[335] in Athena_Backup_System
Status as of 24 October 1996 (and next steps)
daemon@ATHENA.MIT.EDU (Bill Cattey On the Road)
Thu Oct 24 14:18:36 1996
Date: Thu, 24 Oct 96 14:13:01 EDT
From: wdc@MIT.EDU (Bill Cattey On the Road)
To: athena-backup@MIT.EDU
What are our next steps?
Diane and Ted will be doing the large cell test today or tomorrow.
(Miki too will be in attendance.)
Don't forget to save terminal logs for Karlyne's documentation work.
The master spec review is now nearly done, and everyone around the table
feels that the major changes have been done.
The same sort of review of the Slave will begin after the Master Spec
review is put to bed. We expect that there will be few surprises, but
the work may be a little more tedious because the major work will be
verifying the Master-Slave communication.
So, very soon the Master Spec and the Master code will be in a shape to
present to Jeff for his review.
We talked about testing:
The writing of the test plan has been slowed down because vrt has
been spending most of his time with SGI work.
We need unit tests to confirm the APIs work.
Then we need basic system tests -- confirm that the system as we
would use it today DOES work.
We need to document the various system tests we performed.
We also need to confirm that bits coming in from restores are
indeed restored: Files, ACLs, Timestamps.
The test plan writing will focus now on fleshing out the System,
acceptance, catastrophic, and stress test sections.
Any unit testing not yet done which confirms that proper action is
taken on error returns WILL be done.
Need the test plan in reasonable shape to establish system testing by
about the third week of November.
Vrt would like to send the Master (and later the slave) through Purify.
Miki has a little replica of the current AFS backup system and is
verifying that her understanding of how reading AFS tapes is correct.
After that Miki will go back and finish her Functional Spec of the ABS
API's.
Miki's API document will be reviewed by the group. Hope to be done with
it by the third week of November.
Then Miki will modify the slave to teach it how to read the current
backup tapes being made.
After Miki is done with her changes, someone (Probably Miki or Diane)
will make the modifications of the slave necessary for the
self-throttling to prevent overloading the AFS servers, etc. that we
hashed out conceptually weeks ago when Ted and Tom joined the review.
(We colloquially refer to this modification as "Van Jacobsonization of
the Slave.")
At this time we're considering making extensive testing under load
during IAP in parallel with the regular dumps.
We expect to be doing unit testing through December.
A possible migration strategy is to do dumps in parallel, and then
attempt restores with ABS preferentially to the old system. (The old
system would be used in critical initial situations, and would be
available as a fall back. The expectation is that we would eventually
feel comfortable with using ABS all the time.)
