MikroTik

Hello,

My BGP peers keep crashing (3 full routing table). With both routing-test and routing packages (since ROS 3.7 currently I'm using 3.28).
Is MK looking into finding a resolution for this is there any work going ahead????
Is there particular hardware issues with x86 I use hardware from the approaved list.
Could someone give an anwser to this if MK are looking for a solution to the problem (this one is for Normis).
I'm having crashing problems with BGP since ROS 3.7.

Thanks,

Sotiris

just wondering - why do you need 3 full routing tables?..

This happens with a single peer as well. I tested that as well.

is the hardware crashing, or are you just dropping the peer sessions ? What exactly happens, and how far into it. A little more info should help us troubleshoot.

The peers crash. The go from established to idle the entire routing table is lost. And everything has to start all over again dpownloadiung all the routes.
When I log into the box and I try to view the peers it takes some comsiderable amount of time to display the print.
This happens once every couple of days.

Thanks,
Sotiris

Soltiris,

What Ethernet cards are in your setup?

Do you use Marvell PCI-E maybe ?

Hello exe,

Nop I'm using an RB44 gigabit at 1Gbps. I don't know what other information you might need. I've been strugling with this for some time now.


Thanks,

Sotiris

check your memory modules with memtestx. make sure that you don't have any IRQs that aren't being used, ie disable all peripherals in the bios that you are not using. try turning off multi-cpu. how much memory do you have and how much is free?

post a /system resources pci print

also, does the log have anything about it? i wonder if its the remote side terminating the session, or tcp/179 packets getting blocked, etc. Is the hold timer just expiring or is it an abrupt close?

Hello guys,

It is an abrupt shut because all the peers go down at the same time. I have 2GB of memory in the system and 600MB are used. Any systems that can be disabled from the bios and they are not used they have been disabled. The IRQ's look fine they all corespond to periferals I have on the server (i.e. ethernet ports).

The only thing I haven't done so far is to disable the multi-cpu.
On the logs I get the connection has terminated and the system waits for the BGP peers to get reastablished. The firewall allows TCP and UDP connections on port 179. So that shouldn't be the problem.

Thanks,

Sotiris

v3.28 and earlier versions had route redistribution bug. Send supout to support at mikrotik.com.

what kind of bug? could you please be more precise?..

Route redistribution crash. It shows up especially if there are multiple BGP peers.

v3.28 is the currently stable version, do you suggest to use 4.0 devel if i plan to deploy on a similar architecture ?

Im about to switch all my routers to Mikrotik, so i dont want to run into the same problem

synologic, are you using BGP full view?.. if no - then 3.28 is good enough, I believe. or you may want to wait for 3.29

Yes, i will be having 2 global bgp tables and 3 local peering tables of about 6k routes each ... any ideea when 3.29 is going to be released ?

guys, are you Tiers 1? =) why do you need full view? %-)

MT has fixed the bug into the 3.30
I am testing it and seems more stable, no crashes after 3days of operation.

regards
Ros

Hopefully 3.30 will be stable enough for the BGP.

Sotiris

rpingar how many active BGP peers do you have on your testing platform and how many route entries ???


Sotiris

2 peer with full route
28 peers with an average of 100 routes each

regards
Ros

3.30?.. where's 3.29? %)

Its on the RB750s. If 3.30 is stable - give me 3.30 I don't care about 3.29

Chupaka, yes we are using Tier 1 providers

I'm also interested in testing 3.30, so that's right ... where is it ?

You normally have to ask mikrotik if you specifically want to check out a new pre-release version.
And its normally because you've reported the bug that they'll offer you the fix.

MT has fixed the bug into the 3.30
I am testing it and seems more stable, no crashes after 3days of operation.

regards
Ros

We just had a crash on 3.30 after some days, it's the first one on over 20 boxes, which is EXTREMELY better than before.
So far 3.30 seems very stable (routing-test) and the box where crashed is quite a little bit loaded (3 full routes ipv4, 3full routes ipv6 and almost 200 peers ipv6/v4)

please post the supout to MT asasp so they will improve the routing.

We are still quite a bit far a way from competitor but they are improving it.

Still see 100% load whenever I change the filter applied to bgp peers and slowness charging and distributing the fullroute.

regards
Ros

please post the supout to MT asasp so they will improve the routing.

Done two minutes after the crash.. and they have quite a collection of our supouts from 3.23 on

We are still quite a bit far a way from competitor but they are improving it.

MT has done some wonderful steps forward since last year (bgp speaking).
I've been using MT bgp for the last four years and (as you should know, if i got who you are ) up to
last year you would have never forced me to use ROS as a full function border router (you should rember
my e-mail when you joined Namex).
Since the RB1000 has come out ROS versions have become much better on BGP (even if not perfect,
but if the redistribution route crash bug is fixed we are not too far from that), at the point that we started replacing
our quagga boxes with MT boxes.. they finally included some missing basic features (like the remove private as), and
have a good ipv6 support.. i really don't miss any "legacy" routers, maybe except for the lack of something like and RB1000 with optical interfaces (SFP would really be nice), and possibly a "legacy" 48V powered RB "something"

Still see 100% load whenever I change the filter applied to bgp peers and slowness charging and distributing the fullroute.

That's not a bug, you will always see the CPU crunching numbers and load up to 100% when it has to update the route table, that's not an issue, as long as it's handled fine by the BGP process and does not impact any router functionality. It may still be a little slow on loading and redistibuting prefixes, but i've never installed ROS on the same machine as a Quagga box to compare, and after all i don't feel it's that slow in processing routes.

From me is a big ++ and thank to MT for the work done on bgp lately..

karyal,

How was the bgp response on legacy ROS 2.9.51. does it crash like version 3 does or is it more stable????


Thanks,

Sotiris

karyal,

How was the bgp response on legacy ROS 2.9.51. does it crash like version 3 does or is it more stable????

I never used ROS 2.9.x as border bgp router, so i can tell you just that for internal bgp we had no issues (using routing-test)
I would suggest you to upgrade to ROS 3.x though, you should be able to upgrade freely to 3.x if you have a 2.9.51
If not, i strongly suggest you to spend the few dollars of license cost to support MT work...

I wonder, where is v3.30 download link?
Thanks..

karyal,

How was the bgp response on legacy ROS 2.9.51. does it crash like version 3 does or is it more stable????


Thanks,

Sotiris

No crashing.

I wonder, where is v3.30 download link?
Thanks..

no link until release =)

v3.30 is out!

Hopefully 3.30 will solve some issues. ANy first impretions or trials of the 3.30 version will be greatly appreciated.

Thanks,

Sotiris

3.30 was best version of MS DOS

Oh RLY What about 5.0 ?

p.s. 3.30 + routing test working fine with one BGP neighbor

Hello everyone any news on 3.30 has anyone been testing the box with more thatn one peers and more than one advertisment ?????

Also has anyone else noticed that the BGP updates on the MK are picked up with a slow pace ?????


Thanks,


Sotiris

Okay.. i dunno if it's a bug but i've serious problems with advertising my networks on 3.30:
I announce a /22 and a /23 to my nexthop peer.. it works..
For some load balancing between all my peers, i'm splitting those /22 in /24 networks.
On 3.24 it works.. on 3.30 it doesnt... my provider reiceves only the /22 and doesn't see the /24..
It's not a filter issue, until yesterday when i upgraded, everything was working good.
To add, those networks i advertise, aren't shown as Active anymore.
I'm sure the /22 is working since some traffic is passing through..

Okay.. i dunno if it's a bug but i've serious problems with advertising my networks on 3.30:
I announce a /22 and a /23 to my nexthop peer.. it works..
For some load balancing between all my peers, i'm splitting those /22 in /24 networks.
On 3.24 it works.. on 3.30 it doesnt... my provider reiceves only the /22 and doesn't see the /24..
It's not a filter issue, until yesterday when i upgraded, everything was working good.
To add, those networks i advertise, aren't shown as Active anymore.
I'm sure the /22 is working since some traffic is passing through..

Are you using a static route to generate the /24 advertisiment? if yes, check that.

No, i'm using a Loopback IP to generate all the /24 (one for each network)
But i have a route for the /22 and /23 that are working.. is it related? SHOULD it be related?

Hello everyone,

Just to provide my results on 3.30 and BGP using a hardware platform from the approaved list on the wiki web page.

Three peers with full routing tables crashed twice in less than 24 hours. It looks like the problem with the ROS is located on the routing module not the redistribution on 3.30.
The more routes you have in a system the faster the routing module crashes. I have send an spout to the support team.

MK you use the BGP for VPLS so having the routing module crashing on BGP is not that good.
Currently I'm looking into other solutions for BGP until MK finds a solution for the BGP not crashing.
Does anyone else has a different experience with the BGP ????


Thanks,

Sotiris

Hello everyone,

Just to provide my results on 3.30 and BGP using a hardware platform from the approaved list on the wiki web page.

Three peers with full routing tables crashed twice in less than 24 hours. It looks like the problem with the ROS is located on the routing module not the redistribution on 3.30.
The more routes you have in a system the faster the routing module crashes. I have send an spout to the support team.

MK you use the BGP for VPLS so having the routing module crashing on BGP is not that good.
Currently I'm looking into other solutions for BGP until MK finds a solution for the BGP not crashing.
Does anyone else has a different experience with the BGP ????


Thanks,

Sotiris

3.30 routing-test on RB1000. Full feed (>294000 Prefixes) and one iBGP-Peer.
Running for 5 days since update.

Stefan

Xeon dual core 3ghz, Full routing table and 1 Ibgp, working since upgrade (3 days)
About the network announciation issue i did the workaround of giving adding a static for those routes.

Hello everyone,

I got a reply from MK support and they have said that the routing module is not updated anymore. All the bugs are fixed in the routing-test packages. I will have a go at that and let you know.

Medianet: Your config is the basic and by adding static routes the BGP module is not actually used for outbound connections it is just used for inbound routes. Also when you peer with other ISP's usually there are two BGP peers which means a minimum of 500K routes in the routing table. That is where the problems start emerging.

Thanks,

Sotiris