Approximately 250 DBs were sent to Penn from Queen's for bipolar problem debugging.

Summary of DB Repairs at Penn

• The first step was to insure that the Penn repairs did more good than harm. To this end, Penn sent to Queens 15 DBs which were then run through the Queens test stand. Aside from some minor problems symptomatic of initial teething difficulties associated with the Penn test stand, the boards passed the Queens test stand. We are therefore confident that the Penn repairs are doing the right thing.
• We have added Chuck Alexander to the local workforce which will increase our throughput substantially for the short term. We are aiming for 30-40 repaired boards/week. We also have a candidate to replace Chuck once he leaves for Sudbury on 12/15.
• While we have not looked at all the boards, we now have a sufficiently large statistical sample (over 100 DBs) to be able to say that it is unlikely we will find a delightfully small number of problem classes which are easy to diagnose. It looks like there are a wide variety of problems, each of which requires placing the DB into the Penn test stand (as opposed to making a simple DVM measurement, say). The time overhead for placing a DB into a test stand is substantial, so it makes the most sense to run each DB through the full test suite once it has been mounted.
• We have stopped using the level 3 testing code here at Penn to check repaired boards since the Queens test stand confirmed that we are doing the right thing.

Notes:

1. The Queens test stand is eminently capable of keeping up with Penn's rate of DB repairs, but additional manpower is needed at Queens to insure this is the case. Penn needs the quickest feedback possible on repaired DBs to insure that we never venture into the phasespace where we inadvertently end up doing more harm than good.
2. We unearthed 9 DBs which had passed the Queens test stand and which had been sent to Penn for system test purposes. These 9 should be added to the pool of fully tested and passed DBs and may be attached to MBs.
3. The Penn test stand connectors may be damaging the DB connectors. We are taking special care during insertion and removal of connectors, and are having more connectors made.
4. The amount of time it takes to diagnose a DB is ROUGHLY 15 minutes, and 10 minutes to replace a chip (done by a separate person). It then takes about 5 minutes to recheck the DB to see if the replacement has worked. In about 1/4 of the cases, this cycle has to be repeated when another chip on the same DB is replaced.
   The total time per board thus comes to about 20-25 minutes.
   We are currently spending a total of about 40 person-hours/week on this activity, so unless these numbers are way off, we would be done in about 3 weeks if not for the fact that we are going to lose about half of the total labor during finals, which start in about 2 weeks, and over Christmas break.