Page 1 of 1

Mikrotik still chokes on full bgp routing table

Posted: Thu Apr 27, 2006 12:32 pm
by ksswd
Mikrotik is still unable to work properly after downloading full routing table (over 180000 prefixes). This was on MT 2.9.22 with routing and routing-test packages, 512MB memory, PIII 2.4GHz. Both BGP and OSPF while loading full routing table caused mt to go 100% CPU.
Has anybody successfuly tried to use mt as bgp router?

Piotr

Posted: Thu Apr 27, 2006 7:32 pm
by macgaiver
I think that routing table of 180000 prefixes takes lots and lots of calculation and it is just normal to have 100% CPU load on week CPUs

You should try Athlon64 3xxx+

Re: Mikrotik still chokes on full bgp routing table

Posted: Fri Apr 28, 2006 12:51 am
by sten
Mikrotik is still unable to work properly after downloading full routing table (over 180000 prefixes). This was on MT 2.9.22 with routing and routing-test packages, 512MB memory, PIII 2.4GHz. Both BGP and OSPF while loading full routing table caused mt to go 100% CPU.
Has anybody successfuly tried to use mt as bgp router?

Piotr
Your spec's are way too low for said job.

Posted: Fri Apr 28, 2006 1:16 am
by changeip
I think he meant to say P4 2.4 ... not PIII. I don't think they even have p3s above 1.7 or so. A p4 2.4 is enough for bgp i would think. CPU will hit 100% for about 5 minutes but it should not affect packet routing during this time. Is it?

Sam

Posted: Fri Apr 28, 2006 3:55 am
by jp
Ciscos, known for the bloat, can handle 2 full tables pretty easy. The 7500's RSP4 has a R5000 CPU at 200Mhz, and can load both tables in probably less than 2 minutes, which is probably as fast as my peers can dish them out. It's old technology too. I would expect any gigahertz PC to do it in a fraction of the time.

Posted: Fri Apr 28, 2006 8:52 am
by normis
Developers will consider it a bug if it will not use 100% of CPU resources. If it is 100% then it means that cpu is working at it's full potential, and it does not mean that it is at it's limit. Even at 100% it will still work great, it just means that CPU does what it can.

Posted: Fri Apr 28, 2006 9:20 am
by sten
Ciscos, known for the bloat, can handle 2 full tables pretty easy. The 7500's RSP4 has a R5000 CPU at 200Mhz, and can load both tables in probably less than 2 minutes, which is probably as fast as my peers can dish them out. It's old technology too. I would expect any gigahertz PC to do it in a fraction of the time.
Cisco, also known for their 20 odd years (if not more) of developing their IOS have picked up a thing or two during all those years of development. I wouldn't be so quicky to assume such things as;
They have the same architectural challanges, bottlenecks and strengths as a x86 consumer platform. That their code is equally efficient. That the overhead is the same.

Anything zebra bgpd based is, my opinion, a good example in bad programming and inefficient design.

Posted: Fri Apr 28, 2006 9:35 am
by normis
good thing that we don't use zebra :)

Posted: Mon Jun 19, 2006 3:32 am
by rkj
good thing that we don't use zebra :)
2.9.25+ routing-test is based on Quagga, OpenBGPD, XORP or something else ?

BGP code on 2.9.25 seems very good, but it failed to do an export:
"#error exporting /routing rip interface". Will test 2.9.26/routing-test soon to check this out.

Posted: Mon Jun 19, 2006 3:52 am
by Beccara
routing-test is a ground-up rebuild and is not based on anything like quagga IIRC

about BGP

Posted: Mon Jun 19, 2006 9:27 am
by heimdal
I have 2.9.25 with routing-test, works for me but have strange "reload"
on BGP sessions once for last 48 hours. For now this is not a problem,
router with P4 3GHz, 1 G RAM, 4 Intel giga ethernet (one dual + two integrated on Intel Server Board). Load 2x 185 000 routes from two ISPs for about 2 minutes. Works ...except this "reload" ...

Posted: Mon Jun 19, 2006 11:12 am
by grzesjan
Still one thing is not corrected. Mikrotik show status=connect and the other side shows status=established. Session is OK, prefixes are being sent both ways.

Gregor

Posted: Mon Jun 19, 2006 2:46 pm
by nikhil
Still one thing is not corrected. Mikrotik show status=connect and the other side shows status=established. Session is OK, prefixes are being sent both ways.

Gregor
Yes even the uptime in winbox is wrong because uptime should show the connection uptime and not just plain uptime

Re: about BGP

Posted: Mon Jun 19, 2006 4:30 pm
by spirosco
I have 2.9.25 with routing-test, works for me but have strange "reload" on BGP sessions once for last 48 hours.
I was having the same problem with 2.9.25 r-t.
My routing table is not so big as of the guys above :) , its about 300 routes.
With 2.9.26 r-t and with increased hold timers (from 10 to 90) i have 2 days+ without a glitch.
What i have not been able so far to test on 2.9.26, is the correct removal of the routes.
I hope tomorrow to give it a try and hopefully come back with positive results :)