headers
Gobbledegeek | 1 Jul 2006 07:01
Picon

[quagga-users 7155] Re: Quagga-users Digest, Vol 35, Issue 35

>
> Gobble - when you say the system will struggle, what bottlenecks are present
> in the amd64 ht setup? Will running a 64bit linux VS 32bit linux help the
> situation?
>

Hi
 I meant to say you might hit a bottleneck with intel's outdated FSB
not on amd with hypertransport.
The rest of your question is best answered by someone who does system
performance banchmarking on the job for a living. I think if you
google on "wirespeed gigabit linux" you will find a lot of people
who've studied performance and capabilities of today's platforms.
Achieving gigabit thruput was a task even for the folks at Sun on
Solaris and RISC, if I remember right an article I read somewhere...
(:

PS: Why struggle with 10 100mbps nic's when you can have two gigabit
nic's in each system? The more components you have, the more the
probability of failure... the more maintanance and mgmt overhead. I
would inventory my 100mbps nics for reuse in desktops and replace
with gigabit nics on  the server.

Rgrds
GobbledeGeek
[Everything but Gobbledegook.. !!]
Permalink | Reply | 19 comments |
headers
Gobbledegeek | 1 Jul 2006 07:15
Picon

[quagga-users 7156] Re: Quagga-users Digest, Vol 35, Issue 35

>
> I'm pretty disappointed that per-packet routing/ load balancing is not
> available for linux "yet". What is so hard about implementing CEF in linux?
>
> Regards,
> Mike
> CCIE #16395

Design philosophy I guess - BSD's and linux are meant to be general
purpose OS'es not built to perform dedicated tasks like packet
forwarding etc...  I think once you can get a system up and running on
a lan the target is acheived for a general OS. I guess thats why we
have all the embedded s/w companies that strip down BSD's for
dedicated platforms and functions... on the bright side - it keeps the
job market hot!! (:

CEF is proprietory although yes one could design a hash based LB
algorithm with a running count of packets transferred per interface,
for linux and  bsd's. I guess its one of the pitfalls of open source
development model. No out-of-the-box multimedia workstations, nor any
for packet switching... I'm not sure if its fair to expect everything
with people contributing their free time for OSD. (: (Besides we need
the jobs...)

Rgrds
GobbledeGeek
[Everything but Gobbledegook.. !!]
Permalink | Reply | 0 comments |
headers
Michael Bernstein | 1 Jul 2006 19:30
Picon
Favicon

[quagga-users 7157] Re: Quagga-users Digest, Vol 35, Issue 35

Thank you for your reply.

This gigabit scenario is interesting. If I could just share a short story - so a friend of mine manages a firewall with 100Mb nics for a customer where the CPU was maxing out. They had a conference call and the sales engineer told the customer that they will need to upgrade to 1000Mb nics and the firewall will be fine. So they put in the 1000Mb nics and the CPU was still maxing out!!! No one had thought to actually upgrade the firewall hardware.

Another instance is with a PIX535 where a gigabit line card was installed in a 33MHz bus slot. Cisco says you should run the gigabit card in the 66MHz bus slot.

Gobble - when you say intel's outdated FSB, are you referring to P3, P4? I know P4 uses FSB too.  I am reading the wiki now on HyperTransport.

Yeah, you're right. The 100Mb nics are a waste of time. I was just trying to make the b onding work first and then use 1000Mb nic.

Can you tell me your opinion though whether 32bit vs 64bit makes a difference?

Thank you.
Mike


Gobbledegeek <gobbledegeek-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org> wrote:

>
> Gobble - when you say the system will struggle, what bottlenecks are present
> in the amd64 ht setup? Will running a 64bit linux VS 32bit linux help the
> situation?
>

Hi
I meant to say you might hit a bottleneck with intel's outdated FSB
not on amd with hypertransport.
The rest of your question is best answered by someone who does system
performance banchmarking on the job for a living. I think if you
google on "wirespeed gigabit linux" you will find a lot of people
who've studied performance and capabilities of today's platforms.
Achieving gigabit thruput was a task even f or the folks at Sun on
Solaris and RISC, if I remember right an article I read somewhere...
(:

PS: Why struggle with 10 100mbps nic's when you can have two gigabit
nic's in each system? The more components you have, the more the
probability of failure... the more maintanance and mgmt overhead. I
would inventory my 100mbps nics for reuse in desktops and replace
with gigabit nics on the server.



Rgrds
GobbledeGeek
[Everything but Gobbledegook.. !!]
_______________________________________________
Quagga-users mailing list
Quagga-users-UOy77sIEA+cAd7ICUelF/Q@public.gmane.org
http://lists.quagga.net/mailman/listinfo/quagga-users

Talk is cheap. Use Yahoo! Messenger to make PC-to-Phone calls. Great rates starting at 1¢/min. 
_______________________________________________
Quagga-users mailing list
Quagga-users@...
http://lists.quagga.net/mailman/listinfo/quagga-users
Permalink | Reply | 18 comments |
headers
Markus | 1 Jul 2006 22:52

[quagga-users 7158] Re: Quagga-users Digest, Vol 35, Issue 35

On Sat, 1 Jul 2006 10:30:32 -0700 (PDT)
Michael Bernstein <mb_jobs@...> wrote:

> Yeah, you're right. The 100Mb nics are a waste of time. I was just
> trying to make the bonding work first and then use 1000Mb nic.

Just a bit input regarding forwarding performance:

Either I overlooked it or you didn't specify the reason why you need
load balancing. If your only goal is to achieve 500 Mbit/s wirespeed why
not simply use a GigE NIC? I wanted to throw in something regarding
performance: FreeBSD has a nice feature called "polling". I don't know
if Linux has something similar to polling at this time but anyways,
here's a short excerpt from the manual:

-snip-
Device polling (polling for brevity) refers to a technique that lets the
operating system periodically poll devices, instead of relying on the
devices to generate interrupts when they need attention.  This might
seem inefficient and counterintuitive, but when done properly, polling
gives more control to the operating system on when and how to handle
devices, with a number of advantages in terms of system responsiveness
and performance.
-snip-

And the more interesting thing: we recently had to set up a box acting
as a router with FreeBSD and quagga. The hardware was a Dell Xeon
server, nothing too fancy but definitely not outdated hardware. We used
FreeBSD 4.11-STABLE after some bad experiences with 5.x regarding packet
forwarding performance. Didn't try 6.x or 7.x. So, the results were:
500k pps in + 500k pps out, polling enabled, one ipfw entry (default
allow ip from any to any), full routing table (180k-something entries),
all 64 byte UDP packets: 1-2% CPU consumed = 98-99% idle! This was
actually better than what expected at first. The Dell box seems to be
designed properly. Intel GigE NICs onboard and some additional Intel
GigE NIC in a PCI-X slot. We didn't try to generate more pps because our
test stations were already maxed out... 

I remember someone on this list posting that he achieved 1M pps maximum
using a Dell box.. don't remember the OS though. We shall perform some
more tests to see if we can beat this, but at 98-99% CPU idle and 500k
pps I'm confident we will :-)

Regards
Markus
Permalink | Reply | 0 comments |
headers
Gobbledegeek | 2 Jul 2006 16:57
Picon

[quagga-users 7159] Re: Quagga-users Digest, Vol 35, Issue 35

SPeaking specifically about FSB - All intels have been on outdated
technology for a few years now. So has apple now, with their move to
intel.

The new intel platform launch will rectify this soon with improvements
over hypertransport - I don't remember the name, but the technology is
said to be promising.

64bit PCI bus will allow more data thruput per clock cycle - but its
only found on server chipsets not desktop. You will need a 64bit Gige
card ofcourse. But what I'm telling you is layman stuff that everyone
knows. (:

Rgrds

On 7/1/06, Michael Bernstein <mb_jobs@...> wrote:
> Thank you for your reply.
>
> This gigabit scenario is interesting. If I could just share a short story -
> so a friend of mine manages a firewall with 100Mb nics for a customer where
> the CPU was maxing out. They had a conference call and the sales engineer
> told the customer that they will need to upgrade to 1000Mb nics and the
> firewall will be fine. So they put in the 1000Mb nics and the CPU was still
> maxing out!!! No one had thought to actually upgrade the firewall hardware.
>
> Another instance is with a PIX535 where a gigabit line card was installed in
> a 33MHz bus slot. Cisco says you should run the gigabit card in the 66MHz
> bus slot.
>
> Gobble - when you say intel's outdated FSB, are you referring to P3, P4? I
> know P4 uses FSB too.  I am reading the wiki now on HyperTransport.
>
> Yeah, you're right. The 100Mb nics are a waste of time. I was just trying to
> make the bonding work first and then use 1000Mb nic.
>
> Can you tell me your opinion though whether 32bit vs 64bit makes a
> difference?
>
> Thank you.
> Mike
>
>
>
> Gobbledegeek <gobbledegeek@...> wrote:
>
> >
> > Gobble - when you say the system will struggle, what bottlenecks are
> present
> > in the amd64 ht setup? Will running a 64bit linux VS 32bit linux help the
> > situation?
> >
>
> Hi
> I meant to say you might hit a bottleneck with intel's outdated FSB
> not on amd with hypertransport.
> The rest of your question is best answered by someone who does system
> performance banchmarking on the job for a living. I think if you
> google on "wirespeed gigabit linux" you will find a lot of people
> who've studied performance and capabilities of today's platforms.
> Achieving gigabit thruput was a task even for the folks at Sun on
> Solaris and RISC, if I remember right an article I read somewhere...
> (:
>
> PS: Why struggle with 10 100mbps nic's when you can have two gigabit
> nic's in each system? The more components you have, the more the
> probability of failure... the more maintanance and mgmt overhead. I
> would inventory my 100mbps nics for reuse in desktops and replace
> with gigabit nics on the server.
>
>
>
> Rgrds
> GobbledeGeek
> [Everything but Gobbledegook.. !!]
> _______________________________________________
> Quagga-users mailing list
> Quagga-users@...
> http://lists.quagga.net/mailman/listinfo/quagga-users
>
>
>
>
> ________________________________
> Talk is cheap. Use Yahoo! Messenger to make PC-to-Phone calls. Great rates
> starting at 1ยข/min.
>
>

--

-- 
Nonchalantly yours
GobbledeGeek
[Everything but Gobbledegook.. !!]
Permalink | Reply | 16 comments |
headers
Paul Jakma | 2 Jul 2006 17:12
Picon

[quagga-users 7160] Re: per packet load balancing

On Wed, 28 Jun 2006, Michael Bernstein wrote:

> Would it be possible to program a solution for Quagga/Zebra that could 
> do this, maybe implement something in ospfd?

Wrong place - this would have to be done in the kernel of your OS.

regards,
--

-- 
Paul Jakma,
Network Approachability, KISS.           Sun Microsystems, Dublin, Ireland.
http://opensolaris.org/os/project/quagga tel: EMEA x19190 / +353 1 819 9190
Permalink | Reply | 6 comments |
headers
Daniel Pocock | 2 Jul 2006 19:58
Picon

[quagga-users 7161] Redundant load balanced Linux solution


Hi,

I've just documented a redundant, load balanced routing solution for 
Linux servers (e.g. web servers, Asterisk servers) using BGP and OSPF.  
This is based on the recent discussion I started on the list about 
source addresses.

If anyone can see any faults with this or suggest improvements, please 
let me know:

    http://www.readytechnology.co.uk/open/bgp/loadbalanced.html

Regards,

Daniel
Permalink | Reply | 0 comments |
headers
Robbert-Jan Sperna Weiland | 2 Jul 2006 21:36
Picon
Favicon

[quagga-users 7162] BGP session password


Hello,

How can I configure quagga to use password protected BGP sessions?
I read somewhere to use:

neighbor x.x.x.x password the_BGP_session_password
neighbor xxx ASN xxx
neighbor x.x.x.x <etc>

in bgpd.conf. Is this correct? Is it that simple to enable and use password 
protected BGP sessions?
Please let me know.

Thank you very much in advance for any help.
Permalink | Reply | 5 comments |
headers
The Admin | 3 Jul 2006 02:24
Picon
Favicon

[quagga-users 7163] Re: BGP session password

>
>Hello,
>
>How can I configure quagga to use password protected BGP sessions?
>I read somewhere to use:
>
>neighbor x.x.x.x password the_BGP_session_password
>neighbor xxx ASN xxx
>neighbor x.x.x.x <etc>
>
>in bgpd.conf. Is this correct? Is it that simple to enable and use password 
>protected BGP sessions?
>Please let me know.

Actually I just found you first have to state the remote ASN, like:

neighbor xxx ASN xxx
neighbor x.x.x.x password the_BGP_session_password
neighbor x.x.x.x <etc>

>
>Thank you very much in advance for any help.
>

But still, it that everything to it? Can I count on it to work like this? :p
Permalink | Reply | 0 comments |
headers
Paul Jakma | 3 Jul 2006 02:32
Picon

[quagga-users 7164] Re: BGP session password

On Sun, 2 Jul 2006, Robbert-Jan Sperna Weiland wrote:

> in bgpd.conf. Is this correct? Is it that simple to enable and use 
> password protected BGP sessions? Please let me know.

At the moment, this is only available in the FreeBSD port of Quagga.

So if you're using that, then "yes" otherwise "no".

regards,
--

-- 
Paul
Jakma	paul@...	paul@...	Key
ID: 64A2FF6A
Fortune:
Indecision is the true basis for flexibility.
Permalink | Reply | 3 comments |

Gmane