|
| news.inet.tele.dk Fra : Henrik Bruun-Peterse~ |
Dato : 19-01-01 22:45 |
|
Hej
Hvad sker der med denne server?
Kunne ikke forstå at jeg manglede filer både i dk.binaer og dk.binaer.musik
prøvede så lige sunsite for sjov og der lå de alle og flere til.
Det har ALDRIG været et problem før nu så er der nogen der ved hvad der er
sket?
--
Mvh
Henrik
news@nytliv.net
| |
Charlie Chronic (19-01-2001)
| Kommentar Fra : Charlie Chronic |
Dato : 19-01-01 23:44 |
|
:: Kunne ikke forstå at jeg manglede filer både i dk.binaer og dk.binaer.musik
:: Det har ALDRIG været et problem før nu så er der nogen der ved hvad der er
brug news://193.162.153.117/
eller SSL snews://193.162.153.117/ (read-only)
news.tele.dk is on its way out, and will be taken down in a week for
what should be a short time. It has not been able to keep up with
the full news volume of ~250GB/day for some months now, and the machine
ahead of it (in the URL above) does not have retention to keep all
posts during peak hours. Upstream, limited bandwidth results in about
70GB out of these 250GB being lost.
The machine at news://193.162.153.117:143/ (read-only) should have all
the missing articles. As more people read from and post to this machine,
the load on news.tele.dk will go down, and its poor performance should
improve somewhat. At present, I see a world reader load of about 25
max readers, and max 5 to 10 TDC customers, compared with over 1000
TDC customers on news.tele.dk.
This is a test machine and is not the Monsternewsswerver that should be
built Real Soon Now, but this machine will probably remain in operation
until that time, and has many more binariez froups than news.tele.dk.
We apologize to all who are using news.tele.dk.
Kvindeservice
| |
Claus Schmidt (21-01-2001)
| Kommentar Fra : Claus Schmidt |
Dato : 21-01-01 15:59 |
|
"Charlie Chronic" skrev
> brug news://193.162.153.117/
> ... At present, I see a world reader load of about 25
Trying to pump up this figure 4% to 26 and in the
process saying thank you as appropriate (!) I ran
in to this one:
Emne 'Re: news.inet.tele.dk', Konto: '193.162.153.117',
Server: '193.162.153.117', Protokol: NNTP,
Svar fra server: '502 news02.inet.tele.dk is not a public
news swerver. Sorry d00d.', Port: 119, Sikker (SSL): Nej,
Serverfejl: 502, Fejlnummer: 0x800CCCA0
-now, if posting isn't allowed the new account
isn't very much fun after all which just might
explain the poor readership.
--
[claus]
-loyal, paying and pleased TeleDamnark Customer
since whoknowswhen.
| |
Charlie Crackpot (21-01-2001)
| Kommentar Fra : Charlie Crackpot |
Dato : 21-01-01 17:18 |
|
:: > brug news://193.162.153.117/
::
:: Svar fra server: '502 news02.inet.tele.dk is not a public
:: news swerver. Sorry d00d.', Port: 119, Sikker (SSL): Nej,
::
:: -now, if posting isn't allowed the new account
:: isn't very much fun after all which just might
There is a software-enforced limit of two parallel connections, in
place from the time when I was trying to track down unexplained
freezes that seemed to be related to readers making as many as ten
simultaneous connections. (I have since determined that the crashes
were due to stale lockfiles left behind in the database.) This was
supposed to be enough for someone to open one connection for reading
and a second one if needed for posting.
There is a BOFH-enforced limit of two connections on port 143 for
non-customers, which is done by adding the offending IP (which may
sometimes be seen to have opened ten connections) to a blockhole
list so the news swerver just seems to disappear forever. This has
been mostly used against a small number of T-Online and Sympatico
customers who apparently do not read the fine print. Like Bill
Gates said, two connections should be enough for anyone.
If your experience is that you need more than two connections to
run your high-speed line at near-capacity, you need to adjust your
receive window size upward from the default I see most of 8kB size.
There's a web page out there that tells how to do this in 'Doze.
If that still isn't enough, your ISP is probably overbooking its
capacity.
There may also be a rate-of-connection limit in place that prevents
readers from opening a connection, closing it immediately, and then
promptly opening a new one, again because this seemed to be part of
the cause of the earlier crashes. (As it turns out, because these
closed connections were aborting the release of the locks they had
grabbed.) But I'm not sure if I was able to get it to work on this
OS properly -- in fact, it is commented out.
The error you see is the generic failure message, intended to be
returned to non-customers reading from this port, because I have
not changed things to return different banners for the failures of
too many connections or machine-gun-like connect patterns.
I do not know what the Outhouse defaults for connection patterns
are, and whether you can select how many connections to make, but
just for you (CUZ I LIKES YA!!! *smooch*) I'll be adjusting the
limit before the error appears up to, oh, say, four connections
for customers, to avoid this problem. I could drop it completely,
but it is of no benefit to those customers who open six or 20 or
50 connections by default (and adds an extra overhead) and I'd
rather get them to fix their receive window size and number of
connections configuration.
Some pay subscription services have a limit of two connections as
well, returning an error at anything more than this.
The TDC family of companies wishes to take this time to extend our
sincerest apologies and deepest sympathies to all who have been
inconvenienced in their attempts to use this test news swerver.
After our earlier failures in offering it and other machines as a
short-lived replacement for the overworked news.tele.dk, we have
been more cautious than we needed to be in giving you monkeys carte
blanche to trash this machine as you see fit. As this machine has
been running flawlessly for nearly a month, this caution appears to
have been unwarranted. As compensation, we would like to offer a
complimentary 10 seconds of connect time to our flatrate customers
to attempt to bribe you to overlook our gross incompetence, but this
goodwill gesture is currently entangled in red tape imposed by a
number of regulatory bodies and may eventually be denied as an
anti-competitive measure intended to solidify our iron grip upon
our customer base. Instead, we may be required to offer you a
cherry-flavoured all-day sucker, but our crack team
| |
Peter Brodersen (21-01-2001)
| Kommentar Fra : Peter Brodersen |
Dato : 21-01-01 21:13 |
|
On Sun, 21 Jan 2001 16:17:46 +0000 (UTC), Charlie Crackpot
<autocyberretributionmeowbot@plonk.invalid> wrote:
>If your experience is that you need more than two connections to
>run your high-speed line at near-capacity, you need to adjust your
>receive window size upward from the default I see most of 8kB size.
Dette vil vel kun være aktielt i de tilfælde, en bruger har en fortsat
data-stream, fx når han henter pr0n i binaries-grupper.
Men hvis man blot henter en del almindelige nyhedsgrupper samt indlæg
i samme, handler det jo snarere om at news-klienten først henter næste
meddelelse, når den ved om sidste er hentet succesfuldt. Eller fx når
den først har hentet headers for samme grupper. Her foregår der vel en
mere rask tovejskommunikation af nntp-kommandoer, fremfor at der bare
skal ack'es en del tcp-pakker?
--
- Pede
Professionel nørd
Tilhørsforhold/disclaimer: http://www.ter.dk/disclaimer/
| |
Charlie Cat (21-01-2001)
| Kommentar Fra : Charlie Cat |
Dato : 21-01-01 22:19 |
|
:: >If your experience is that you need more than two connections to
:: >run your high-speed line at near-capacity, you need to adjust your
:: >receive window size upward from the default I see most of 8kB size.
But not all the way up to 370k! Unless you want to. More typical
would be 32k to 64k or 128k...
:: Dette vil vel kun være aktielt i de tilfælde, en bruger har en fortsat
:: data-stream, fx når han henter pr0n i binaries-grupper.
Which 90% of the people reading from the test box, and 99% of those
who open six or so connections, are doing. Text just isn't doing
so well anymore.
:: Men hvis man blot henter en del almindelige nyhedsgrupper samt indlæg
:: i samme, handler det jo snarere om at news-klienten først henter næste
:: meddelelse, når den ved om sidste er hentet succesfuldt. Eller fx når
Pipelining. While not documented in the handshaking, this is in fact
supported except when doing a mode change, which usually happens only
at initial connect time or when posting.
If one is receiving a list of messages, there is no reason for the
next command to be given to depend on the results of the command that
is in progress. If the current message to be retrieved returns an
error, that won't change whether the next message is available or
not (unless the error is fatal, which is pretty rare and simply has
the result that all queued commands are lost as the connection closes).
>>> ARTICLE 356
>>> ARTICLE 357
>>> ARTICLE 358
<<< Here comes article 356! Wait for it!
<<< [Headers]
<<< [Body]
<<< .
<<< Here comes article 357! Gotta catch 'em all!
<<< [Headers]
<<< [Body]
<<< .
>>> ARTICLE 359
>>> ARTICLE 360
<<< Sorry d00d, article 358 is CANCELLED!
<<< Here comes article 359! Open wide!
<<< [Headers]
etc.
is a completely legal exchange and usually works with most software.
Saves time on slower connections.
:: den først har hentet headers for samme grupper. Her foregår der vel en
:: mere rask tovejskommunikation af nntp-kommandoer, fremfor at der bare
There are certain parts of the protocol where the following command
does in fact depend on the response returned by the previous command.
For example, the GROUP command must wait for the greeting. The range
of articles requested by the XOVER command will depend on the numbers
returned by the GROUP command. The choice of ARTICLE commands may
depend on the results of GROUP or XOVER, or maybe not XOVER if one
takes the chance of asking for missing articles, but each ARTICLE
command is independent of the previous command.
<<< 200 news.Contactel.CZ zpravy a zmrzlina servis vitame vas! Prosim!
>>> GROUP news.admin.pedophile.barry-bouwsma
<<< 211 150 36328 36492
At this point, you know the low and high numbers in the group. If
you use Outhouse, you will then give the command
>>> XOVER 36328-36492
If you have a newsreader that keeps track that the last message you
read was up to 36416, then you would give the command
>>> XOVER 36417-36492
in order to provide information about the articles to the reader.
This could be simplified for Outhouse, which seems to always download
all headers every time for some design reason unknown to me, with
>>> XOVER 1-
in every group, so it would not depend on the numbers returned, unless
the group has been deleted. Of course, this means downloading several
megabytes of headers in large multimedia groups, but hey, if you use
Outhouse, it knows what you want.
The behaviours I see but do not understand include closing the connection
after downloading an article, then opening a new one to download the
next, which results in inefficient idle time waiting for the new
connection to be established and to get the response to the GROUP
command. And more than three or four connections, which result in
constant total throughput, but a reduced per-connection transfer rate.
:: skal ack'es en del tcp-pakker?
Multiple connections are more useful on h
| |
Claus Schmidt (21-01-2001)
| Kommentar Fra : Claus Schmidt |
Dato : 21-01-01 15:47 |
|
"Charlie Chronic" wrote
> brug news://193.162.153.117/
> ... At present, I see a world reader load of about 25
thx. Make that 26 from now
--
[claus]
| |
Rune Klausen (21-01-2001)
| Kommentar Fra : Rune Klausen |
Dato : 21-01-01 23:40 |
|
"Charlie Chronic" <autocyberretributionmeowbot@plaank.invalid> wrote in
message news:94afvt$pct$1@news.talkline.nl...
> :: Kunne ikke forstå at jeg manglede filer både i dk.binaer og
dk.binaer.musik
> :: Det har ALDRIG været et problem før nu så er der nogen der ved hvad der
er
>
> brug news://193.162.153.117/
> eller SSL snews://193.162.153.117/ (read-only)
>
> news.tele.dk is on its way out, and will be taken down in a week for
> what should be a short time. It has not been able to keep up with
> the full news volume of ~250GB/day for some months now, and the machine
> ahead of it (in the URL above) does not have retention to keep all
> posts during peak hours. Upstream, limited bandwidth results in about
> 70GB out of these 250GB being lost.
>
> The machine at news://193.162.153.117:143/ (read-only) should have all
> the missing articles.
Well it's doesn't still missing a lot of heeads in binaries groups eg.
alt.binaries.x-files
-Rune
| |
Charlie Cherubim (22-01-2001)
| Kommentar Fra : Charlie Cherubim |
Dato : 22-01-01 00:29 |
|
:: > posts during peak hours. Upstream, limited bandwidth results in about
:: > 70GB out of these 250GB being lost.
:: >
:: > The machine at news://193.162.153.117:143/ (read-only) should have all
:: > the missing articles.
::
:: Well it's doesn't still missing a lot of heeads in binaries groups eg.
:: alt.binaries.x-files
True. As I noted, while a full feed is near 250GB per day, more or
less, trying to push all the data over a single 100Mbit/sec ethernet
card, when we are attempting to deliver about 350Mbit/sec at peak
times, is resulting in only 170GB/day reaching our filtering machine.
This shows as missing articles in the non-danish binaries froups.
Missing text articles are probably due to a VM problem, as can be seen
by comparing two hierarchies carrying the same gated mailing lists
(such as the missing posts in sol.lists.* that can be seen in muc.lists.*
and vice versa)
At long last, gigabit ethernet cards have arrived, but turned out to
be incompatible with the operating system running on the binaries
box, so our attempts to solve this problem late last week failed.
I'm not sure if attempts will be made to obtain Tigon chipset gigabit
ethernet cards that other sites are using successfully, or to bribe
a local ethernet driver hacker into porting Linux source code to work
on the machines, or to wait until the maintainer of the Intel driver
k0de is able to fix it to support these new cards, which he said he
would try to do this coming week.
When this, the most serious performance bottleneck is solved, there
should be adequate bandwidth available between the machines so this
problem of missing articles (several dozens of thousands per day or
more) will finally be solved, at least, for a few more months. This
should (if there are no bottlenecks that appear elsewhere) result in
full completion on the test machine you tried.
The condition of news.tele.dk is slightly better than before, because
of the weekend when people pretend to have lives, and so there should
be many fewer missing articles there than earlier in the week. But
because of the higher Sunday reader load, backlogs have formed again
(and they will be even larger when the missing 50 to 70GB/day are able
to make it downstream).
We hope that a voluntary migration of Readers Who Care from news.tele.dk
to the test machine at news://193.162.153.117/ will result in a much
lower load on news.tele.dk and better performance. This migration is
going slowly, with perhaps 20 customers max reading and posting via
this test machine at any given time, but it's a start. This number
should increase as the planned downtime of news.tele.dk this coming
Friday approaches, and we hope that binary-sucking customers who do
discover this test machine (if it stays up) are satisfied enough to
continue to use it while the MonsterBox is being built.
The TDC family of companies wishes to apologize to afflicted customers
who were hoping that the promised delivery of the gigabit ethernet
cards would have solved the most serious of the problems last week.
There is probably a lesson to be learned from this, like, don't count
your chickens before they rip your lips off. Thank you for your patience.
| |
tjah@lphunde.dk (22-01-2001)
| Kommentar Fra : tjah@lphunde.dk |
Dato : 22-01-01 10:51 |
|
On Sun, 21 Jan 2001 23:29:29 +0000 (UTC), Charlie Cherubim
<autocyberretributionmeowbot@plonk.invalid> wrote:
> This migration is
>going slowly, with perhaps 20 customers max reading and posting via
>this test machine at any given time, but it's a start.
En årsag til manglende migration:
"Error reported by server: 502 You have no hostname to match your IP
in DNS. Goodbye."
--
MVH
Bruno Christensen
| |
Jan Chrillesen (22-01-2001)
| Kommentar Fra : Jan Chrillesen |
Dato : 22-01-01 13:06 |
|
tjah@lphunde.dk wrote:
> En årsag til manglende migration:
>
> "Error reported by server: 502 You have no hostname to match your IP
> in DNS. Goodbye."
Så få en in-addr.arpa registrering af dine IP adresser. Det er ikke et
problem med newsserveren.
/jan
| |
Charlie Chiseltooth (22-01-2001)
| Kommentar Fra : Charlie Chiseltooth |
Dato : 22-01-01 13:24 |
|
:: En årsag til manglende migration:
:: "Error reported by server: 502 You have no hostname to match your IP
Access to both ports 563 and 143, for both TDC customers and non-
customers, requires a DNS PTR entry so that I can keep track of
usage of the swerver, and where readers are coming from.
All TDC customers should have access to the normal port 119, that
is closed to non-customers, even if you do not have a reverse DNS
entry. Try this link: news://193.162.153.117/
There is at least one other customer with no reverse DNS who is
able to connect and read from port 119, so it should work for you.
If it does not, <usenet@tdk.net> wants to know about it.
If you do not get an immediate response, there is probably a problem
with your IP's DNS configuration that should also cause delays when
connecting to news.tele.dk too. You probably want to report this to
<usenet@tdk.net> to be forwarded to the appropriate person, unless
it is policy to have accounts with DNS PTR lookups that timeout.
If as a customer, you get an error when connecting to port 119 that
says that news02 is not a public news swerver, then you probably have
an IP address that is missing from our too-long list of IP ranges
that belong to TDC or one of its tentacles throughout europe. Again,
please contact <usenet@tdk.net> so we can update our access list.
However, the IP address you used to make this post seems to be on
our guest list (twice even!) so you should have no problems. With
port 119, that is. True, you can't use the SSL test port now...
Non-customers of TDC who have no reverse DNS entry are out of luck,
as port 119 is closed to you, and the other two ports require a
functional PTR entry. Sorry, but a name tells me much more than
just an IP, and whaddya expect for free anyhows?
| |
Lars Balker Rasmusse~ (22-01-2001)
| Kommentar Fra : Lars Balker Rasmusse~ |
Dato : 22-01-01 14:05 |
|
Charlie Chiseltooth <autocyberretributionmeowbot@plonk.invalid> writes:
> and whaddya expect for free anyhows?
Beer?
--
Lars Balker Rasmussen "Woo hoo!?"
| |
Charlie Cheapskate (22-01-2001)
| Kommentar Fra : Charlie Cheapskate |
Dato : 22-01-01 14:36 |
|
"lars@balker.org" <lars@balker.org> skrev i en
nyhedsmeddelelse:ubsszkaa4.fsf@balker.org...
:: > and whaddya expect for free anyhows?
::
:: Beer?
Oh, a bit slow on the uptake, are we? You shall find that the topic has
already been covered yesterday along with several other highly technical
points in a heavily complex thread entitled `anal biverkninger' in the
customer support and technical discussion froup dk.snak.mudderkastning.
You are probably going to attempt to hold me hostage until you get the
pleasure next month of hearing me ask for ten litres of hemp oil, as
you lost your chance for Irish Kraftige, sorry to say. (Well, not really)
| |
Peter Alfredsen (19-01-2001)
| Kommentar Fra : Peter Alfredsen |
Dato : 19-01-01 23:54 |
|
On Fri, 19 Jan 2001 22:45:11 +0100, "Henrik Bruun-Petersen"
<news@nytliv.net> wrote:
>Hej
>Hvad sker der med denne server?
>
>Kunne ikke forstå at jeg manglede filer både i dk.binaer og dk.binaer.musik
>prøvede så lige sunsite for sjov og der lå de alle og flere til.
>
>Det har ALDRIG været et problem før nu så er der nogen der ved hvad der er
>sket?
Den har taget et styrt. Trafikken på USENET er steget for meget
til, at den kan følge med.
--
Peter Alfredsen (peteralf@fabel.dk) |"A name indicates what we seek. An
"We reject kings, presidents and voting.| address indicates where it is. A
We believe in rough consensus and | route indicates how we get there."
running code." -- David Clark (MIT) | -- Jon Postel
| |
|
|