At 09:07 2/21/2001 -0800, you wrote:
>There is a problem with this approach. Suppose everybody did it that
>way. Then at exactly the same time each week, everybody in a given
>time zone would be trying to CVSup from their nearby mirrors. The
>mirrors would quickly fill up, and you'd be retrying for hours, along
>with everybody else in your time zone.
That starts off around a "probably true" and progresses on to "nearly
ludicrous." ;)
1. The higher you go in bandwidth (to end-users), the less users that have
that kind of bandwidth. While the possibility of the mirrors getting
congested is certainly possible, it's just as improbable. Let Y be the
number of cvsup mirrors in the US, and X be the number of freebsd boxes in
the US. Now, X/Y is the ratio of freebsd boxes to cvsup mirrors. It
doesn't take a rocket scientist to determine that number will be
ridiculously overburdened, probably in the neighborhood of a quarter
million to one (NOTE: I'm totally pulling that number out of my ass, but
it doesn't seem too unreasonable. As of Jan. 2001, out of 27 million sites
checked, 15 million were running apache. 250k of them being freebsd
would mean about 1 in 60 are freebsd.. I expect that number is probably
higher.).
With 250,000 estimated sites cvsupping to each server, If each server was
assigned a unique time to (just begin) it's cvsup, spaced one hour apart,
it would take 28 years for each server to cvsup just once. This is
obviously not the case. Possible explainations are that I've overestimated
the number of machines cvsupping (just 10k machines though would take over
a year); that the machines cvsupping, in general, have far less bandwidth
than the machines running the cvsup mirrors; etc.
The bottom line of this explaination is simple. If every machine running
freebsd was also cvsupping, it wouldn't matter -what- time they decided to
do it at, the servers would be under full load constantly.. so saying "if
everyone in one timezone did that, it would drag the server down" is a moot
point. The servers do get dragged down, but if you'll notice, in general
the lower numbered mirror you go with, the slower it is, almost any time of
day, on any day of the month. People are just too lazy to pick a mirror
thats close to them, or further down the "cvsupit" list than the first one
they see.
>It's much better to pick a random time and put it in your crontab.
I disagree. I think it would be "much better" if this was truly a concern,
to change the time that your periodics are executed, or to simply pick a
different mirror that isn't saturated when your default periodic runs.
...
In any case, to the original poster, regardless of how you decide to
schedule your updates, you'll alleviate the problem by doing as was
suggested and reading the manpage, or (a better solution imho, but reading
the manpage anyway is always a great idea) condensing the three seperate
cvsupfiles you have into one and only invoking one copy of cvsup to begin with.
-------signature file-------
PGP Key Fingerprint:
446B 7718 B219 9F1E 43DD 8E4A 6BE9 D739 CCC5 7FD7
"I don't think [Linux] will be very successful in the long run."
"My experience and some of my friends' experience is that Linux is quite
unreliable. Microsoft is really unreliable but Linux is worse."
-Ken Thompson, Interview May 1999.
http://www.freebsd.org
FreeBSD - The Power to Serve
http://www.rfnj.org
Radio Free New Jersey - 375 streams - 96kbps @ 44.1khz
http://namespace.org -- http://name.space
Resist the ICANN! Support name.space!
To Unsubscribe: send mail to [EMAIL PROTECTED]
with "unsubscribe freebsd-stable" in the body of the message
