Mailing List Archive: Linux-HA: Users

Question

Index | Next | Previous | View Threaded

hack at altavista

Sep 9, 1999, 6:57 AM

Post #1 of 14 (1774 views)
Permalink

Question

Hi folks,

The ISP in the Netherlands that I'm currently working for, want me to implement a scalable, high available (load balancing) mailserver. I have been looking around and think the best way to do this is by having a daemon listen to port 110 (POP3). This daemon will first ask the user's name and then handle the traffic between the selected mailserver (the one with, for example, the lowest traffic) and the user. I have a few questions, maybe you folks can help me:

- Since this daemon will be a SPOF (single point of failure), what do I have to pay extra attention to?

- Am I right when I think that there will be a maximum of 30,000 users at the same time (since the daemon will have to open both sockets to the user and the mailserver)? Can DNS be used as a solution for this?

- Does anyone know of a daemon (or any software in general) that does something similar to what I want to do or an URL with more info on this subject (POP3 specific)?

Any help is appreciated!

Greetings, Norbert de Jonge

----------------------------------------------------------------
Get your free email from AltaVista at http://altavista.iname.com

lmb at teuto

Sep 9, 1999, 7:16 AM

Post #2 of 14 (1734 views)
Permalink

Question [In reply to]

On 1999-09-09T09:57:40,
hack [at] altavista said:

> The ISP in the Netherlands that I'm currently working for, want me to
> implement a scalable, high available (load balancing) mailserver. I have
> been looking around and think the best way to do this is by having a daemon
> listen to port 110 (POP3). This daemon will first ask the user's name and
> then handle the traffic between the selected mailserver (the one with, for
> example, the lowest traffic) and the user. I have a few questions, maybe you
> folks can help me:

This is what "LinuxVirtualServer" is abut. You may want to check out
http://www.LinuxVirtualServer.org/, subscribe to the mailing list, and repost
your question there ;-)

Sincerely,
Lars Marowsky-Brée

--
Lars Marowsky-Brée
Network Management

teuto.net Netzdienste GmbH - DPN Verbund-Partner

alanr at bell-labs

Sep 9, 1999, 7:17 AM

Post #3 of 14 (1724 views)
Permalink

Question [In reply to]

hack [at] altavista wrote:
>
> Hi folks,
>
> The ISP in the Netherlands that I'm currently working for, want me to implement a scalable, high available (load balancing) mailserver. I have been looking around and think the best way to do this is by having a daemon listen to port 110 (POP3). This daemon will first ask the user's name and then handle the traffic between the selected mailserver (the one with, for example, the lowest traffic) and the user. I have a few questions, maybe you folks can help me:
>
> - Since this daemon will be a SPOF (single point of failure), what do I have to pay extra attention to?
>
> - Am I right when I think that there will be a maximum of 30,000 users at the same time (since the daemon will have to open both sockets to the user and the mailserver)? Can DNS be used as a solution for this?
>
> - Does anyone know of a daemon (or any software in general) that does something similar to what I want to do or an URL with more info on this subject (POP3 specific)?
>
> Any help is appreciated!
>
> Greetings, Norbert de Jonge

Sounds like you might want to look into the Linux Virtual Server project.
They're doing about a similar thing for web servers, which might work directly
for your application. The can work together with Linux-HA software (heartbeat)
to make the front end highly available. You can find it pointed to off the
Linux-HA home page, or can go directly there:

http://proxy.iinchina.net/~wensong/ippfvs/

-- Alan Robertson
alanr [at] bell-labs

Andy at activetools

Sep 16, 1999, 1:51 AM

Post #4 of 14 (1726 views)
Permalink

Question [In reply to]

hack [at] altavista wrote:
>
> The ISP in the Netherlands that I'm currently working for, want me to implement a
> scalable, high available (load balancing) mailserver. I have been looking around
> and think the best way to do this is by having a daemon listen to port 110 (POP3).

Hi Norbet,

I suggest that you look also at:

http://beta.turbolinux.com/cluster/

LPA

PS: If you have any further questions, please do not hesitate to contact
me.

alanr at bell-labs

Sep 16, 1999, 5:00 AM

Post #5 of 14 (1728 views)
Permalink

Question [In reply to]

"Andrej Brodnik (Andy)" wrote:
>
> hack [at] altavista wrote:
> >
> > The ISP in the Netherlands that I'm currently working for, want me to implement a
> > scalable, high available (load balancing) mailserver. I have been looking around
> > and think the best way to do this is by having a daemon listen to port 110 (POP3).
>
> Hi Norbet,
>
> I suggest that you look also at:
>
> http://beta.turbolinux.com/cluster/

Since the TurboLinux folks don't make this as clear as they could...

The core function of their product is really a subset of the code of the Linux
Virtual Server project repackaged for commercial use.

-- Alan Robertson
alanr [at] bell-labs

Andy at activetools

Sep 16, 1999, 5:13 AM

Post #6 of 14 (1729 views)
Permalink

Question [In reply to]

Alan Robertson wrote:
>
> "Andrej Brodnik (Andy)" wrote:
> >
> > hack [at] altavista wrote:
> > >
> > > The ISP in the Netherlands that I'm currently working for, want me to implement a
> > > scalable, high available (load balancing) mailserver. I have been looking around
> > > and think the best way to do this is by having a daemon listen to port 110 (POP3).
> >
> > Hi Norbet,
> >
> > I suggest that you look also at:
> >
> > http://beta.turbolinux.com/cluster/
>
> Since the TurboLinux folks don't make this as clear as they could...
>
> The core function of their product is really a subset of the code of the Linux
> Virtual Server project repackaged for commercial use.

Hi Alan,

I don't want YAFW (yet another flame war), but I think I have to say
this
for your and other readers information. Namely, that your claim might
used
to be true in previous versions (I was not involved in that stage), but
in
version 4 this is not true any more. It is programmed completely from
scratch.
For more details see

http://beta.turbolinux.com/cluster/doc-v4/faq/faq-1.html#ss1.1

To summarize, TurboCluster is designed and programmed from scratch and
hence it
is known where it is headeing and what improvements are forthcomming.
But to make
the message short -- the product is meant for heavy duty 24/7 service at
the ISP
servers with a simple and intuitive (GUI) support for the administrator.

LPA

alanr at bell-labs

Sep 16, 1999, 5:22 AM

Post #7 of 14 (1726 views)
Permalink

Question [In reply to]

"Andrej Brodnik (Andy)" wrote:
>
> Alan Robertson wrote:
> >
> > "Andrej Brodnik (Andy)" wrote:
> > >
> > > hack [at] altavista wrote:
> > > >
> > > > The ISP in the Netherlands that I'm currently working for, want me to implement a
> > > > scalable, high available (load balancing) mailserver. I have been looking around
> > > > and think the best way to do this is by having a daemon listen to port 110 (POP3).
> > >
> > > Hi Norbet,
> > >
> > > I suggest that you look also at:
> > >
> > > http://beta.turbolinux.com/cluster/
> >
> > Since the TurboLinux folks don't make this as clear as they could...
> >
> > The core function of their product is really a subset of the code of the Linux
> > Virtual Server project repackaged for commercial use.
>
> Hi Alan,
>
> I don't want YAFW (yet another flame war), but I think I have to say
> this
> for your and other readers information. Namely, that your claim might
> used
> to be true in previous versions (I was not involved in that stage), but
> in
> version 4 this is not true any more. It is programmed completely from
> scratch.
> For more details see
>
> http://beta.turbolinux.com/cluster/doc-v4/faq/faq-1.html#ss1.1
>
> To summarize, TurboCluster is designed and programmed from scratch and
> hence it
> is known where it is headeing and what improvements are forthcomming.
> But to make
> the message short -- the product is meant for heavy duty 24/7 service at
> the ISP
> servers with a simple and intuitive (GUI) support for the administrator.
>
> LPA

That's good information. Sounds like a quote from a Microsoft press release :-)

It's unfortunate that they also made this claim about their earlier releases...
It certainly wasn't true then.

-- Alan Robertson
alanr [at] bell-labs

alan at lxorguk

Sep 16, 1999, 5:27 AM

Post #8 of 14 (1718 views)
Permalink

Question [In reply to]

> To summarize, TurboCluster is designed and programmed from scratch and
> hence it
> is known where it is headeing and what improvements are forthcomming.
> But to make
> the message short -- the product is meant for heavy duty 24/7 service at
> the ISP
> servers with a simple and intuitive (GUI) support for the administrator.

Was that a deliberate attempt to impune Wensong's code or just crass choice
of words.

ipvs _is_ already running in heavy load situations. I've also subjected it to
a security verification involving beating on it with 100Mbits of simulated
hostile attackers.

Alan

lmb at teuto

Sep 16, 1999, 5:38 AM

Post #9 of 14 (1726 views)
Permalink

Question [In reply to]

On 1999-09-16T14:13:56,
"Andrej Brodnik (Andy)" <Andy [at] activetools> said:

> To summarize, TurboCluster is designed and programmed from scratch and hence
> it is known where it is headeing and what improvements are forthcomming.

I can't follow the "hence" here.

Anyway: The kernel patches are going to be available under the GPL, as they
must be. The monitoring software is not a problem at all - I wrote a
quick&dirty solution in one day and am working on a really pretty solution as
time permits.

> But to make the message short -- the product is meant for heavy duty 24/7
> service at the ISP servers with a simple and intuitive (GUI) support for the
> administrator.

Uhm. The LinuxVirtualServer project _is_ meant for heavy duty 24/7 service and
HAS already been deployed for those tasks.

No need to speak about the future here ;-)

The GUI tools - I had a bird tell me that Red Hat might be working on
something which you can find in Raw Hide or Lorax.

Sincerely,
Lars Marowsky-Brée

--
Lars Marowsky-Brée
Network Management

teuto.net Netzdienste GmbH

sct at redhat

Sep 17, 1999, 7:31 AM

Post #10 of 14 (1733 views)
Permalink

Question [In reply to]

Hi,

On Thu, 16 Sep 1999 14:38:42 +0200, Lars Marowsky-Bree <lmb [at] teuto>
said:

> The GUI tools - I had a bird tell me that Red Hat might be working on
> something which you can find in Raw Hide or Lorax.

I've no idea about the GUI bit but:

ftp://sunsite.doc.ic.ac.uk/pub/Mirrors/ftp.redhat.com/lorax/i386/RedHat/RPMS
-rw-r--r-- 1 root root 83267 Aug 16 19:52 heartbeat-0.4.1-3.i386.rpm
-rw-r--r-- 1 root root 8799 Aug 29 23:14 ipvsadm-1.1-1.i386.rpm

the core heartbeat and ipvs code already appears to be in the public
lorax release.

--Stephen

wanger at redhat

Sep 17, 1999, 9:05 AM

Post #11 of 14 (1731 views)
Permalink

Question [In reply to]

On Thu, 16 Sep 1999 14:38:42 +0200, Lars Marowsky-Bree wrote: =

>The GUI tools - I had a bird tell me that Red Hat might be working on
>something which you can find in Raw Hide or Lorax.

Ok, I'm the "bird". :) Yes, we've got a very nice GUI tool for LVS, =

client service monitoring, load balancing, and heartbeating put =

together. It also has a daemon that runs in the background keeping =

track of all this stuff. I know, I know, I should have released this =

sooner, but for a multitude of reasons, I haven't yet. I'm currently =

traveling, but early next week, when I return to the office, I'll be =

putting the new stuff out.

I'll post back here once I've got it available. Oh, and BTW, it is =

_all_ GPL'd code. :)

Mike

-----------------------------------------------------------------------
Mike Wangsmo Red Hat, Inc =

"I've seen this before in Montana! Its snowing, nobody lick a flag
pole" -- Peggy Hill

lists at thohal

Jan 13, 2010, 1:40 PM

Post #12 of 14 (1697 views)
Permalink

Re: Question [In reply to]

Hello Mark,

Am Mittwoch, den 13.01.2010, 15:57 -0500 schrieb Yeary, Mark A.:
> Hello All,
>
> I'm looking for advice. I'm using drbd-8.3.5. I have created a
> drbd cluster active/active with gfs2 on the two nodes. Can I access
> this cluster from other nodes besides the two I've created drbd on and
> maintain sync. In other words, can I add four nodes to the gfs2 cluster
> (only two are the drbd) and access it from the four nodes using gfs2 and
> maintain sync? Note: I am not increasing my drbd cluster size (still
> two nodes), I am just adding gfs2 nodes to make a total of four.

But how should ne these "just-gfs-nodes" be able to see or attach these
storage, since your drbd-device is _not_ shared to this hosts. Your
Nodes are not able to attch

I would suggest, you want to export your drbd-device through aoe, iscsi,
fcoe or whatever.

A high-available storage-server through linux-ha could be the name to
the next topic you open ;-)

> I want to say that drbd is great :-) I have been using it for ten
> years. Philip, you won't remember me but I still owe you a drink :-)
>
> Mark

hth,

Thomas

> _______________________________________________
> Linux-HA mailing list
> Linux-HA [at] lists
> http://lists.linux-ha.org/mailman/listinfo/linux-ha
> See also: http://linux-ha.org/ReportingProblems

_______________________________________________
Linux-HA mailing list
Linux-HA [at] lists
http://lists.linux-ha.org/mailman/listinfo/linux-ha
See also: http://linux-ha.org/ReportingProblems

dejanmm at fastmail

Jan 14, 2010, 5:04 AM

Post #13 of 14 (1686 views)
Permalink

Re: Question [In reply to]

Hi,

On Wed, Jan 13, 2010 at 03:57:50PM -0500, Yeary, Mark A. wrote:
> Hello All,
>
> I'm looking for advice. I'm using drbd-8.3.5. I have created a
> drbd cluster active/active with gfs2 on the two nodes. Can I access
> this cluster from other nodes besides the two I've created drbd on and
> maintain sync. In other words, can I add four nodes to the gfs2 cluster
> (only two are the drbd) and access it from the four nodes using gfs2 and
> maintain sync? Note: I am not increasing my drbd cluster size (still
> two nodes), I am just adding gfs2 nodes to make a total of four.

Perhaps to resend to the drbd list.

Cheers,

Dejan

> I want to say that drbd is great :-) I have been using it for ten
> years. Philip, you won't remember me but I still owe you a drink :-)
>
> Mark
> _______________________________________________
> Linux-HA mailing list
> Linux-HA [at] lists
> http://lists.linux-ha.org/mailman/listinfo/linux-ha
> See also: http://linux-ha.org/ReportingProblems
_______________________________________________
Linux-HA mailing list
Linux-HA [at] lists
http://lists.linux-ha.org/mailman/listinfo/linux-ha
See also: http://linux-ha.org/ReportingProblems

florian at hastexo

Feb 7, 2012, 9:43 AM

Post #14 of 14 (143 views)
Permalink

Re: question [In reply to]

On Fri, Feb 3, 2012 at 12:10 AM, Amel <mail [at] amel> wrote:
> Hey hey,
>
> I am new to HA Heartbeat, and I am wondering about one thing as I could not find any info about it on Your page.
>
> Can we run the HA heartbeat across the WAN links if the servers are on the different subnets ??
> In case "yes" what about the bandwidth usage ? Â is it stable ?

Yes, when using "ucast" links. Or, if you're using Corosync instead of
Heartbeat, using the udpu transport. Bandwidth is really not your
issue though, latency and the inherent unreliability of your transport
network are. So what you are trying to do is clearly not recommended.

Perhaps you want to tell us what you're really trying to achieve, so
we can suggest more suitable alternatives?

Cheers,
Florian

--
Need help with High Availability?
http://www.hastexo.com/now
_______________________________________________
Linux-HA mailing list
Linux-HA [at] lists
http://lists.linux-ha.org/mailman/listinfo/linux-ha
See also: http://linux-ha.org/ReportingProblems

Index | Next | Previous | View Threaded

Interested in having your list archived? Contact Gossamer Threads