Re: Where to find modules for pre-compiled debian kernel

[Tzafrir Cohen]

On Mon, Sep 22, 2008 at 10:33:52PM +0300, Geoffrey S. Mendelson wrote:
> I have access to a "virtual machine" running Debian, kernel 
> "2.6.18-fza-028stab053.5-686-bigmem".
>
> I need to add a module to the kernel, but it currently does not exist.
> I have permission of the owner of the real machine to do anything I
> want to the virtual one, and have root access via sudo.

To add? Or to build?

Can you modprobe a module from a openvz guest?

>
> What I lack is Debian knowledge to this point. I'm too spoilt by Red
> Hat and it's derivatives, YUM. 

Loading a module is the same everywhere.

>
> I assume this was a pre-compiled kernel, can I find the modules for it
> without having to recompile them?

I hope so. Is that a .ko file?

-- 
Tzafrir Cohen | [EMAIL PROTECTED] | VIM is
http://tzafrir.org.il || a Mutt's
[EMAIL PROTECTED] ||  best
ICQ# 16849754 || friend

=
To unsubscribe, send mail to [EMAIL PROTECTED] with
the word "unsubscribe" in the message body, e.g., run the command
echo unsubscribe | mail [EMAIL PROTECTED]

Where to find modules for pre-compiled debian kernel

[Geoffrey S. Mendelson]

I have access to a "virtual machine" running Debian, 
kernel "2.6.18-fza-028stab053.5-686-bigmem".


I need to add a module to the kernel, but it currently does not exist.
I have permission of the owner of the real machine to do anything I
want to the virtual one, and have root access via sudo.

What I lack is Debian knowledge to this point. I'm too spoilt by Red
Hat and it's derivatives, YUM. 


I assume this was a pre-compiled kernel, can I find the modules for it
without having to recompile them?

Thanks, Geoff.
--
Geoffrey S. Mendelson, Jerusalem, Israel [EMAIL PROTECTED]  N3OWJ/4X1GM

=
To unsubscribe, send mail to [EMAIL PROTECTED] with
the word "unsubscribe" in the message body, e.g., run the command
echo unsubscribe | mail [EMAIL PROTECTED]

Re: Telling libtool/automake to be less anal

[Ehud Karni]

On Mon, 22 Sep 2008 20:28:18 Shachar Shemesh wrote:
>
> Ehud Karni wrote:
> >
> > will adding a simple rule to make a real link, something like
> > ln -f libplugin.so.0.1.2 plugin.so
> > work for you ?
> >
> Right now I'm not working on a Mozilla plugin (I'm writing a plugin for
> net-snmp agent, which is less anal about such things), so a link is
> neither necessary nor particularly important. When writing Mozilla
> plugins, the problem is not creating the links (running "make install"
> create those links with no extra work). The problem is that mozilla
> does, roughly, the following:
> - Scan all files in the plugin directory
> - For each file: As long as the type is "symlink", resolve where it
> points to.
> - Check whether the file name ends with ".so"
>
> So the above plan (link from .so to the real name) will simply not work.
> Please don't ask me why that is what Mozilla does.

Shachar, I suggested a real (hard) link, not a symbolic link.
The -f was to ensure re-linking even if the target exist.

Ehud.


--
 Ehud Karni   Tel: +972-3-7966-561  /"\
 Mivtach - Simon  Fax: +972-3-7966-667  \ /  ASCII Ribbon Campaign
 Insurance agencies   (USA) voice mail and   X   Against   HTML   Mail
 http://www.mvs.co.il  FAX:  1-815-5509341  / \
 GnuPG: 98EA398D Better Safe Than Sorry

=
To unsubscribe, send mail to [EMAIL PROTECTED] with
the word "unsubscribe" in the message body, e.g., run the command
echo unsubscribe | mail [EMAIL PROTECTED]

Re: Telling libtool/automake to be less anal

[Shachar Shemesh]

Ehud Karni wrote:


will adding a simple rule to make a real link, something like
ln -f libplugin.so.0.1.2 plugin.so
work for you ?
  
Right now I'm not working on a Mozilla plugin (I'm writing a plugin for 
net-snmp agent, which is less anal about such things), so a link is 
neither necessary nor particularly important. When writing Mozilla 
plugins, the problem is not creating the links (running "make install" 
create those links with no extra work). The problem is that mozilla 
does, roughly, the following:

- Scan all files in the plugin directory
- For each file: As long as the type is "symlink", resolve where it 
points to.

- Check whether the file name ends with ".so"

So the above plan (link from .so to the real name) will simply not work. 
Please don't ask me why that is what Mozilla does.


Shachar

=
To unsubscribe, send mail to [EMAIL PROTECTED] with
the word "unsubscribe" in the message body, e.g., run the command
echo unsubscribe | mail [EMAIL PROTECTED]

Re: Telling libtool/automake to be less anal

[Ehud Karni]

On Sun, 21 Sep 2008 17:24:49 Shachar Shemesh wrote:
>
> Case in point: You want to develop a mozilla plugin. It is harmless but
> makes no sense to have the plug-in's name begin with "lib". Also, if the
> plugin's name does not end in ".so", Mozilla will refuse to load it.
> Worse, Mozilla explicitly resolves symbolic links, so merely putting the
> file name as libtool created it (libplugin.so.0.1.2) and placing a
> symbolic link to it will not work. The file has to actually be called
> "libplugin.so".

will adding a simple rule to make a real link, something like
ln -f libplugin.so.0.1.2 plugin.so
work for you ?

Ehud.


--
 Ehud Karni   Tel: +972-3-7966-561  /"\
 Mivtach - Simon  Fax: +972-3-7966-667  \ /  ASCII Ribbon Campaign
 Insurance agencies   (USA) voice mail and   X   Against   HTML   Mail
 http://www.mvs.co.il  FAX:  1-815-5509341  / \
 GnuPG: 98EA398D Better Safe Than Sorry

=
To unsubscribe, send mail to [EMAIL PROTECTED] with
the word "unsubscribe" in the message body, e.g., run the command
echo unsubscribe | mail [EMAIL PROTECTED]

Re: Firefox conformity: fighting from within

[Arie Skliarouk]

Hi,

Just a small good news: Today BGU upgraded version of HighLearn, and
the new version does work with Firefox.

-- 
Arie

=
To unsubscribe, send mail to [EMAIL PROTECTED] with
the word "unsubscribe" in the message body, e.g., run the command
echo unsubscribe | mail [EMAIL PROTECTED]

Re: MySQL over DRBD (was: Re: Clustering/Failover Project)

[Oleg Goldshmidt]

"Amos Shapira" <[EMAIL PROTECTED]> writes:

> BUT now that we are finally moving from DNS-based fail-over (due to
> lack of Virtual IP support in the hosting product we originally had)
> to Virtual-IP (using Linux Virtual Server) we see how MySQL
> master-master replication (using simple two-sided replication, e.g.
> http://www.howtoforge.com/mysql-5-master-master-replication-fedora-8,
> NOT the "MySQL Cluster" product which last time I read about it it
> sounded like a bad joke) would benefit us -
> 1. It means that fail-over time will be virtually zero (compared to a
> few seconds to bring up the secondary)
> 2. It means that we can take advantage of both MySQL servers in
> parallel - for higher capacity

It looks like you are also interested in load balancing / scalability
and are looking for active/active failover as a part of that. AFAIK,
LinuxHA has active/active mode, but I have never used it. From Noam's
description it seems that active/passive is what he is looking for.

>> [2] RedHat is the only exception I know of, I suppose because they
>>have their own clustering product - of course RPMs are there.
>
> Exception to what? Linux-HA and DRBD comes packaged in CentOS 5 and we
> have all software from rpm's, we didn't have to compile anything.

RPMs are there, but AFAIK they are not included by default in RHEL
(the "RedHat" we, as a commercial company, normally deal with, as
opposed to CentOS). Other distros (including, e.g., SLES) have LinuxHA
in by default. We have it here on RHEL, actually it was the first
platform we tried it on and it was up and running in no time,
providing failover to a rather sophisticated bunch of servers with
MySQL inside.

-- 
Oleg Goldshmidt | [EMAIL PROTECTED]

=
To unsubscribe, send mail to [EMAIL PROTECTED] with
the word "unsubscribe" in the message body, e.g., run the command
echo unsubscribe | mail [EMAIL PROTECTED]

Re: Looking for IGMP (multicast) server-client solution

[Baruch Even]

On Mon, Sep 22, 2008 at 1:39 PM, Lev Olshvang <[EMAIL PROTECTED]> wrote:

>
> Hi Baruch,
>
> Thanks for response,
> I will put my comments close to your statements, please see below
>
>
>
> Thaks,
>
>
> Baruch Even wrote:
>
> On Mon, Sep 22, 2008 at 11:38 AM, Lev Olshvang <[EMAIL PROTECTED]> wrote:
>
>
> Hi,
>
> I have following issue to solve:
>
> The network have one server which sends multicast streams to several
> multicast groups.
>
> I would like it to stop  streaming if no client is listening, and  start
> streaming again when client is joining multicast group.
>
>
> Normally for multicast the server never receives any notification if there
> are clients or not. It just sends data and if the clients want it, they ask
> for it. To control the server actual sending you need an external protocol
> to notify the server that there is a new client and when the client leaves.
>
>As such, I think  IGMP serves this matter.  IGMP defines host - router
> conversation and I see IGMP messages from a host
>   (in my terms client).  ( client application use setsockopt(
> IP_ADD_MEMBERSHIP) ) .
>
> The server box have as linux kernel + mrouted counting for this groups and
> making proper entries in /proc/net/igmp,
> /proc/net/ip_mf_cache ,...
>

Yes.

>
> If the clients are connected to the server without a router between them,
> you could implement a program that listens on the interface for the IGMP
> messages and controls the server when to send and when not.
>
>   Since the kernel have this information - which group are active on every
> interface, it is very logically to suppose that kernel may easily filter out
> the multicast packets to the non-active group. It seems to me it should work
> this way.
> But these functionality should be implemented in kernel space ( it is
> similar to netfilter,) the question is how to turn it on ?
>

It has it but multicast routing doesn't work this way. There is no way to
"turn it on" simply because there is no such functionality. The multicast
sender will continue to send its data on the output interface regardless of
any routing rules. The routing rules can only add additional interfaces to
send to.

>
>
> If there is a router between the server and the clients you are out of luck
> as the router expects the multicast packets to just be sent constantly to
> its interface for it to consider this multicast group active at all. A
> router will not send IGMP messages to you!
>
> I think you wrong here, I see mrouted is taking the responsibility of
> querier and periodically ask the interface neighbours to confirm every
> registered multicast group relationship.
>

That's the router-client communication. I meant that the router will not
send the multicast sender an IGMP join request.

>
>
>
> I see from sniffing, that clients  indeed  send IGMP join group protocol
> message  , and I wonder how to make linux server to handle this join and
> manage multicast group.
>
>
> If you send data to the interface the clients will receive it, I'm not sure
> what is your problem statement here.
>
>I do not want to overload LAN in  a data nobody is interesting  (no
> clients running)
>

Then the case is closed. Implement your own protocol on top to notify the
server when to stop transmission.

There is one way you can try but it's likely not to work. If you send your
data to a tun device that has no output or a loopback device it might be
that mrouted will properly forward from it to the external interfaces when
clients show up there but the multicast sender will continue to send and
burn cpu time, you just won't see the output on the network.




I saw that kernel have provisions to manage multicast groups through
setsockopt(), but I did not find how to get join/leave group notification at
user
space.

I know that there are multicast routers , like mrouted and pimd and I am
trying to configure thema at they moment but I  doubt that I can configure
only one interface because it contradict routing concept, isn't it ?


mrouted and pimd will monitor for IGMP messages and control the kernel
multicast routing fabric. They will instruct the kernel where to forward
multicast packets, but you only need that if you have multiple interfaces on
your computer or to make a multicast router that sits between your multicast
server and clients.

   I expected them to turn outgoing packet filtering on the base of active
group information kernel has.
That never happens.

Baruch

MySQL over DRBD (was: Re: Clustering/Failover Project)

[Amos Shapira]

2008/9/22 Oleg Goldshmidt <[EMAIL PROTECTED]>:
> Noam Rathaus <[EMAIL PROTECTED]> writes:
>
>> I'm also not interested in replies telling me there's a project that
>> does exactly that (and only requires a few tiny customizations that
>> will take days).
>
> Well, how about a couple of hours, including RTFM? LinuxHA (heartbeat,
> http://www.linux-ha.org) + DRBD (http://www.drbd.org,
> http://linux-ha.org/DRBD) is the most common linux HA solution, works
> great with MySQL (http://www.mysql.com/products/enterprise/drbd.html),
> comes with most linux distros [1,2], and thus is likely already
> installed on your system.

(I changed the subject since this is about replying to Oleg, not
within the scope of the original post).

We are using MySQL over DRBD mostly because I had to set it up in a
hurry and we were already using DRBD for PostgresQL and other files.

BUT now that we are finally moving from DNS-based fail-over (due to
lack of Virtual IP support in the hosting product we originally had)
to Virtual-IP (using Linux Virtual Server) we see how MySQL
master-master replication (using simple two-sided replication, e.g.
http://www.howtoforge.com/mysql-5-master-master-replication-fedora-8,
NOT the "MySQL Cluster" product which last time I read about it it
sounded like a bad joke) would benefit us -
1. It means that fail-over time will be virtually zero (compared to a
few seconds to bring up the secondary)
2. It means that we can take advantage of both MySQL servers in
parallel - for higher capacity

I have yet to get around to switch to MySQL clustering but I think if
you have the time and the option then it's a better long-term
investment than plain DRBD.

>
> Anyway, I am not available for this gig, so feel free to pay whoever
> does this for you. If I were you, I would insist on hearing real hard
> arguments before deciding on a competing solution.
>
> One such argument may be "we absolutely cannot move MySQL data onto a
> separate partition, because (a really good argument goes here)".
>
> [1] Debian included
> [2] RedHat is the only exception I know of, I suppose because they
>have their own clustering product - of course RPMs are there.

Exception to what? Linux-HA and DRBD comes packaged in CentOS 5 and we
have all software from rpm's, we didn't have to compile anything.

Cheers,

--Amos

PS - The link to HowtoForge above, which I just noticed while writing
this message, looks like excellent news since until now I saw only
tutorials on how to do this on MySQL 4, which didn't provide support.

PPS - Personally, I prefer PostgresQL, and maybe use PL/Proxy
(https://developer.skype.com/SkypeGarage/DbProjects/PlProxy) or
http://www.postgresql.at/english/downloads_e.html for clustering, e.g.
see http://highscalability.com/skype-plans-postgresql-scale-1-billion-users.

=
To unsubscribe, send mail to [EMAIL PROTECTED] with
the word "unsubscribe" in the message body, e.g., run the command
echo unsubscribe | mail [EMAIL PROTECTED]

Re: Looking for IGMP (multicast) server-client solution

[Lev Olshvang]

Hi Baruch,

Thanks for response,
I will put my comments close to your statements, please see below



Thaks,


Baruch Even wrote:

  
  On Mon, Sep 22, 2008 at 11:38 AM, Lev Olshvang <[EMAIL PROTECTED]>
wrote:
  
  
Hi,

I have following issue to solve:

The network have one server which sends multicast streams to several
multicast groups.

I would like it to stop  streaming if no client is listening, and
 start streaming again when client is joining multicast group.
  
Normally for multicast the server never receives any notification if
there are clients or not. It just sends data and if the clients want
it, they ask for it. To control the server actual sending you need an
external protocol to notify the server that there is a new client and
when the client leaves.
  
  
  

   As such, I think  IGMP serves this matter.  IGMP defines host -
router conversation and I see IGMP messages from a host 
  (in my terms client).  ( client application use setsockopt(
IP_ADD_MEMBERSHIP) ) .

The server box have as linux kernel + mrouted counting for this groups
and making proper entries in /proc/net/igmp, 
/proc/net/ip_mf_cache ,...

  
  
  
If the clients are connected to the server without a router between
them, you could implement a program that listens on the interface for
the IGMP messages and controls the server when to send and when not.
  
  
  

  Since the kernel have this information - which group are active on
every interface, it is very logically to suppose that kernel may easily
filter out the multicast packets to the non-active group. It seems to
me it should work this way.
But these functionality should be implemented in kernel space ( it is
similar to netfilter,) the question is how to turn it on ?

  
  
  
If there is a router between the server and the clients you are out of
luck as the router expects the multicast packets to just be sent
constantly to its interface for it to consider this multicast group
active at all. A router will not send IGMP messages to you!
  
  
  

I think you wrong here, I see mrouted is taking the responsibility of
querier and periodically ask the interface neighbours to confirm every
registered multicast group relationship.


  
  
   
  
  I see from sniffing, that clients  indeed  send IGMP join
group protocol message  , and I wonder how to make linux server to
handle this join and manage multicast group.
  
If you send data to the interface the clients will receive it, I'm not
sure what is your problem statement here. 
  
  
  

   I do not want to overload LAN in  a data nobody is interesting  (no
clients running) 

  
  
  
  
I saw that kernel have provisions to manage multicast groups through
setsockopt(), but I did not find how to get join/leave group
notification at user
space.

I know that there are multicast routers , like mrouted and pimd and I
am trying to configure thema at they moment but I  doubt that I can
configure only one interface because it contradict routing concept,
isn't it ?
  
mrouted and pimd will monitor for IGMP messages and control the kernel
multicast routing fabric. They will instruct the kernel where to
forward multicast packets, but you only need that if you have multiple
interfaces on your computer or to make a multicast router that sits
between your multicast server and clients.
  
  
  

   I expected them to turn outgoing packet filtering on the base of
active group information kernel has.


  
  
  
If the server app is on a machine and the clients are connected to it
directly (as in with a switch or a hub) you don't need mrouted. 
  
   
  Thanks ,
Lev
  
Baruch
  
  
  





=
To unsubscribe, send mail to [EMAIL PROTECTED] with
the word "unsubscribe" in the message body, e.g., run the command
echo unsubscribe | mail [EMAIL PROTECTED]

Re: Looking for IGMP (multicast) server-client solution

[Baruch Even]

On Mon, Sep 22, 2008 at 11:38 AM, Lev Olshvang <[EMAIL PROTECTED]> wrote:

>
> Hi,
>
> I have following issue to solve:
>
> The network have one server which sends multicast streams to several
> multicast groups.
>
> I would like it to stop  streaming if no client is listening, and  start
> streaming again when client is joining multicast group.


Normally for multicast the server never receives any notification if there
are clients or not. It just sends data and if the clients want it, they ask
for it. To control the server actual sending you need an external protocol
to notify the server that there is a new client and when the client leaves.

If the clients are connected to the server without a router between them,
you could implement a program that listens on the interface for the IGMP
messages and controls the server when to send and when not.

If there is a router between the server and the clients you are out of luck
as the router expects the multicast packets to just be sent constantly to
its interface for it to consider this multicast group active at all. A
router will not send IGMP messages to you!


> I see from sniffing, that clients  indeed  send IGMP join group protocol
> message  , and I wonder how to make linux server to handle this join and
> manage multicast group.


If you send data to the interface the clients will receive it, I'm not sure
what is your problem statement here.

>
> I saw that kernel have provisions to manage multicast groups through
> setsockopt(), but I did not find how to get join/leave group notification at
> user
> space.
>
> I know that there are multicast routers , like mrouted and pimd and I am
> trying to configure thema at they moment but I  doubt that I can configure
> only one interface because it contradict routing concept, isn't it ?


mrouted and pimd will monitor for IGMP messages and control the kernel
multicast routing fabric. They will instruct the kernel where to forward
multicast packets, but you only need that if you have multiple interfaces on
your computer or to make a multicast router that sits between your multicast
server and clients.

If the server app is on a machine and the clients are connected to it
directly (as in with a switch or a hub) you don't need mrouted.


> Thanks ,
> Lev


Baruch

Looking for IGMP (multicast) server-client solution

[Lev Olshvang]

Hi,

I have following issue to solve:

The network have one server which sends multicast streams to several 
multicast groups.


I would like it to stop  streaming if no client is listening, and  start 
streaming again when client is joining multicast group.


I see from sniffing, that clients  indeed  send IGMP join group protocol 
message  , and I wonder how to make linux server to handle this join and 
manage multicast group.


I saw that kernel have provisions to manage multicast groups through 
setsockopt(), but I did not find how to get join/leave group 
notification at user

space.

I know that there are multicast routers , like mrouted and pimd and I am 
trying to configure thema at they moment but I  doubt that I can 
configure only one interface because it contradict routing concept, 
isn't it ?


Thanks ,
Lev

=
To unsubscribe, send mail to [EMAIL PROTECTED] with
the word "unsubscribe" in the message body, e.g., run the command
echo unsubscribe | mail [EMAIL PROTECTED]