test 3 installation comments/questions

Michael Fulbright msf at redhat.com
Tue Oct 14 19:50:57 UTC 2003


On Tue, 2003-10-14 at 15:13, Don wrote:
> Michael,
> I like the new install progress bar being reduced to a single "overall
> install bar", I agree with your comment about "visual noise" regarding
> individual package progress.
> 

Thanks I was getting really tired of that same old screen. Plus now
marketting can thank me for all that new space for ads.  To be honest I
really prefer the blue look for Fedora versus the red look of
traditional Red Hat stuff.

What I did was setup some code (its still in anaconda commmented out)
that output the elapsed time after each package was installed.  Looking
at this I found if you wait about 100 packages into the install and
compute the average bytes/sec up to that point, divide the total number
of bytes in the install by this, and then multiply by around 1.5, you
could get the total install time pretty closely. Of course this assumes
that the ftp connection doesn't slow down, etc.

I compute the remaining time as the remaining bytes divided by the
average bytes/sec for the install. I then average this over the last 10
packages. This settles down the number somewhat.

Finally I round the number up to a 5 minute boundary for time greater
than 10 minutes, and a 2 minute boundary for less than 10 minutes.

All of this gives a reasonably monotonic sequences of remaining time
that is close enough you'll know if you have time to go grab a sandwich
or not.  I haven't really tried this with upgrades so I don't know how
accurate the timing is in that case.  The difference is we turn off some
db syncing for installs, since if the install blows up we don't care if
the rpm database is intact. On upgrades we turn on some paranoid stuff
so if the install breaks your rpm db has some chance of being intact.
This slows things down in a way that could affect the timing.

If anyone wants to experiment with this for their class project let me
know. I don't really want to look at it ever again :)

Michael Fulbright
msf at redhat.com








More information about the test mailing list