Date: Thu, 9 Jul 1998 07:03:31 -0500 From: NOC-IPAD <sysadmin@greeves.mfn.org> To: NOC-IPAD <sysadmin@greeves.mfn.org>, "freebsd-questions@FreeBSD.ORG" <freebsd-questions@FreeBSD.ORG>, "'rafie@innocent.com'" <rafie@innocent.com>, "'Greg Lehey'" <grog@lemis.com> Subject: RE: Availability & Cost Message-ID: <01BDAB07.AEDA6130@noc.mfn.org>
next in thread | raw e-mail | index | archive | help
>On Thursday, 9 July 1998 at 6:07:36 -0500, NOC-IPAD wrote: >> On Thursday, July 09, 1998 12:49 AM, Greg Lehey wrote: >>> On Thursday, 9 July 1998 at 5:02:22 -0500, NOC-IPAD wrote: >>>> On Wednesday, July 08, 1998 10:51 PM, Rafie wrote: >>>>> >>>> Use a mirror site! Freebsd's main server is >>>> (/FLAMESUIT ON) >>>> very overloaded, >>>> (/FLAMESUIT OFF) >>>> and on a connection which dies going through MCI's teeny-tiny >>>> little pipe! >>> >>> It's a good thing you have your flame suit on. This advice is >>> completely off the point. What evidence do you have of your claims? >>> I've always had excellent response times from both ftp.FreeBSD.org >>> (really wcarchive.cdrom.com, the world's largest ftp server), and >>> www.FreeBSD.org (really freefall.FreeBSD.org). I've just tried again, >>> and again I had excellent response times. >> >> You may want to note that it is also 0100 here, >Ah. At first I didn't understand what you meant, since your machine >claims it's 6 am. But you're right, and your machine is wrong: it's >just past 1 am Just goes to show you shouldn't believe everything you read! >> and at the site (I seem to recall it was in Texas, if not, it is >> *still* not exactly "primetime"). >Which site? wcarchive and freefall are both in California. Which would put their local time at 2300 and the local time on the east coast at 0200. Considering that the US comprises the *vast* majority of internet traffic, you are proving my point - this is one of the best possible time to look at this from the perspective of wuarchive/freefall, yet, they are still pushing 2700+ users across their ftp connection. >> I have had nothing but poor response times from the main server, and in >> fact don't even use it any more if I can avoid it. When I see a notice telling >> me that I am user 2990 of 3000, and the response time is cruddy (at best), >> my first thought is the machine. BUT: I have done the requisite traceroutes, >> and as I said, MCI's pipe *stinks*. Nevertheless, as a *user*, I have noticed >> correlations between trivial response times and user count: extreme ones. >Well, obviously the data flow through "the pipe" has an influence. >But don't forget that there isn't just one "pipe". How does this >traceroute compare to yours? This is done from a 33.6 kb/s PPP link >(thus the 120 - 150 ms at the start). You'll notice an additional 250 >to 300 ms between Melbourne and Bloomington (across the Pacific), but >after that you can't see any difference. <traceroutes snipped> Ive *never* seen traceroutes this good, to be perfectly honest. In fact, I am used to seeing delays anywhere from 1000ms to 5000ms at the MCI connection. >On the other hand, looking at you, we see: >=== grog@freefall (/dev/ttypa) /d/users/grog 7 -> traceroute greeves.mfn.org >traceroute to greeves.mfn.org (204.238.179.3), 30 hops max, 40 byte packets > 1 gate-free.cdrom.com (204.216.27.17) 1.430 ms 1.750 ms 1.509 ms <snip> >14 * * * >15 * * * > >In other words, > >1. You don't use MCI at all. >2. Within crl.net, the speeds are fine. >3. There are a lot of dropouts in sprintlink.net, but the speeds > still seem acceptable >4. Your machine doesn't appear to be connected. Greg, you've been in the business *how many* years? You can't spot a firewall yet? C'mon, gimme a break! > >I'd like to see your view of this. > >> My *last-ever* encounter with this *very overloaded* server was when it took >> me over an hour to (a) get a free port to log in to, and then (b) load a trivial >> (fixit image) via ftp. I have seen that box scream, but only during the wee >> hours of the morning, on weekdays. > >I don't doubt that you had this experience, but I very much doubt the >conclusions that you draw from it. I can't remember ever having to >wait for a port on wcarchive, though it's possible. When was this? >Which machine? > wuarchive. November through December of 1997. Even wrote the core team about it, got a jumpy reply (a lot like yours!) from someone (actually, I believe it was from Jordan, but I can't swear to it anymore). >> It is unfortunate that the FBSD ppl take these observations as >> personal attacks (as you seem to have done), > >Not at all. I'm picking on this one not because I'm upset, but >because it's wrong and will tend to mislead people. > I stand by my statements. They are not wrong, they are observations made as an educated "user" of wuarchive/crl. They are just as valid as your observations of prime performance. >> they are not meant that way. However, your like or dislike of these >> do not invalidate them. When the user count is high, response time >> stinks to high heaven. > >As I said at the beginning, you have brought no evidence of this. I >get very good response time. You have brought no evidence of this either - sorry, a 2 block ftp at 2300 does not count in my book. Maybe a couple of hundred megs during prime time would convince me tho ;-) > >Here's an example: > >$ time ftp ftp://ftp.cdrom.com/archive-info/slow.txt >Connected to wcarchive.cdrom.com. >220 wcarchive.cdrom.com FTP server (Version DG-3.1.11 Wed Jul 1 02:21:57 PDT 1998) ready. >331 Guest login ok, send your email address as password. >230-Welcome to wcarchive - home FTP site for Walnut Creek CDROM. >230-There are currently 2771 users out of 3200 possible. <ftp session snipped> >221 Goodbye! > >real 0m23.121s >user 0m1.055s >sys 0m0.029s > >In other words, from end to end, including setting up and logging in, >we had a delay of 23 seconds. The file itself is worth reading--why >don't you pick it up and tell me how fast you get it. > Not a valid test, insufficient data. And you, as an industry professional, *know* this! >> Of course, that's one of the main reasons for mirrors in the first >> place. Don't get bent outta shape, just add a box to even the load >> (or deal with it when people say that mirrors are the only way to >> go)... > >I frequently find that our Australian mirror is slower than freefall. I've never used that mirror, but now that I have it on good authority, I'll be certain to stay away from it! > >Greg J.A. Terranson <YAWN> Good Night Greg! To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-questions" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?01BDAB07.AEDA6130>