When routes are deleted (for ex., terminating a lot of PPP tunnels for some reason in same time), it seems like each removed route is sent into separate OSPF packet that causes troubles with OSPF in some cases (for ex., there's troubles with old quagga). Is there a possibility to add some rate-limiting (for ex., limit to 10 messages per second), and aggregate separate 'route deleted' messages to one packet?
OSPF doesn't delete "routes", it sends LSAs. It seems that you are redistributing a lot of single IP addresses in OSPF (as external routes). So there's one LSA type 5 for each single redistributed /32. OSPF is not the proper choice for this usecase, and it's not a Quagga or BIRD issue. BGP is the protocol to use for such cases. Many routes, few messages, scalability.
Le 14 juil. 2015 à 09:20, Andrew <nitr0@seti.kr.ua> a écrit :
When routes are deleted (for ex., terminating a lot of PPP tunnels for some reason in same time), it seems like each removed route is sent into separate OSPF packet that causes troubles with OSPF in some cases (for ex., there's troubles with old quagga).
Is there a possibility to add some rate-limiting (for ex., limit to 10 messages per second), and aggregate separate 'route deleted' messages to one packet?
Olivier, Whilst what you say is correct as far as it goes (and BGP may well be a better idea), if I believe the OSPF protocol works it permits more than one LSA to be encapsulated into each OSPF link state update packet, so Andrew's assertion that aggregating LSA changes into a smaller number of packets would be a good idea is valid. I believe what he is asking for is the similar to Cisco's OSPF LSA throttling: http://www.cisco.com/c/en/us/td/docs/ios/12_0s/feature/guide/fsolsath.html Note even before that feature was introduced Cisco throttled LSAs:
Prior to the OSPF LSA Throttling feature, LSA generation was rate-limited for 5 seconds. That meant that changes in an LSA could not be propagated in milliseconds, so the OSPF network could not achieve millisecond convergence.
and I think he's merely asking for similar behaviour. Alex On 14 Jul 2015, at 15:47, Olivier Benghozi <olivier.benghozi@wifirst.fr> wrote:
OSPF doesn't delete "routes", it sends LSAs. It seems that you are redistributing a lot of single IP addresses in OSPF (as external routes). So there's one LSA type 5 for each single redistributed /32.
OSPF is not the proper choice for this usecase, and it's not a Quagga or BIRD issue. BGP is the protocol to use for such cases. Many routes, few messages, scalability.
Le 14 juil. 2015 à 09:20, Andrew <nitr0@seti.kr.ua> a écrit :
When routes are deleted (for ex., terminating a lot of PPP tunnels for some reason in same time), it seems like each removed route is sent into separate OSPF packet that causes troubles with OSPF in some cases (for ex., there's troubles with old quagga).
Is there a possibility to add some rate-limiting (for ex., limit to 10 messages per second), and aggregate separate 'route deleted' messages to one packet?
-- Alex Bligh
I mean host routes mass removing. Maybe OSPF really isn't best choice for this. Earlier, with quagga, there was no troubles with it. I didn't look deeply to RFC, so I thouht that single LSA can carry multiple prefixes. Possibly quagga just sends packets slower... 14.07.2015 17:47, Olivier Benghozi пишет:
OSPF doesn't delete "routes", it sends LSAs. It seems that you are redistributing a lot of single IP addresses in OSPF (as external routes). So there's one LSA type 5 for each single redistributed /32.
OSPF is not the proper choice for this usecase, and it's not a Quagga or BIRD issue. BGP is the protocol to use for such cases. Many routes, few messages, scalability.
Le 14 juil. 2015 à 09:20, Andrew <nitr0@seti.kr.ua> a écrit :
When routes are deleted (for ex., terminating a lot of PPP tunnels for some reason in same time), it seems like each removed route is sent into separate OSPF packet that causes troubles with OSPF in some cases (for ex., there's troubles with old quagga).
Is there a possibility to add some rate-limiting (for ex., limit to 10 messages per second), and aggregate separate 'route deleted' messages to one packet?
On Tue, Jul 14, 2015 at 10:20:22AM +0300, Andrew wrote:
When routes are deleted (for ex., terminating a lot of PPP tunnels for some reason in same time), it seems like each removed route is sent into separate OSPF packet that causes troubles with OSPF in some cases (for ex., there's troubles with old quagga).
Is there a possibility to add some rate-limiting (for ex., limit to 10 messages per second), and aggregate separate 'route deleted' messages to one packet?
AFAIK, multiple flush LSAs can be packed to one LSA when flushed together (i.e. the routes are deleted together), but there is no time-limit. If these routes are deleted as separate (but immediate) events, then it is possible that the OSPF packet with just one flushed LSA is sent before waiting for next route delete (and therefore next flushed LSA. -- Elen sila lumenn' omentielvo Ondrej 'Santiago' Zajicek (email: santiago@crfreenet.org) OpenPGP encrypted e-mails preferred (KeyID 0x11DEADC3, wwwkeys.pgp.net) "To err is human -- to blame it on a computer is even more so."
18.07.2015 00:36, Ondrej Zajicek пишет:
On Tue, Jul 14, 2015 at 10:20:22AM +0300, Andrew wrote:
When routes are deleted (for ex., terminating a lot of PPP tunnels for some reason in same time), it seems like each removed route is sent into separate OSPF packet that causes troubles with OSPF in some cases (for ex., there's troubles with old quagga).
Is there a possibility to add some rate-limiting (for ex., limit to 10 messages per second), and aggregate separate 'route deleted' messages to one packet? AFAIK, multiple flush LSAs can be packed to one LSA when flushed together (i.e. the routes are deleted together), but there is no time-limit. If these routes are deleted as separate (but immediate) events, then it is possible that the OSPF packet with just one flushed LSA is sent before waiting for next route delete (and therefore next flushed LSA.
Maybe it'll be good to add some delay (for ex., 10ms) before flush LSA will be sent?
Am 18.07.2015 um 10:54 schrieb Andrew:
18.07.2015 00:36, Ondrej Zajicek пишет:
On Tue, Jul 14, 2015 at 10:20:22AM +0300, Andrew wrote:
When routes are deleted (for ex., terminating a lot of PPP tunnels for some reason in same time), it seems like each removed route is sent into separate OSPF packet that causes troubles with OSPF in some cases (for ex., there's troubles with old quagga).
Is there a possibility to add some rate-limiting (for ex., limit to 10 messages per second), and aggregate separate 'route deleted' messages to one packet? AFAIK, multiple flush LSAs can be packed to one LSA when flushed together (i.e. the routes are deleted together), but there is no time-limit. If these routes are deleted as separate (but immediate) events, then it is possible that the OSPF packet with just one flushed LSA is sent before waiting for next route delete (and therefore next flushed LSA.
Maybe it'll be good to add some delay (for ex., 10ms) before flush LSA will be sent? please allow me to quote in this context: Am 08.06.2015 um 13:18 schrieb Ondrej Zajicek: On Mon, Jun 08, 2015 at 12:36:07PM +0200, Gruber Gerold wrote:
please allow me to add the pointer to the also wished "noise suppression" in OSPF, where a delay between appearance or disappearance of a route and it's propagation to the neighbours would also improve the stability of the landscape and lower the "noise" caused by flapping routes. (Cisco boxes behave this way, from our observations.)
Hi
This is AFAIK fixed in 1.5.0 - LSAs are originated at most once per MinLSInterval (5 s).
So if there is a "time trigger" for sending LSAs: should it not be possible to filter (in the sense drop vanish/reappear pairs) or aggregate multiple changes into one LSA message, as Andrew suggests? Best regards Gerold
participants (5)
-
Alex Bligh -
Andrew -
Gruber Gerold -
Olivier Benghozi -
Ondrej Zajicek