Re: [JDEV] Autoupdate

2002-07-16 Thread DJ Adams

On Tue, Jul 16, 2002 at 09:39:05AM -0600, Ben Schumacher wrote:
> It relies on S2S. The current autoupdate protocol (which could be cleaned
> up a little) is designed so that clients send a presence packet with the
> following format:
>   
> 
> Where '1251235123' is a unique identifier, '0.1.2.3' is the current
> version number, and update.jabber.org is (of course) the target host. If

FWIW, there's some more background here:

http://www.pipetree.com/jabber/update.html

cheers
dj
___
jdev mailing list
[EMAIL PROTECTED]
http://mailman.jabber.org/listinfo/jdev



RE: [JDEV] Autoupdate

2002-07-16 Thread Jonathan Pobst

I think he meant how do you tell the server when a new version is available.
For the update.jabber.org, per my understanding, this was tied to
maintaining a record of your client on the JabberCentral.org website, and
telling it (via I assume a web interface), that a new version was available.
That record was then queried by the server agent on jabber.org and compared
to the version number the client sent in to see if a newer version was
available.

hope that helps,
pobst

-Original Message-
From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED]]On Behalf Of
Sami Haahtinen
Sent: Tuesday, July 16, 2002 11:26 AM
To: [EMAIL PROTECTED]
Subject: Re: [JDEV] Autoupdate


On Tue, Jul 16, 2002 at 04:04:29AM -0400, Kriggs wrote:
> How does the server a client is connected to know that there is a new
version
> available? Does the server query some database when the client asks? From
> what I understand, the server DOES have something to do with this...

like Ben already explained, you can send a certain type of presence to
the agent, with my agent we use format like:


  


where client is the name of the client in question (which is marked to
the database) and 50 is the current version of the client. the server
will return new version(s) as a result to that query

ofcourse, this is the case with my agent, but the common idea is that
the client sends a request and tells it's current version with it. you
could also do it so that you trigger the agent to probe you, but just
telling the agent your version is the simplest and fastest way..

Sami

--
  -< Sami Haahtinen >-
  -[ Notify immediately if you do not receive this message ]-
-< 2209 3C53 D0FB 041C F7B1  F908 A9B6 F730 B83D 761C >-
___
jdev mailing list
[EMAIL PROTECTED]
http://mailman.jabber.org/listinfo/jdev

___
jdev mailing list
[EMAIL PROTECTED]
http://mailman.jabber.org/listinfo/jdev



Re: [JDEV] Autoupdate

2002-07-16 Thread Sami Haahtinen

On Tue, Jul 16, 2002 at 04:04:29AM -0400, Kriggs wrote:
> How does the server a client is connected to know that there is a new version 
> available? Does the server query some database when the client asks? From 
> what I understand, the server DOES have something to do with this...

like Ben already explained, you can send a certain type of presence to
the agent, with my agent we use format like:


  


where client is the name of the client in question (which is marked to
the database) and 50 is the current version of the client. the server
will return new version(s) as a result to that query

ofcourse, this is the case with my agent, but the common idea is that
the client sends a request and tells it's current version with it. you
could also do it so that you trigger the agent to probe you, but just
telling the agent your version is the simplest and fastest way..

Sami

-- 
  -< Sami Haahtinen >-
  -[ Notify immediately if you do not receive this message ]-
-< 2209 3C53 D0FB 041C F7B1  F908 A9B6 F730 B83D 761C >-
___
jdev mailing list
[EMAIL PROTECTED]
http://mailman.jabber.org/listinfo/jdev



Re: [JDEV] 100k users...

2002-07-16 Thread Ben Schumacher

What was your login rate? What backend where you using? Where all the
connections fully authenticated? Sorry, just some questions I felt like
asking. I imagine logging into 100K users over the course of 10 minutes
would have killed the server (166-167 logins per second) if you where
running xdb_file. All Linux/Intel boxes?

Just curious.

bs.

On Tue, 16 Jul 2002, Ryan Eatmon wrote:
> This was just passive connections.  I can't find any good load test
> programs that are easy to use, and haven't had time to sit down and
> write one.

___
jdev mailing list
[EMAIL PROTECTED]
http://mailman.jabber.org/listinfo/jdev



Re: [JDEV] Autoupdate

2002-07-16 Thread Ben Schumacher

On Tue, 16 Jul 2002, Kriggs wrote:
> How does the server a client is connected to know that there is a new version
> available? Does the server query some database when the client asks? From
> what I understand, the server DOES have something to do with this...

It relies on S2S. The current autoupdate protocol (which could be cleaned
up a little) is designed so that clients send a presence packet with the
following format:
  

Where '1251235123' is a unique identifier, '0.1.2.3' is the current
version number, and update.jabber.org is (of course) the target host. If
your server doesn't have S2S configured, then autoupdate won't work.

Cheers,

bs.

___
jdev mailing list
[EMAIL PROTECTED]
http://mailman.jabber.org/listinfo/jdev



RE: [JDEV] id attr in message packets

2002-07-16 Thread Constantin Nickonov

You could also make the 'thread' attribute hold more than one value (choose
your delimeter), i.e.,

  root;parent;grandparent;etc

Since your client need only communicate with others like it, you can both
generate and make sense of this value at any point.

> -Original Message-
> From: Sean Wheeler [mailto:[EMAIL PROTECTED]]
> Sent: Monday, July 15, 2002 8:58 PM
> To: [EMAIL PROTECTED]
> Subject: Re: [JDEV] id attr in message packets
> 
> 
> 
> Thanks for the explanation. Your point about message ordering is well
> taken, I imagine many would complain if servers did not ensure ordered
> arrival.  I now understand the id tag is not really for pinning down
> messages in the conversation but more as a client 
> convenience. This sucks,
> so since I'm not building an IM client and I don't have to 
> interact with
> any other clients in my project I might have to veer from the 
> standard for
> my purposes.
> 
> I think my problem does exist and in fact is a general one if you
> want/need to think of an interchange as a nested tree 
> structure. A reply
> is a child node, and your point is well taken that preserving 
> ordering of
> messages from a single client helps a lot. But since the tree 
> is deeply
> nested, how can you make reference to any node in the tree?  My first
> interpretation of what a thread "should be" is a reference to the root
> node.  This is actually the case, but in a more limited way, since the
> protocol only creates a tree of depth 1.
> 
> One scheme allowing 'n-depth tree' conversations is for a 
> child to be able
> to identify its parent, and this is what I was looking at the 
> id tag for.
> Another more elegant scheme is to use a tag that refers to 
> the root node
> of the implicit tree structure, call it 'root', or 'conversation' or
> whatever, and stick it in an x tag in another namespace.  Given the
> protocol and the common usage of id, this creates the kind of tree I'm
> looking for, but there's no assurance that the tree structure is
> accurately represented if a third party that doesnt have the full
> conversation history is thrown in.  For this the implicit root of the
> conversation would have to identify itself somehow, either by adding
> another tag in the x namespace, setting thread and root to 
> the same value,
> etc...
> 
> I'll have to take a closer look at JEP-0033.
> 
> Sean
> 
> 
> On Mon, 15 Jul 2002, Ben Schumacher wrote:
> 
> > On Sun, 14 Jul 2002, Sean Wheeler wrote:
> > > On Sun, 14 Jul 2002, Jeremy Nickurak wrote:
> > > >
> > > > The id attribute is intended to be used so that the 
> sender and receiver
> > > > can refer to particular messages. For example, someone 
> might implement a
> > > > "addendum" message type/namespace that changes some 
> element of the
> > > > previous message:
> > > >
> > > > 
> > > >
> > > >   
> > > >  See also http://somepage.com/ for more details.
> > > >   
> > > >
> > > > 
> > > >
> > > > That way, the  tag's "id" attribute can be 
> uniquesly associated
> > > > with a particular mesage bar@media has already sent to 
> foo@media.
> > >
> > > Thanks!
> > >
> > > Yes, this is precisely the kind of thing i'm looking for. 
>  I also need the
> > > ability to refer to the reply that foo sends, and I can't 
> really figure
> > > out how to do this without some non-standard crutch.
> > >
> > > Maybe a rough ascii diagram helps.
> > >
> >[...snip...]
> >
> > Sean-
> >
> > I think you are struggling to find a solution to a problem 
> that doesn't
> > really exist. Since all Jabber messages arrive in order (at 
> least with
> > current server implementations), there is no need to know 
> explicitly that
> > a user sends two messages as a reply to a single message 
> from the other
> > end. It is implicit that if a message is sent by bar to 
> foo, and then foo
> > sends two responses on the same , they are 
> obviously "replying" to
> > the original message by bar. In addition, id numbers used 
> in most clients
> > don't relate directly to conversations. Most client 
> libraries will simply
> > increase the value of the id each time a packet is sent. id's are
> > generally not intended for this purpose. In fact, there is 
> no guarantee
> > that other clients will behave in the way you have 
> described, and I would
> > even go so far as to venture a guess that very few currently do.
> >
> > For example, considering the following exchange:
> >
> > 1) foo sends a message to bar with id='1'
> > 2) bar replies to message from foo without an id
> > 3) foo sends an iq request in iq:register to component x with id='2'
> > 4) foo replies to message from bar with id='3'
> > 5) component x sends iq result to foo with id='2'
> > 6) bar replies to message from foo without an id
> >
> > ... etc.
> >
> > You'll notice that bar does not use id's in its message 
> packets. This
> > behavior can be expected, because it is considered an 
> optional attribute
> > in the spec. (see: 
> http://jabber.org/ietf/dr

Re: [JDEV] 100k users...

2002-07-16 Thread anghel

I am also interested in the Farming How-To. Could you please post a message on 
jdev when you will make it public? I think that there are a lot of people out 
there interested in this subject.

Thanx

Dan

Citazione Ryan Eatmon <[EMAIL PROTECTED]>:

> 
> All will be made clear when I can find time to finish my Jabber Server 
> Farming How-To...  Must find time...
> 
> I'm going to go in details about the setup, tweaks, versions, etc...
> 
> Basic hardware overview:
> 
> c2s machines are dual proc dual nic, each running two jadc2s.
> Five of those.
> one main jabberd they all talk to.
> 
> 
> Julian Missig wrote:
> 
> > Any info you could give us on the hardware, the OS, the tweaks you had 
> > to do the the OS?
> > 
> > Julian
> > 
> > Ryan Eatmon wrote:
> > 
> >>
> >> Just to let everyone know the good news, I just got a 1.4.2 server 
> >> plus jadc2s up to 100k concurrent connections.  And there was still 
> >> room to go higher, I just ran out of time and test boxes.
> >>
> > 
> > ___
> > jdev mailing list
> > [EMAIL PROTECTED]
> > http://mailman.jabber.org/listinfo/jdev
> 
> 
> ___
> jdev mailing list
> [EMAIL PROTECTED]
> http://mailman.jabber.org/listinfo/jdev
> 


___
jdev mailing list
[EMAIL PROTECTED]
http://mailman.jabber.org/listinfo/jdev



Re: [JDEV] 100k users...

2002-07-16 Thread anghel

So you have about 20.000 users per machine and 10.000 users per jadc2s? Are 
you using a load balancer or just connect the users directly to each jadc2s? I 
didn't understand if you are already using farming for the jadc2s or not. In 
this case I think that the load balancing could(should) be done through the 
farming mechanism. And one more question, what is "dual nic"?:)

thanx

Dan


Citazione Ryan Eatmon <[EMAIL PROTECTED]>:

> 
> All will be made clear when I can find time to finish my Jabber Server 
> Farming How-To...  Must find time...
> 
> I'm going to go in details about the setup, tweaks, versions, etc...
> 
> Basic hardware overview:
> 
> c2s machines are dual proc dual nic, each running two jadc2s.
> Five of those.
> one main jabberd they all talk to.
> 
> 
> Julian Missig wrote:
> 
> > Any info you could give us on the hardware, the OS, the tweaks you had 
> > to do the the OS?
> > 
> > Julian
> > 
> > Ryan Eatmon wrote:
> > 
> >>
> >> Just to let everyone know the good news, I just got a 1.4.2 server 
> >> plus jadc2s up to 100k concurrent connections.  And there was still 
> >> room to go higher, I just ran out of time and test boxes.
> >>
> > 
> > ___
> > jdev mailing list
> > [EMAIL PROTECTED]
> > http://mailman.jabber.org/listinfo/jdev
> 
> 
> ___
> jdev mailing list
> [EMAIL PROTECTED]
> http://mailman.jabber.org/listinfo/jdev
> 


___
jdev mailing list
[EMAIL PROTECTED]
http://mailman.jabber.org/listinfo/jdev



Re: [JDEV] 100k users...

2002-07-16 Thread Ryan Eatmon


All will be made clear when I can find time to finish my Jabber Server 
Farming How-To...  Must find time...

I'm going to go in details about the setup, tweaks, versions, etc...

Basic hardware overview:

c2s machines are dual proc dual nic, each running two jadc2s.
Five of those.
one main jabberd they all talk to.


Julian Missig wrote:

> Any info you could give us on the hardware, the OS, the tweaks you had 
> to do the the OS?
> 
> Julian
> 
> Ryan Eatmon wrote:
> 
>>
>> Just to let everyone know the good news, I just got a 1.4.2 server 
>> plus jadc2s up to 100k concurrent connections.  And there was still 
>> room to go higher, I just ran out of time and test boxes.
>>
> 
> ___
> jdev mailing list
> [EMAIL PROTECTED]
> http://mailman.jabber.org/listinfo/jdev


___
jdev mailing list
[EMAIL PROTECTED]
http://mailman.jabber.org/listinfo/jdev



Re: [JDEV] 100k users...

2002-07-16 Thread Ryan Eatmon


This was just passive connections.  I can't find any good load test 
programs that are easy to use, and haven't had time to sit down and 
write one.

Dan Anghel wrote:

> Hi,
> 
> I just want to ask if the 100k concurrent connections were passive or active? 
> If you tried some load tests with message sending, etc.
> 
> Thanx
> Dan
> 
> On Tuesday 16 July 2002 05:50 am, you wrote:
> 
>>Just to let everyone know the good news, I just got a 1.4.2 server plus
>>jadc2s up to 100k concurrent connections.  And there was still room to
>>go higher, I just ran out of time and test boxes.
>>
> ___
> jdev mailing list
> [EMAIL PROTECTED]
> http://mailman.jabber.org/listinfo/jdev
> 


___
jdev mailing list
[EMAIL PROTECTED]
http://mailman.jabber.org/listinfo/jdev



Re: [JDEV] 100k users...

2002-07-16 Thread Dan Anghel

Hi,

I just want to ask if the 100k concurrent connections were passive or active? 
If you tried some load tests with message sending, etc.

Thanx
Dan

On Tuesday 16 July 2002 05:50 am, you wrote:
> Just to let everyone know the good news, I just got a 1.4.2 server plus
> jadc2s up to 100k concurrent connections.  And there was still room to
> go higher, I just ran out of time and test boxes.
___
jdev mailing list
[EMAIL PROTECTED]
http://mailman.jabber.org/listinfo/jdev



Re: [JDEV] Autoupdate

2002-07-16 Thread Kriggs

On July 16, 2002 02:02 am, you wrote:
> On Mon, Jul 15, 2002 at 04:35:08PM -0400, Kriggs wrote:
> > So it is the client's job to talk directly to the autoupdate service
> > provider?
>
> yes, the autoupdate-agent has no way of knowing which clients are in use
> everywhere in the world.
>
> also it is a much better thing to send the autoupdate request through
> jabber than to poll a website or something, as the client has already
> made it's connection to jabber server, you have less work and more
> reliable way of passing on the request.

Hmmm. Maybe I should ask my question a different way.

How does the server a client is connected to know that there is a new version 
available? Does the server query some database when the client asks? From 
what I understand, the server DOES have something to do with this...
___
jdev mailing list
[EMAIL PROTECTED]
http://mailman.jabber.org/listinfo/jdev