Hi, gang.
(Some notes, gets more interesting towards the end.)
>>>>> "Avery" == Avery Pennarun <[EMAIL PROTECTED]> writes:
Avery> On Mon, Mar 04, 2002 at 11:39:44PM -0800, Daniel Quinlan
Avery> wrote:
Avery> This is on purpose -- netselect
So, this discussion arguably belongs to netselect itself.
Avery> gives you the ip address if the name resolves to more than
Avery> one address. For example, if netselect gave you
Avery> "http.us.debian.org" as your fastest server, it wouldn't do
Avery> you any good.
Daniel> I understand, but it's a completely unreliable strategy.
Daniel> I can't put a numeric address into any configuration file
Daniel> (with the exception of /etc/resolv.conf) because numeric
Daniel> addresses change -- it's part of the reason why DNS
Daniel> exists.
Avery> Yes you can. Just regenerate the config file every now and
Avery> then using netselect like you did in the first place.
Avery> People's network services change, and the Internet
Avery> rearranges itself regularly. The fastest/closest server
Avery> one day might not be the fastest server the next day.
Avery> In fact, since it looks like you're doing time servers, at
Avery> least cronyd appears to _recommend_ using numeric IP's for
Avery> some reason.
Daniel> It would be better to search all addresses and present an
Daniel> average and/or all results. Something like:
Daniel>
Daniel> hostname.foo.com aa.bb.cc.dd 0.40
Daniel>
Daniel> hostname.foo.com ww.xx.yy.zz 0.83
Daniel>
Daniel> hostname.foo.com average 0.62
Daniel>
Daniel> At this point, you can see what is happening and decide to
Daniel> use a different host, or if http.us.debian.org is (on
Daniel> average) fast enough compared with the other options, just
Daniel> go with that.
Avery> No, this would be useless, at least for http.us.debian.org.
Avery> The speeds of those hosts vary wildly, so the extremely
Avery> fast host would be ignored because the irrelevant slow ones
Avery> punish it.
I think this is just what I am experiencing now. Some packages
dripple along, some flood in.
Avery> Look at the results for me:
Avery> ~ $ netselect -vv http.us.debian.org Running netselect to
Avery> choose 1 out of 6 addresses.
Avery> ...............................................
Avery> 128.101.36.192 75 ms 19 hops 40% ok ( 2/ 5) 141.213.4.21
Avery> 399 ms 16 hops 62% ok ( 5/ 8) 192.25.206.10 246 ms 23 hops
Avery> 20% ok ( 1/ 5) 35.9.37.225 53 ms 19 hops 90% ok ( 9/10)
Avery> 208.185.25.38 19 ms 11 hops 90% ok ( 9/10) 204.152.189.120
Avery> 115 ms 14 hops 90% ok ( 9/10)
Avery> 208.185.25.38 is _way_ faster than 141.213.4.21. They're
Avery> not even comparable.
(So, this is arguably a question of the Debian system administration -
hosts under the same DNS name *should* be comparable.)
Daniel> Options to implement different behaviors/strategies would
Daniel> be acceptable. Just using the first numeric address isn't
Daniel> very useful.
Avery> It's not the first numeric address. It's the fastest one.
Importamt point.
Avery> I don't believe anything else is worthwhile.
I do. I am also concerned about IP Addresses in config files, as they
tie me to internals of the administration of the service. What I've
done is to change netselect-apt to report the best result of a server
with a unique address also and, for comparison, the netselect ratings.
Fairly straightforward, and it gives you a simple option to use an IP
address or a name.
These differences are not *that* dramatic:
============================================================
The fastest named server (rated 140) seems to be:
http://ftp.tiscali.de/pub/debian/debian/
But there is an unnamed server with rating 109:
http://194.97.2.69/debian/
============================================================
(Yes, the wording is misleading - I should have read this thread
before hacking this.)
Avery> You can reverse the number back into a name (not likely the
Avery> same name you started with) using the 'host' command if you
Avery> want. Normally (but not always) these reversed names have
Avery> single IP addresses.
Avery> If you really, really want to reduce the usefulness of your
Avery> netselect results, you can use the tag feature if you want:
Avery> ~ $ netselect -vv http.us.debian.org:http.us.debian.org
Avery> Running netselect to choose 1 out of 6 addresses.
Avery> ...................................................
Avery> 128.101.36.192:http.us.debian.org 65 ms 19 hops 50% ok ( 2/
Avery> 4)
Avery> 141.213.4.21:http.us.debian.org 316 ms 16 hops 90% ok (
Avery> 9/10)
Avery> 192.25.206.10:http.us.debian.org 228 ms 23 hops 33% ok ( 1/
Avery> 3)
Avery> 35.9.37.225:http.us.debian.org 56 ms 19 hops 90% ok ( 9/10)
Avery> 208.185.25.38:http.us.debian.org 19 ms 11 hops 90% ok (
Avery> 9/10)
Avery> 204.152.189.120:http.us.debian.org 116 ms 14 hops 90% ok (
Avery> 9/10)
Avery> Then you can implement your own averaging strategy if you
Avery> want.
The averaging strategy I'd want would collect the results for all DNS
names and discard those with too widely differing ratings (what do
those statisticians call it? Variance?). Then, report the best-run
most reliable service. But I won't go into this, as will apparantly
nobody else.
I can post a diff to 0.3-6 if you're interested.
Greetings,
Felix
--
To UNSUBSCRIBE, email to [EMAIL PROTECTED]
with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
