Skip to contentSkip to conversation

Texas UML instances and OLSR+BATMAN tests

B.A.T.M.A.N. Benjamin Henrion 6 participants 21 messages Jun 8, 2007 – Oct 9, 2007 1 message not in the archive

lore
Share & tools
lore

Benjamin Henrion

author
#0 lore raw

Hi,

I would like to test OLSR and BATMAN on UML instances, since I don't have the hardware to make real tests.

Would it be possible to have a copy of the binaries running on the simulator of http://texas.funkfeuer.at/, in order to launch it on my computer?

Best,

--
Benjamin Henrion <[email protected]>
FFII Brussels - +32-484-566109 - +32-2-4148403

#1 lore raw

If you want to code on it you can even test there. But only one tester at a time please

alle the best,
a.

Quoted 11 lines · Benjamin Henrion

Hi,

I would like to test OLSR and BATMAN on UML instances, since I don't have the hardware to make real tests.

Would it be possible to have a copy of the binaries running on the simulator of http://texas.funkfeuer.at/, in order to launch it on my computer?

Best,

Quoted 8 lines

--
Benjamin Henrion <[email protected]>
FFII Brussels - +32-484-566109 - +32-2-4148403
_______________________________________________
B.A.T.M.A.N mailing list
[email protected]
https://list.open-mesh.net/mm/listinfo/b.a.t.m.a.n

---
there's no place like 127.0.0.1

author
#2 lore raw

No I am looking for UML kernels + rootfs (which includes olsr or batman) I can boot on my 32bit computer. Texas seems to be a 64bit machine.

Do you sources and/or config files for the UML kernel? Where did you get the rootfs?

Best,

Quoted 38 lines · Aaron Kaplan

If you want to code on it you can even test there. But only one tester at a time please

alle the best,
a.

Quoted 11 lines · Benjamin Henrion

Hi,

I would like to test OLSR and BATMAN on UML instances, since I don't have the hardware to make real tests.

Would it be possible to have a copy of the binaries running on the simulator of http://texas.funkfeuer.at/, in order to launch it on my computer?

Best,

Quoted 8 lines

--
Benjamin Henrion <[email protected]>
FFII Brussels - +32-484-566109 - +32-2-4148403
_______________________________________________
B.A.T.M.A.N mailing list
[email protected]
https://list.open-mesh.net/mm/listinfo/b.a.t.m.a.n

---
there's no place like 127.0.0.1

_______________________________________________
B.A.T.M.A.N mailing list
[email protected]
https://list.open-mesh.net/mm/listinfo/b.a.t.m.a.n

Signature · 2 lines
Benjamin Henrion <bhenrion at ffii.org>
FFII Brussels - +32-484-566109 - +32-2-4148403
#3 lore raw

I am currently trying to get a grip on vnuml (
http://www.dit.upm.es/vnumlwiki/index.php/Tutorial ) which has good howtos and tutorial for building host- and uml- kernels and root_fs.

Especially vnuml offers a framework for virutal network configuration based on xml-configuration files. I think texas-kernel is also patched for vnuml enhancements but for texas's root_fs and network-configuration scripts I am not shure.
Maybe aaron can provide some more background on that?

In order to releas texas from unproductive trial-and-error configuration testing (and feeling less intervening when applying core-kernel-patches or installing new system-libraries) I started to setup a small vnuml environment on my home-pc. Therefore I found also the following links quite useful:

http://www.dit.upm.es/vnumlwiki/index.php/Installation

http://www.enic.fr/people/landru/viminal/vnuml.gentoo/how-to/vnuml-gentoo-guide.html
( gentoo oriented setup, host-kernel-enhancement, uml-kernel, root_fs )

http://www.dit.upm.es/vnumlwiki/index.php/Howto
( further links for root_fs howtos )

@aaron:
do you have a link explaining from-scratch busybox-based uml-rootfs generation (as being used on texas) ?

ciao,
axel

Quoted 44 lines · Benjamin Henrion

No I am looking for UML kernels + rootfs (which includes olsr or batman) I can boot on my 32bit computer. Texas seems to be a 64bit machine.

Do you sources and/or config files for the UML kernel? Where did you get the rootfs?

Best,

Quoted 34 lines · Aaron Kaplan

If you want to code on it you can even test there. But only one tester at a time please

alle the best,
a.

Quoted 11 lines · Benjamin Henrion

Hi,

I would like to test OLSR and BATMAN on UML instances, since I don't have the hardware to make real tests.

Would it be possible to have a copy of the binaries running on the simulator of http://texas.funkfeuer.at/, in order to launch it on my computer?

Best,

Quoted 7 lines

--
Benjamin Henrion <[email protected]>
FFII Brussels - +32-484-566109 - +32-2-4148403
_______________________________________________
B.A.T.M.A.N mailing list
[email protected]
https://list.open-mesh.net/mm/listinfo/b.a.t.m.a.n

---
there's no place like 127.0.0.1

_______________________________________________
B.A.T.M.A.N mailing list
[email protected]
https://list.open-mesh.net/mm/listinfo/b.a.t.m.a.n

#4 lore raw
Quoted 10 lines · Axel Neumann

I am currently trying to get a grip on vnuml (
http://www.dit.upm.es/vnumlwiki/index.php/Tutorial ) which has good howtos and tutorial for building host- and uml- kernels and root_fs.

Especially vnuml offers a framework for virutal network configuration based on xml-configuration files. I think texas-kernel is also patched for vnuml enhancements but for texas's root_fs and network-configuration scripts I am not shure.
Maybe aaron can provide some more background on that?

I _think_ zethix (in the CC) briefly looked into vnuml but it seemed to complicated. The scripts were all borked and quite unusable for us at least.

Quoted 10 lines

In order to releas texas from unproductive trial-and-error configuration testing (and feeling less intervening when applying core-kernel-patches or installing new system-libraries) I started to setup a small vnuml environment on my home-pc. Therefore I found also the following links quite useful:

http://www.dit.upm.es/vnumlwiki/index.php/Installation

http://www.enic.fr/people/landru/viminal/vnuml.gentoo/how-to/vnuml-gentoo-guide.html
( gentoo oriented setup, host-kernel-enhancement, uml-kernel, root_fs )

Quoted 7 lines

http://www.dit.upm.es/vnumlwiki/index.php/Howto
( further links for root_fs howtos )

@aaron:
do you have a link explaining from-scratch busybox-based uml-rootfs generation (as being used on texas) ?

yes

with an existing root_fs:

mount -o loop root_fs /mnt
then just copy over things as needed. That is what I did . works!

To create an initial root_fs:

dd if=/dev/zero of=roof_fs bs=1M count=XXX
Then mount this root_fs from inside an existing UML instance and fdisk it (./linux ubd1=root_fs ) and mkfs it

voila!

a.

PS:
zethix was recently working on a way to do the simulation completely without UML but with olsr_switch which has still some bugs. Maybe some similar batman_switch application can be usefull?

---
there's no place like 127.0.0.1

#5 lore raw
Quoted 18 lines · Aaron Kaplan
Quoted 5 lines · Axel Neumann

I am currently trying to get a grip on vnuml (
http://www.dit.upm.es/vnumlwiki/index.php/Tutorial ) which has good howtos
and tutorial for building host- and uml- kernels and root_fs.

Quoted 9 lines

Especially vnuml offers a framework for virutal network configuration based on
xml-configuration files. I think texas-kernel is also patched for vnuml
enhancements but for texas's root_fs and network-configuration scripts I am
not shure.
Maybe aaron can provide some more background on that?

Quoted 3 lines

I _think_ zethix (in the CC) briefly looked into vnuml but it seemed to complicated.
The scripts were all borked and quite unusable for us at least.

Yes, as far as I remember, vnuml needs a lot of things in order to work... well, things that all
linux distros usually have, but not busybox. It was something like ... install its boot scripts in the
vnuml'd system, make it mount an fs on the host, then use ssh for something and so on and so
on. But the biggest problem we had was actually... that it was constantly making our poor system
crash. Anyway, I guess we just had to drop it in order to move on. I do believe there are some nice
things about vnuml, but ...

Quoted 41 lines
Quoted 16 lines

In order to releas texas from unproductive trial-and-error configuration
testing (and feeling less intervening when applying core-kernel- patches or
installing new system-libraries) I started to setup a small vnuml environment
on my home-pc. Therefore I found also the following links quite useful:

http://www.dit.upm.es/vnumlwiki/index.php/Installation

http://www.enic.fr/people/landru/viminal/vnuml.gentoo/how-to/vnuml-
gentoo-guide.html
( gentoo oriented setup, host-kernel-enhancement, uml-kernel, root_fs )

Quoted 8 lines

http://www.dit.upm.es/vnumlwiki/index.php/Howto
( further links for root_fs howtos )

@aaron:
do you have a link explaining from-scratch busybox-based uml-rootfs generation
(as being used on texas) ?

yes

with an existing root_fs:

mount -o loop root_fs /mnt
then just copy over things as needed. That is what I did . works!

To create an initial root_fs:

dd if=/dev/zero of=roof_fs bs=1M count=XXX
Then mount this root_fs from inside an existing UML instance and fdisk it (./linux ubd1=root_fs )
and mkfs it

voila!

Quoted 7 lines

a.

PS:
zethix was recently working on a way to do the simulation completely without UML but with olsr_switch
which has still some bugs. Maybe some similar batman_switch application can be usefull?

Briefly, the olsr_switch works like this: in olsrd, there is this special 'driver' that doesn't use
a netif for transmission, but a tcp connection instead. that connection goes to the olsr_switch.
Also, there are some patches preventing all instances to modify the same routing table. So,
in order to make batman work with olsr_switch (or whatever) I guess modifications might be
needed.

There is also an N-th way of doing it, probably: pseudo interfaces (like tap) instead of
tcp connections. Nevertheless, all batmans (or olsrds) will share the same routing table.

---

Quoted 1 line

there's no place like 127.0.0.1

Yes, there is - ::1, only bigger ;)

#6 lore raw
Quoted 4 lines · zethix or something

---
there's no place like 127.0.0.1

Yes, there is - ::1, only bigger ;)

haha !!! cooool answer :)

author
#7 lore raw
In reply to zethix or something
Quoted 99 lines · zethix or something
Quoted 5 lines · Aaron Kaplan
Quoted 2 lines · Axel Neumann

I am currently trying to get a grip on vnuml (
http://www.dit.upm.es/vnumlwiki/index.php/Tutorial )

which has good

Quoted 12 lines
Quoted 12 lines

howtos
and tutorial for building host- and uml- kernels and root_fs.

Especially vnuml offers a framework for virutal network configuration based on
xml-configuration files. I think texas-kernel is also patched for vnuml
enhancements but for texas's root_fs and network-configuration scripts I am
not shure.
Maybe aaron can provide some more background on that?

Quoted 3 lines

I _think_ zethix (in the CC) briefly looked into vnuml but it seemed to complicated.
The scripts were all borked and quite unusable for us at least.

Yes, as far as I remember, vnuml needs a lot of things in order to work... well, things that all
linux distros usually have, but not busybox. It was something like ... install its boot scripts in the
vnuml'd system, make it mount an fs on the host, then use ssh for something and so on and so
on. But the biggest problem we had was actually... that it was constantly making our poor system
crash. Anyway, I guess we just had to drop it in order to move on. I do believe there are some nice
things about vnuml, but ...

Quoted 9 lines
Quoted 9 lines

In order to releas texas from unproductive trial-and-error configuration
testing (and feeling less intervening when applying core-kernel- patches or
installing new system-libraries) I started to setup a small vnuml environment
on my home-pc. Therefore I found also the following links quite useful:

Quoted 3 lines
Quoted 3 lines

http://www.dit.upm.es/vnumlwiki/index.php/Installation

http://www.enic.fr/people/landru/viminal/vnuml.gentoo/how-to/vnuml-

Quoted 29 lines
Quoted 4 lines

gentoo-guide.html
( gentoo oriented setup, host-kernel-enhancement, uml-kernel, root_fs )

Quoted 8 lines

http://www.dit.upm.es/vnumlwiki/index.php/Howto
( further links for root_fs howtos )

@aaron:
do you have a link explaining from-scratch busybox-based uml-rootfs generation
(as being used on texas) ?

yes

with an existing root_fs:

mount -o loop root_fs /mnt
then just copy over things as needed. That is what I did . works!

To create an initial root_fs:

dd if=/dev/zero of=roof_fs bs=1M count=XXX
Then mount this root_fs from inside an existing UML instance and fdisk it (./linux ubd1=root_fs )
and mkfs it

voila!

Quoted 7 lines

a.

PS:
zethix was recently working on a way to do the simulation completely without UML but with olsr_switch
which has still some bugs. Maybe some similar batman_switch application can be usefull?

Briefly, the olsr_switch works like this: in olsrd, there is this special 'driver' that doesn't use
a netif for transmission, but a tcp connection instead. that connection goes to the olsr_switch.
Also, there are some patches preventing all instances to modify the same routing table. So,
in order to make batman work with olsr_switch (or whatever) I guess modifications might be
needed.

Quoted 4 lines

There is also an N-th way of doing it, probably: pseudo interfaces (like tap) instead of
tcp connections. Nevertheless, all batmans (or olsrds) will share the same routing table.

What about using OpenVZ instances (each instance has its own TCP IP stack I think)?

I did not manage to have a working 32bit UML kernel+busybox rootfs for testing OLSR and/or batman (ifconfig did not work out with the busybox rootfs version of
http://uml.nagafix.co.uk/BusyBox-1.5.0/BusyBox-1.5.0-x86-root_fs.bz2
and kernel version vmlinux-2.6.18.1-bb2), if someone can recommend me one version for each of those.

--
Benjamin Henrion <bhenrion at ffii.org>
FFII Brussels - +32-484-566109 - +32-2-4148403

#8 lore raw
Quoted 1 line
Quoted 9 lines · Benjamin Henrion

What about using OpenVZ instances (each instance has its own TCP IP stack I think)?

I did not manage to have a working 32bit UML kernel+busybox rootfs for testing OLSR and/or batman (ifconfig did not work out with the busybox rootfs version of
http://uml.nagafix.co.uk/BusyBox-1.5.0/BusyBox-1.5.0-x86-root_fs.bz2
and kernel version vmlinux-2.6.18.1-bb2), if someone can recommend me one version for each of those.

Well, I have not tried openVZ. I would be interested in it.
UML is quite tricky to get right. As soon as our texas server is up and running again, I can give you kernel .config file which worked for me.

best,
aaron.

---
there's no place like 127.0.0.1

#9 lore raw

Hello,

We have used OpenVZ (linux-2.6.18-openvz-028.035.1) for emulating batman and olsr.
Further background information about the setup is given here:
http://open-mesh.net/batman/doc/evaluation/setup-details
and about the general evaluation methodology here:
http://open-mesh.net/batman/doc/evaluation

Not yet documented but:
Recently also managed to run openVZ (with many virtuaized debian systems) inside of qemu-kvm. Used this for proof of concept, testing and performance evaluation.

This has the advantage of:
- having all relevant data in a qemu image.
- The image could be easily transferred to another machine (actually Iam waiting to move it to texas).
- could be easily backupped
- Any misconfiguration inside qemu does never harm your real system !!

The following worked:
- On 2GHz core duo notebook (qemu using only one processor and 800MB Ram)
- More than 100 virtual instances idling in parrallel
- About 60 virtual instances running olsr or batman in parrallel

- use brctl, ebtables, TC, and NETEM for configuring virtual networks with
- dedicated links
- dedicated packetloss, delay for unicast and broadcast traffic

- Having all the tools like wireshark, tcpdump, ... for network monitoring,.

- Being able to monitor all virtualized VZ processes from the qemu system.

If somebody is interested in using or even helping to document and extend that system, would be really appreciated.

ciao,
axel

Quoted 29 lines · Aaron Kaplan
Quoted 9 lines · Benjamin Henrion

What about using OpenVZ instances (each instance has its own TCP IP stack I think)?

I did not manage to have a working 32bit UML kernel+busybox rootfs for testing OLSR and/or batman (ifconfig did not work out with the busybox rootfs version of
http://uml.nagafix.co.uk/BusyBox-1.5.0/BusyBox-1.5.0-x86-root_fs.bz2
and kernel version vmlinux-2.6.18.1-bb2), if someone can recommend me one version for each of those.

Well, I have not tried openVZ. I would be interested in it. UML is quite tricky to get right. As soon as our texas server is up and running again, I can
give you kernel .config file which worked for me.

best,
aaron.

---
there's no place like 127.0.0.1

_______________________________________________
B.A.T.M.A.N mailing list
[email protected]
https://list.open-mesh.net/mm/listinfo/b.a.t.m.a.n

author
#10 lore raw
Quoted 31 lines · Axel Neumann

Hello,

We have used OpenVZ (linux-2.6.18-openvz-028.035.1) for emulating batman and olsr.
Further background information about the setup is given here:
http://open-mesh.net/batman/doc/evaluation/setup-details
and about the general evaluation methodology here:
http://open-mesh.net/batman/doc/evaluation

Not yet documented but:
Recently also managed to run openVZ (with many virtuaized debian systems) inside of qemu-kvm. Used this for proof of concept, testing and performance evaluation.

This has the advantage of:
- having all relevant data in a qemu image.
- The image could be easily transferred to another machine (actually Iam waiting to move it to texas).
- could be easily backupped
- Any misconfiguration inside qemu does never harm your real system !!

The following worked:
- On 2GHz core duo notebook (qemu using only one processor and 800MB Ram)
- More than 100 virtual instances idling in parrallel
- About 60 virtual instances running olsr or batman in parrallel

- use brctl, ebtables, TC, and NETEM for configuring virtual networks with
- dedicated links
- dedicated packetloss, delay for unicast and broadcast traffic

Quoted 6 lines

- Having all the tools like wireshark, tcpdump, ... for network monitoring,.

- Being able to monitor all virtualized VZ processes from the qemu system.

If somebody is interested in using or even helping to document and extend that system, would be really appreciated.

I have tested the OpenVZ livecd with a small window manager (consumes 128MB in RAM), and with 10 instances running batman or olsr, it works fine with 512MB of RAM, but my problem is that the livecd does not supports ebtables it seems, which I need to configure for avoiding bridge shortcuts between the nodes.

Does anybody has experience with ebtables? Does it needs a kernel patch? Or it is just user-space?

OpenVZ looks very promising for network simulators, it consumes very few memory, especially if you use a mounted HD to store the filesystem.

--
Benjamin Henrion <bhenrion at ffii.org>
FFII Brussels - +32-484-566109 - +32-2-4148403

#11 lore raw

Hello

Quoted 7 lines · Benjamin Henrion
Quoted 6 lines · Axel Neumann

Hello,

We have used OpenVZ (linux-2.6.18-openvz-028.035.1) for emulating batman and olsr.
Further background information about the setup is given here:
http://open-mesh.net/batman/doc/evaluation/setup-details

[snap]

Quoted 9 lines
Quoted 9 lines

The following worked:
- On 2GHz core duo notebook (qemu using only one processor and 800MB Ram) - More than 100 virtual instances idling in parrallel
- About 60 virtual instances running olsr or batman in parrallel

- use brctl, ebtables, TC, and NETEM for configuring virtual networks with - dedicated links
- dedicated packetloss, delay for unicast and broadcast traffic

[snap]

Quoted 9 lines

I have tested the OpenVZ livecd with a small window manager (consumes 128MB in RAM), and with 10 instances running batman or olsr, it works fine with 512MB of RAM, but my problem is that the livecd does not supports ebtables it seems, which I need to configure for avoiding bridge shortcuts between the nodes.

Does anybody has experience with ebtables? Does it needs a kernel patch? Or it is just user-space?

If you follow the link in the above lines, you will find a line like:

... patched again to fix a bug in the ebtables part of the 2.6.18 kernel extracted from patch-2.6.18.8 from
http://www.kernel.org/pub/linux/kernel/v2.6/ and downloadable here...

Why not having a look ?

Quoted 12 lines

OpenVZ looks very promising for network simulators, it consumes very few memory, especially if you use a mounted HD to store the filesystem.

--
Benjamin Henrion <bhenrion at ffii.org>
FFII Brussels - +32-484-566109 - +32-2-4148403
_______________________________________________
B.A.T.M.A.N mailing list
[email protected]
https://list.open-mesh.net/mm/listinfo/b.a.t.m.a.n

#12 lore raw

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Quoted 6 lines · Axel Neumann
Quoted 6 lines

I have tested the OpenVZ livecd with a small window manager (consumes 128MB in RAM), and with 10 instances running batman or olsr, it works fine with 512MB of RAM, but my problem is that the livecd does not supports ebtables it seems, which I need to configure for avoiding bridge shortcuts between the nodes.

Quoted 8 lines
Quoted 2 lines

Does anybody has experience with ebtables? Does it needs a kernel patch? Or it is just user-space?

If you follow the link in the above lines, you will find a line like:

... patched again to fix a bug in the ebtables part of the 2.6.18 kernel extracted from patch-2.6.18.8 from
http://www.kernel.org/pub/linux/kernel/v2.6/ and downloadable here...

in openvz usuall all ebtables and iptables modules are supported in the 0 instance. modules that can be used inside of virtuall instances have to be prepared for this. but this shouldn't be a problem for your setup, so i guess the problem may be somewhere else.

by the way, if you use reiserfs + raid + lvm you have to switch off the 4k stack patch or your system will run instable :)

kindly regards
daniel
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.6 (GNU/Linux)
Comment: GnuPT 2.7.2
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org

iD8DBQFHC0Wuy/mkIQp7AD0RAqK3AKC5sVgI7aNVn3+DST8EgbvqTrGOoQCeNTcj SMjDi5igzJtn45LbA7F+w70=
=9F1u
-----END PGP SIGNATURE-----

#13 lore raw

Hello,

Quoted 4 lines

in openvz usuall all ebtables and iptables modules are supported in the 0 instance. modules that can be used inside of virtuall instances have to be prepared for this. but this shouldn't be a problem for your setup, so i guess the problem may be somewhere else.

As fare as I remember there have been some promlems with the development versions of openVZ. With the stable version for kernel 2.6.18 everything worked fine.
Except, the ebtables module oopsed when applying more than 25 rules !! Because I wanted to configure virtual networks with up to 100 nodes this was a problem. The mentioned patch solved this problem.

ciao
/axel

#14 lore raw

Just yesterday I had a brief talk with Herbert from vserver. He mentioned that it should be ok to use different routing tables in each vserver instance. If this works out then I would like to know if vserver is a viable alternative. Vserver is very low overhead. One of the problems that I had with UML is the high (soft-)IRQ load. this produces permanent context switches and no work gets done. (if you start 1000 instances that is!) :)

Can somebody test what happens if you start many instances in openVZ and vserver? I hope our dev server is up and living soon again.

best,
a.

Quoted 14 lines · Axel Neumann

Hello,

Quoted 4 lines

in openvz usuall all ebtables and iptables modules are supported in the 0 instance. modules that can be used inside of virtuall instances have to be prepared for this. but this shouldn't be a problem for your setup, so i guess the problem may be somewhere else.

As fare as I remember there have been some promlems with the development versions of openVZ. With the stable version for kernel 2.6.18 everything worked fine.
Except, the ebtables module oopsed when applying more than 25 rules !! Because I wanted to configure virtual networks with up to 100 nodes this was a problem. The mentioned patch solved this problem.

Quoted 7 lines

ciao
/axel
_______________________________________________
B.A.T.M.A.N mailing list
[email protected]
https://list.open-mesh.net/mm/listinfo/b.a.t.m.a.n

---
there's no place like 127.0.0.1

author
#15 lore raw

Aaron Kaplan <[email protected]> [071009]:

Quoted 11 lines

Just yesterday I had a brief talk with Herbert from vserver. He mentioned that it should be ok to use different routing tables in each vserver instance. If this works out then I would like to know if vserver is a viable alternative. Vserver is very low overhead. One of the problems that I had with UML is the high (soft-)IRQ load. this produces permanent context switches and no work gets done. (if you start 1000 instances that is!) :)

Can somebody test what happens if you start many instances in openVZ and vserver?

Many = which number?

I have started 10 of those, and it works fine on a 1500Mhz machine with 512MB of RAM on a livecd system with minimal debian filesystems on a harddisk.

--
Benjamin Henrion <[email protected]>
FFII Brussels - +32-484-566109 - +32-2-4148403

#16 lore raw

with UML i had 1500 instances in parallel

lg,
a.

Quoted 19 lines · Benjamin Henrion

Aaron Kaplan <[email protected]> [071009]:

Quoted 11 lines

Just yesterday I had a brief talk with Herbert from vserver. He mentioned that it should be ok to use different routing tables in each vserver instance. If this works out then I would like to know if vserver is a viable alternative. Vserver is very low overhead. One of the problems that I had with UML is the high (soft-)IRQ load. this produces permanent context switches and no work gets done. (if you start 1000 instances that is!) :)

Can somebody test what happens if you start many instances in openVZ and vserver?

Many = which number?

I have started 10 of those, and it works fine on a 1500Mhz machine with 512MB of RAM on a livecd system with minimal debian filesystems on a harddisk.

Quoted 8 lines

--
Benjamin Henrion <[email protected]>
FFII Brussels - +32-484-566109 - +32-2-4148403
_______________________________________________
B.A.T.M.A.N mailing list
[email protected]
https://list.open-mesh.net/mm/listinfo/b.a.t.m.a.n

---
there's no place like 127.0.0.1

#17 lore raw
Quoted 1 line · Aaron Kaplan

with UML i had 1500 instances in parallel

There should be much more possible since olsrd uses way less (> 30% or so)
CPU time (and probably RAM - no one ever measured that AFAIK really. But Hannes killed code and lots of malloc()s/free()s) than 3 months ago.

As for Vserver: Yes, that might be another trial worth (especially if the vservers 99% identical) but AFAIK it needs a patched kernel. And "Vserver on UML": Herbert should know better if one can mange to run that. At least playing around is less risky.

Bernd

Signature · 3 lines
Firmix Software GmbH                   http://www.firmix.at/
mobil: +43 664 4416156                 fax: +43 1 7890849-55
          Embedded Linux Development and Services
#18 lore raw

yep, I think so too! Definitely worth a try.
@Bertl: would you help us to get vserver running as the basis for olsr (mesh routing demon) tests? @Bernd: texas is at your place right now? I think I will just order a new mainboard.

best,
a.

Quoted 5 lines

As for Vserver: Yes, that might be another trial worth (especially if the vservers 99% identical) but AFAIK it needs a patched kernel. And "Vserver on UML": Herbert should know better if one can mange to run that. At least playing around is less risky.

Quoted 12 lines

Bernd

Signature · 9 lines
Firmix Software GmbH                   http://www.firmix.at/
mobil: +43 664 4416156                 fax: +43 1 7890849-55
          Embedded Linux Development and Services


_______________________________________________
B.A.T.M.A.N mailing list
[email protected]
https://list.open-mesh.net/mm/listinfo/b.a.t.m.a.n

---
there's no place like 127.0.0.1

A message that is not in the archive (sent off-list or not yet received): [email protected]

#20 lore raw
Quoted 9 lines · Herbert Poetzl
Quoted 5 lines · Aaron Kaplan

yep, I think so too! Definitely worth a try.

@Bertl: would you help us to get vserver running
as the basis for olsr (mesh routing demon) tests?

sure, but please get me some idea what you want to do first :)

ok, so we have the following issue (sorry): we want to test B.a.t.m.a.n (https://www.open-mesh.net/) and olsrd (http://olsr.funkfeuer.at, http://wiki.funkfeuer.at//index.php/Olsrd-ng) on our test server. That test server has 8GB or RAM and is currently in repair.
The idea is that in order to create a massive test with many instances of olsrd or b.a.t.m.a.n. we need lots of virtual hosts, they should all be connected via a virtual network. This network needs a possibility of (real) packetloss between each link.

Furthermore each instance needs its own routing table.

when?
anytime, we can start on some devel server with a few instances. As soon as our 8GB RAM machine is repaired we should test it there.

best,
aaron.

#21 lore raw
In reply to Benjamin Henrion

Hello!

Quoted 4 lines · Benjamin Henrion

Hi,

I would like to test OLSR and BATMAN on UML instances, since I don't have the hardware to make real tests.

I assume you are looking for 64bit UML-binaries. Then, if you have access to texas you can use the binary at /home/neumann/batmand .
use

$ file /home/neumann/batmand
to verify they were truely compiled for 64 bit.

but even better would be to download the latest stable and compile it yourself on a 64bit-system like texas with:
wget
http://downloads.open-mesh.net/batman/stable/sources/batmand_0.2.0rc2-current_sources.tgz
tar xvzf batmand_0.2.0rc2-current_sources.tgz
cd batmand_..
make
strip batmand # (optionally) to reduce the size of the binary

Anyway you'll find a binary attached to this mail, based on batmand_0.2.0rc2-rv406_sources.tgz statically linked and stripped and tested on the 64-bit-uml-instances on texas

ciao,
axel

Quoted 14 lines

Would it be possible to have a copy of the binaries running on the simulator of http://texas.funkfeuer.at/, in order to launch it on my computer?

Best,

--
Benjamin Henrion <[email protected]>
FFII Brussels - +32-484-566109 - +32-2-4148403
_______________________________________________
B.A.T.M.A.N mailing list
[email protected]
https://list.open-mesh.net/mm/listinfo/b.a.t.m.a.n