[2359] in Release_7.7_team
Re: splitting release across multiple nights
daemon@ATHENA.MIT.EDU (Jonathon Weiss)
Wed Jul 19 21:16:23 2000
Message-Id: <200007200116.VAA06110@detraction.mit.edu>
From: Jonathon Weiss <jweiss@MIT.EDU>
To: Garry Zacheiss <zacheiss@MIT.EDU>
cc: release-team@MIT.EDU
In-reply-to: Your message of "Wed, 19 Jul 2000 16:08:26 EDT."
<200007192008.QAA16253@riff-raff.mit.edu>
Date: Wed, 19 Jul 2000 21:16:16 -0400
> Last year, we split up the release by platform type, doing Suns
> one night, Indys the next, and O2s on the third night, the idea being to
> avoid driving both the network and AFS servers into the ground, and
> keeping machine updating well into the morning. Do we want to do this
> again this year? I'm ambivalent at this point, but someone else might
> have stronger feelings.
Thanks for remembering this Garry.
Here is a back of the envelope calculation for w20
25 ultras * 120M = 3G
60 SS{4,5}s * 100M = 6G
(The sizes are based on the assumption that all of the data on / and
/usr has to be copied during an update.
If I've done the math right (modulo rounding), this will *fill* a
10Mbps ethernet for 2h15m. (tho it looks like w20 has a fast ethernet).
Looking at this from the server-side: Based on the athena-inventory
database it looks like there are ~500 ultras, and ~750 sparcs. This
totals 135G. If we assume that the AFS servers are split between 2
100Mbs pipes*, we can say we have 67G to push through each pipe.
Assuming we have the full bandwidth of those pipes (which we won't)
that will take about 1h30m. Even if we assume we only get half of the
bandwidth, 3 hours seems like it ought to be safe.** Still, there will
be enough data flying that I'd feel more comfortable updatiing the
SGIs on a different night, and possibly running the hesiod propogation
early for the suns.
What do other people think?
Jonathon
* For those craving details, those are the pipes connecting e40-rtr to
the rest of the world, and the FastEthernet (serving both 179 and 158)
in w20.
** If we were really paranoid we could change all of the hesiod for
cluster machines and do them on a different night than the suns in
labs and offices, but unless someone checks my assumptions and math
and finds somethign really amiss, this is probably overly paranoid.
