want some fixes, now.
 
 Please advise on the status of the investigation of each ILLOP that has
 been reported (see below), and what action will be taken to escalate
 Engineering's attention to this critical situation. 
  
  1) ILLOP #31336 - a work-around was provided to avoid a crash caused by
  the function WITH-OUTPUT-TO-STRING; we don't have a confirmation, but
  this crash has not been reported since then.  This problem has not "gone
  away", however, since other 3.0 customers may run into this specific
  situation, or a related one.
  
  2) ILLOP #10265 - "running out of address space" - reported at various
  times, but not one of the frequent ones.
  
  3) ILLOP #27553 - seems to happen in GC scavenger - originally happened
  rarely, now seen on many systems, including John Hardin.
  
  4) ILLOP #27037 - originally reported by Pat McGarrity, PCs provided to
  Dave Goodine 7/29; reported last week by Steve Dahlstrom; was seen
  yesterday at AAAI. Garbage collector running out of regions?
  
  5) ILLOP #10446 - associated with exponentiation of large numbers, e.g.
  (^ 3000 3000); reported to Engineering in April.  This has now been seen
  several times by Jim Thomas.
   
  6) ILLOP #31345 - happens to Jim Thomas in a program that does heavy
  numerical analysis; very frequent.  I have asked him to try to pin down
  some code that will reproduce this, but he is not sure if he will be
  able to do so. 
  
  7) ILLOP #3712 - only happened once, to John Hardin while booting.
  System booted next time.
 
 As you can see, the 27553 and 27037 crashes seem to be most urgent;
 31345 is also causing one person a lot of trouble.  #10446 has been on
 Engineering's "priority 1" list for a long time.
 
 KMC