Mailing List Archive: Cisco: NSP

OSPF fast hellos

Index | Next | Previous | View Threaded

ben.steele at internode

Oct 28, 2008, 10:39 PM

Post #1 of 9 (1458 views)
Permalink

OSPF fast hellos

Anyone currently using this in a fairly demanding environment? Ie 5-10Gbs+
Campus/DC model.

Curious as to whether you've had any/many false dead peers with such a short
interval, subsecond dead peer detection does sound very temping though.

Cheers

Ben

_______________________________________________
cisco-nsp mailing list cisco-nsp [at] puck
https://puck.nether.net/mailman/listinfo/cisco-nsp
archive at http://puck.nether.net/pipermail/cisco-nsp/

rodunn at cisco

Oct 29, 2008, 6:10 AM

Post #2 of 9 (1417 views)
Permalink

Re: OSPF fast hellos [In reply to]

Why don't you use BFD instead. It's designed with something called
pseudo preemption from an OS scheduler perspective that helps
reduce false positives and the fact that BFD frames are handled
under interrupt and not process scheduled for rx/tx.

Rodney

On Wed, Oct 29, 2008 at 04:09:45PM +1030, Ben Steele wrote:
> Anyone currently using this in a fairly demanding environment? Ie 5-10Gbs+
> Campus/DC model.
>
>
>
> Curious as to whether you've had any/many false dead peers with such a short
> interval, subsecond dead peer detection does sound very temping though.
>
>
>
> Cheers
>
>
>
> Ben
>
>
>
>
>
> _______________________________________________
> cisco-nsp mailing list cisco-nsp [at] puck
> https://puck.nether.net/mailman/listinfo/cisco-nsp
> archive at http://puck.nether.net/pipermail/cisco-nsp/
_______________________________________________
cisco-nsp mailing list cisco-nsp [at] puck
https://puck.nether.net/mailman/listinfo/cisco-nsp
archive at http://puck.nether.net/pipermail/cisco-nsp/

jackson.tim at gmail

Oct 29, 2008, 6:40 AM

Post #3 of 9 (1404 views)
Permalink

Re: OSPF fast hellos [In reply to]

When is BFD going to not be limited to 7600/12k/CRS?

And when can we get BFD on an SVI (or back on an SVI, iirc SRB supports
this, but SRC doesn't?) or a port-channel?

Until Cisco actually has BFD working on more than a few platforms, I'll
stick with fast hellos since it seems to work on more platforms and in more
configurations...

--
Tim

On Wed, Oct 29, 2008 at 8:10 AM, Rodney Dunn <rodunn [at] cisco> wrote:

> Why don't you use BFD instead. It's designed with something called
> pseudo preemption from an OS scheduler perspective that helps
> reduce false positives and the fact that BFD frames are handled
> under interrupt and not process scheduled for rx/tx.
>
> Rodney
>
> On Wed, Oct 29, 2008 at 04:09:45PM +1030, Ben Steele wrote:
> > Anyone currently using this in a fairly demanding environment? Ie
> 5-10Gbs+
> > Campus/DC model.
> >
> >
> >
> > Curious as to whether you've had any/many false dead peers with such a
> short
> > interval, subsecond dead peer detection does sound very temping though.
> >
> >
> >
> > Cheers
> >
> >
> >
> > Ben
> >
> >
> >
> >
> >
> > _______________________________________________
> > cisco-nsp mailing list cisco-nsp [at] puck
> > https://puck.nether.net/mailman/listinfo/cisco-nsp
> > archive at http://puck.nether.net/pipermail/cisco-nsp/
> _______________________________________________
> cisco-nsp mailing list cisco-nsp [at] puck
> https://puck.nether.net/mailman/listinfo/cisco-nsp
> archive at http://puck.nether.net/pipermail/cisco-nsp/
>
_______________________________________________
cisco-nsp mailing list cisco-nsp [at] puck
https://puck.nether.net/mailman/listinfo/cisco-nsp
archive at http://puck.nether.net/pipermail/cisco-nsp/

rodunn at cisco

Oct 29, 2008, 7:07 AM

Post #4 of 9 (1404 views)
Permalink

Re: OSPF fast hellos [In reply to]

On Wed, Oct 29, 2008 at 08:40:16AM -0500, Tim Jackson wrote:
> When is BFD going to not be limited to 7600/12k/CRS?
>

It's on the ISR's.
Other support (ie: WAN interfaces) are in the works.

> And when can we get BFD on an SVI (or back on an SVI, iirc SRB supports this,
> but SRC doesn't?) or a port-channel?

Haven't checked on that.

>
> Until Cisco actually has BFD working on more than a few platforms, I'll stick
> with fast hellos since it seems to work on more platforms and in more
> configurations...

That's fair.

Rodney

>
> --
> Tim
>
> On Wed, Oct 29, 2008 at 8:10 AM, Rodney Dunn <rodunn [at] cisco> wrote:
>
> Why don't you use BFD instead. It's designed with something called
> pseudo preemption from an OS scheduler perspective that helps
> reduce false positives and the fact that BFD frames are handled
> under interrupt and not process scheduled for rx/tx.
>
> Rodney
>
> On Wed, Oct 29, 2008 at 04:09:45PM +1030, Ben Steele wrote:
> > Anyone currently using this in a fairly demanding environment? Ie
> 5-10Gbs+
> > Campus/DC model.
> >
> >
> >
> > Curious as to whether you've had any/many false dead peers with such a
> short
> > interval, subsecond dead peer detection does sound very temping though.
> >
> >
> >
> > Cheers
> >
> >
> >
> > Ben
> >
> >
> >
> >
> >
> > _______________________________________________
> > cisco-nsp mailing list cisco-nsp [at] puck
> > https://puck.nether.net/mailman/listinfo/cisco-nsp
> > archive at http://puck.nether.net/pipermail/cisco-nsp/
> _______________________________________________
> cisco-nsp mailing list cisco-nsp [at] puck
> https://puck.nether.net/mailman/listinfo/cisco-nsp
> archive at http://puck.nether.net/pipermail/cisco-nsp/
>
>
_______________________________________________
cisco-nsp mailing list cisco-nsp [at] puck
https://puck.nether.net/mailman/listinfo/cisco-nsp
archive at http://puck.nether.net/pipermail/cisco-nsp/

pshuleski at gmail

Oct 29, 2008, 11:23 AM

Post #5 of 9 (1395 views)
Permalink

Re: OSPF fast hellos [In reply to]

Changing the hello/dead timers combined with ispf, and tuning spf
throttle/lsa throttle timers, works extremely well to bring down the
convergence time of OSPF to around what eigrp does natively. You dont want
to adjust the timers without the throttle control. There should be a best
practices document on the timer values somewhere on cisco.com. Can't seem
to find it at the moment. I was using that formula on a large, mixed
OC-48/10G ospf routed network without any issues.

--Pete

On Wed, Oct 29, 2008 at 1:39 AM, Ben Steele <ben.steele [at] internode>wrote:

> Anyone currently using this in a fairly demanding environment? Ie 5-10Gbs+
> Campus/DC model.
>
>
>
> Curious as to whether you've had any/many false dead peers with such a
> short
> interval, subsecond dead peer detection does sound very temping though.
>
>
>
> Cheers
>
>
>
> Ben
>
>
>
>
>
> _______________________________________________
> cisco-nsp mailing list cisco-nsp [at] puck
> https://puck.nether.net/mailman/listinfo/cisco-nsp
> archive at http://puck.nether.net/pipermail/cisco-nsp/
>
_______________________________________________
cisco-nsp mailing list cisco-nsp [at] puck
https://puck.nether.net/mailman/listinfo/cisco-nsp
archive at http://puck.nether.net/pipermail/cisco-nsp/

ben.steele at internode

Oct 29, 2008, 2:36 PM

Post #6 of 9 (1399 views)
Permalink

Re: OSPF fast hellos [In reply to]

Because I couldn't see bfd support for 3750's, best it can do is UDLD,
otherwise that would be my preferred method.

Are you advising against fast hello's? Have you seen many issues with people
using them?

-----Original Message-----
From: Rodney Dunn [mailto:rodunn [at] cisco]
Sent: Wednesday, 29 October 2008 11:41 PM
To: Ben Steele
Cc: cisco-nsp [at] puck
Subject: Re: [c-nsp] OSPF fast hellos

Why don't you use BFD instead. It's designed with something called
pseudo preemption from an OS scheduler perspective that helps
reduce false positives and the fact that BFD frames are handled
under interrupt and not process scheduled for rx/tx.

Rodney

On Wed, Oct 29, 2008 at 04:09:45PM +1030, Ben Steele wrote:
> Anyone currently using this in a fairly demanding environment? Ie 5-10Gbs+
> Campus/DC model.
>
>
>
> Curious as to whether you've had any/many false dead peers with such a
short
> interval, subsecond dead peer detection does sound very temping though.
>
>
>
> Cheers
>
>
>
> Ben
>
>
>
>
>
> _______________________________________________
> cisco-nsp mailing list cisco-nsp [at] puck
> https://puck.nether.net/mailman/listinfo/cisco-nsp
> archive at http://puck.nether.net/pipermail/cisco-nsp/

No virus found in this incoming message.
Checked by AVG - http://www.avg.com
Version: 8.0.175 / Virus Database: 270.8.4/1752 - Release Date: 28/10/2008
10:04 AM

_______________________________________________
cisco-nsp mailing list cisco-nsp [at] puck
https://puck.nether.net/mailman/listinfo/cisco-nsp
archive at http://puck.nether.net/pipermail/cisco-nsp/

rodunn at cisco

Oct 30, 2008, 10:30 AM

Post #7 of 9 (1395 views)
Permalink

Re: OSPF fast hellos [In reply to]

On Thu, Oct 30, 2008 at 08:06:36AM +1030, Ben Steele wrote:
> Because I couldn't see bfd support for 3750's, best it can do is UDLD,
> otherwise that would be my preferred method.
>
> Are you advising against fast hello's?

No totally.

Have you seen many issues with people
> using them?

Yes. They have to be scheduled on the CPU as a process and that is more
variable because IOS is run to completion, except for psuedo preemption
added for BFD.

Even that isn't 100% bullet proof but it's better than OSPF fast hellos
from that perspective.

>
> -----Original Message-----
> From: Rodney Dunn [mailto:rodunn [at] cisco]
> Sent: Wednesday, 29 October 2008 11:41 PM
> To: Ben Steele
> Cc: cisco-nsp [at] puck
> Subject: Re: [c-nsp] OSPF fast hellos
>
> Why don't you use BFD instead. It's designed with something called
> pseudo preemption from an OS scheduler perspective that helps
> reduce false positives and the fact that BFD frames are handled
> under interrupt and not process scheduled for rx/tx.
>
> Rodney
>
> On Wed, Oct 29, 2008 at 04:09:45PM +1030, Ben Steele wrote:
> > Anyone currently using this in a fairly demanding environment? Ie 5-10Gbs+
> > Campus/DC model.
> >
> >
> >
> > Curious as to whether you've had any/many false dead peers with such a
> short
> > interval, subsecond dead peer detection does sound very temping though.
> >
> >
> >
> > Cheers
> >
> >
> >
> > Ben
> >
> >
> >
> >
> >
> > _______________________________________________
> > cisco-nsp mailing list cisco-nsp [at] puck
> > https://puck.nether.net/mailman/listinfo/cisco-nsp
> > archive at http://puck.nether.net/pipermail/cisco-nsp/
>
> No virus found in this incoming message.
> Checked by AVG - http://www.avg.com
> Version: 8.0.175 / Virus Database: 270.8.4/1752 - Release Date: 28/10/2008
> 10:04 AM
_______________________________________________
cisco-nsp mailing list cisco-nsp [at] puck
https://puck.nether.net/mailman/listinfo/cisco-nsp
archive at http://puck.nether.net/pipermail/cisco-nsp/

frnkblk at iname

Oct 30, 2008, 8:38 PM

Post #8 of 9 (1396 views)
Permalink

Re: OSPF fast hellos [In reply to]

If you can get BFD support worked into the 3750ME, we wouldn't have to mess
with OSPF fast hellos. =)

Frank

-----Original Message-----
From: cisco-nsp-bounces [at] puck
[mailto:cisco-nsp-bounces [at] puck] On Behalf Of Rodney Dunn
Sent: Thursday, October 30, 2008 12:30 PM
To: Ben Steele
Cc: cisco-nsp [at] puck
Subject: Re: [c-nsp] OSPF fast hellos

On Thu, Oct 30, 2008 at 08:06:36AM +1030, Ben Steele wrote:
> Because I couldn't see bfd support for 3750's, best it can do is UDLD,
> otherwise that would be my preferred method.
>
> Are you advising against fast hello's?

No totally.

Have you seen many issues with people
> using them?

Yes. They have to be scheduled on the CPU as a process and that is more
variable because IOS is run to completion, except for psuedo preemption
added for BFD.

Even that isn't 100% bullet proof but it's better than OSPF fast hellos
from that perspective.

>
> -----Original Message-----
> From: Rodney Dunn [mailto:rodunn [at] cisco]
> Sent: Wednesday, 29 October 2008 11:41 PM
> To: Ben Steele
> Cc: cisco-nsp [at] puck
> Subject: Re: [c-nsp] OSPF fast hellos
>
> Why don't you use BFD instead. It's designed with something called
> pseudo preemption from an OS scheduler perspective that helps
> reduce false positives and the fact that BFD frames are handled
> under interrupt and not process scheduled for rx/tx.
>
> Rodney
>
> On Wed, Oct 29, 2008 at 04:09:45PM +1030, Ben Steele wrote:
> > Anyone currently using this in a fairly demanding environment? Ie
5-10Gbs+
> > Campus/DC model.
> >
> >
> >
> > Curious as to whether you've had any/many false dead peers with such a
> short
> > interval, subsecond dead peer detection does sound very temping though.
> >
> >
> >
> > Cheers
> >
> >
> >
> > Ben
> >
> >
> >
> >
> >
> > _______________________________________________
> > cisco-nsp mailing list cisco-nsp [at] puck
> > https://puck.nether.net/mailman/listinfo/cisco-nsp
> > archive at http://puck.nether.net/pipermail/cisco-nsp/
>
> No virus found in this incoming message.
> Checked by AVG - http://www.avg.com
> Version: 8.0.175 / Virus Database: 270.8.4/1752 - Release Date: 28/10/2008
> 10:04 AM
_______________________________________________
cisco-nsp mailing list cisco-nsp [at] puck
https://puck.nether.net/mailman/listinfo/cisco-nsp
archive at http://puck.nether.net/pipermail/cisco-nsp/

_______________________________________________
cisco-nsp mailing list cisco-nsp [at] puck
https://puck.nether.net/mailman/listinfo/cisco-nsp
archive at http://puck.nether.net/pipermail/cisco-nsp/

Curtis at GreenKey

Oct 31, 2008, 5:39 PM

Post #9 of 9 (1392 views)
Permalink

Re: OSPF fast hellos [In reply to]

And beware the showstopper in 12.2S and 12.4T where BFD causes an NMI
reset. Doh!

It haunted me for many moons before my stubborness prevailed. And a sharp
Escalations Engineer at Cisco sleuthed it out and eventually generated bug
ID CSCek75694 with a fix that is supposed to wind back into the mainline
trains by EOY.

../C

Yesterday Frank Bulk said:

> If you can get BFD support worked into the 3750ME, we wouldn't have to mess
> with OSPF fast hellos. =)
>
> Frank
>
> -----Original Message-----
> From: cisco-nsp-bounces [at] puck
> [mailto:cisco-nsp-bounces [at] puck] On Behalf Of Rodney Dunn
> Sent: Thursday, October 30, 2008 12:30 PM
> To: Ben Steele
> Cc: cisco-nsp [at] puck
> Subject: Re: [c-nsp] OSPF fast hellos
>
> On Thu, Oct 30, 2008 at 08:06:36AM +1030, Ben Steele wrote:
>> Because I couldn't see bfd support for 3750's, best it can do is UDLD,
>> otherwise that would be my preferred method.
>>
>> Are you advising against fast hello's?
>
> No totally.
>
> Have you seen many issues with people
>> using them?
>
> Yes. They have to be scheduled on the CPU as a process and that is more
> variable because IOS is run to completion, except for psuedo preemption
> added for BFD.
>
> Even that isn't 100% bullet proof but it's better than OSPF fast hellos
> from that perspective.
>
>
>
>>
>> -----Original Message-----
>> From: Rodney Dunn [mailto:rodunn [at] cisco]
>> Sent: Wednesday, 29 October 2008 11:41 PM
>> To: Ben Steele
>> Cc: cisco-nsp [at] puck
>> Subject: Re: [c-nsp] OSPF fast hellos
>>
>> Why don't you use BFD instead. It's designed with something called
>> pseudo preemption from an OS scheduler perspective that helps
>> reduce false positives and the fact that BFD frames are handled
>> under interrupt and not process scheduled for rx/tx.
>>
>> Rodney
>>
>> On Wed, Oct 29, 2008 at 04:09:45PM +1030, Ben Steele wrote:
>>> Anyone currently using this in a fairly demanding environment? Ie
> 5-10Gbs+
>>> Campus/DC model.
>>>
>>>
>>>
>>> Curious as to whether you've had any/many false dead peers with such a
>> short
>>> interval, subsecond dead peer detection does sound very temping though.
>>>
>>>
>>>
>>> Cheers
>>>
>>>
>>>
>>> Ben
>>>
>>>
>>>
>>>
>>>
>>> _______________________________________________
>>> cisco-nsp mailing list cisco-nsp [at] puck
>>> https://puck.nether.net/mailman/listinfo/cisco-nsp
>>> archive at http://puck.nether.net/pipermail/cisco-nsp/
>>
>> No virus found in this incoming message.
>> Checked by AVG - http://www.avg.com
>> Version: 8.0.175 / Virus Database: 270.8.4/1752 - Release Date: 28/10/2008
>> 10:04 AM
> _______________________________________________
> cisco-nsp mailing list cisco-nsp [at] puck
> https://puck.nether.net/mailman/listinfo/cisco-nsp
> archive at http://puck.nether.net/pipermail/cisco-nsp/
>
> _______________________________________________
> cisco-nsp mailing list cisco-nsp [at] puck
> https://puck.nether.net/mailman/listinfo/cisco-nsp
> archive at http://puck.nether.net/pipermail/cisco-nsp/
>

_______________________________________________
cisco-nsp mailing list cisco-nsp [at] puck
https://puck.nether.net/mailman/listinfo/cisco-nsp
archive at http://puck.nether.net/pipermail/cisco-nsp/

Index | Next | Previous | View Threaded

Interested in having your list archived? Contact Gossamer Threads