RARE user and assistance email list

[Alexander Gall <>]

Hi csaba

Very useful, thanks.  The only drawback might be that it's a bit more
tricky to troubleshoot routing issues when the FIB no longer matches
the RIB.  Is there a way to display what optimiziations have been done
explicitly?

Also, while you're at it, did you consider dropping prefixes from the
FIB that have the same next-hop as the default (if one exists) and no
covering prefix?

--
Alex

On Thu, 5 Jan 2023 06:03:19 +0100, mc36 <> said:

> sorry for the spam again but seemingly we're having the stuff in place:
> this night the autosummary (global) knob arrived to all the routing 
> protocols...
> and starting from now, we can do transparent fib compression without 
> altering the rib:

> here is the rib of the v1 vrf:

> r1#show ipv4 route v1
> typ  prefix        metric  iface      hop      time
> C    1.1.1.0/24    0/0     sdn1       null     00:00:28
> LOC  1.1.1.1/32    0/1     sdn1       null     00:00:28
> C    1.1.2.0/24    0/0     sdn2       null     00:00:28
> LOC  1.1.2.1/32    0/1     sdn2       null     00:00:28
> C    1.1.3.0/24    0/0     sdn3       null     00:00:28
> LOC  1.1.3.1/32    0/1     sdn3       null     00:00:28
> C    1.1.4.0/24    0/0     sdn4       null     00:00:28
> LOC  1.1.4.1/32    0/1     sdn4       null     00:00:28
> C    2.2.2.101/32  0/0     loopback0  null     00:00:28
> S    2.2.2.103/32  1/0     sdn1       1.1.1.2  00:00:28
> S    2.2.2.104/32  1/0     sdn2       1.1.2.2  00:00:28
> S    2.2.2.105/32  1/0     sdn3       1.1.3.2  00:00:28
> S    2.2.2.106/32  1/0     sdn4       1.1.4.2  00:00:28
> S    2.2.2.107/32  1/0     sdn4       1.1.4.2  00:00:28

> here is the programmed fib, the 2.2.2.106/32 & 2.2.2.107/32 got 
> autoaggregated to 2.2.2.106/31

> r1#show p4lang p4 done-route4 1
> typ  prefix        metric  iface      hop      time
> C    1.1.1.0/24    0/0     sdn1       null     00:00:41
> LOC  1.1.1.1/32    0/1     sdn1       null     00:00:41
> C    1.1.2.0/24    0/0     sdn2       null     00:00:41
> LOC  1.1.2.1/32    0/1     sdn2       null     00:00:41
> C    1.1.3.0/24    0/0     sdn3       null     00:00:41
> LOC  1.1.3.1/32    0/1     sdn3       null     00:00:41
> C    1.1.4.0/24    0/0     sdn4       null     00:00:41
> LOC  1.1.4.1/32    0/1     sdn4       null     00:00:41
> C    2.2.2.101/32  0/0     loopback0  null     00:00:41
> S    2.2.2.103/32  1/0     sdn1       1.1.1.2  00:00:41
> S    2.2.2.104/32  1/0     sdn2       1.1.2.2  00:00:41
> S    2.2.2.105/32  1/0     sdn3       1.1.3.2  00:00:41
> S    2.2.2.106/31  1/0     sdn4       1.1.4.2  00:00:41

> here you can spot that the dataplane does not report nor byte not packet 
> counters to the 2.2.2.106/32 and 2.2.2.107/32
> simply because these two prefixes get compressed to 2.2.2.106/31, and the 
> dataplane reports to that but that prefix is
> not in the rib, only in the fib:

> r1#show ipv4 counter v1
>                transmit              receive
> prefix        pack     byte         pack  byte  time
> 1.1.1.0/24    10+0     640+0        0+0   0+0   00:00:53
> 1.1.1.1/32    0+8      0+624        0+0   0+0   00:00:53
> 1.1.2.0/24    10+0     640+0        0+0   0+0   00:00:53
> 1.1.2.1/32    0+9      0+702        0+0   0+0   00:00:53
> 1.1.3.0/24    10+0     640+0        0+0   0+0   00:00:53
> 1.1.3.1/32    0+9      0+702        0+0   0+0   00:00:53
> 1.1.4.0/24    10+0     640+0        0+0   0+0   00:00:53
> 1.1.4.1/32    0+9      0+702        0+0   0+0   00:00:53
> 2.2.2.101/32  40+40    2560+3120    0+0   0+0   00:00:53
> 2.2.2.103/32  10+30    640+2340     0+0   0+0   00:00:53
> 2.2.2.104/32  10+30    640+2340     0+0   0+0   00:00:53
> 2.2.2.105/32  10+1141  640+1252215  0+0   0+0   00:00:53
> 2.2.2.106/32  10       640          0     0     00:00:53
> 2.2.2.107/32  0        0            0     0     00:00:53

> and finally the config knobs of the p4lang server:

> r1#show running-config server p4lang p4
> server p4lang p4
>   export-vrf v1
>   export-compress4 v1
>   export-compress6 v1
>   export-port sdn1 1 10 0 0 0
>   export-port sdn2 2 10 0 0 0
>   export-port sdn3 3 10 0 0 0
>   export-port sdn4 4 10 0 0 0
>   interconnect ethernet2
>   vrf v9
>   exit
> !

> r1#

> br,
> cs


> On 1/5/23 05:18, mc36 wrote:
>> sorry for the spam, while added the test cases i found a bug that 
>> prevented proper compression,
>> so the final numbers are 900k-->261k for v4 and 160k-->52k for v6, see 
>> attached...
>> with this boost, we have plenty of headroom for both v4 and v6 in an nren 
>> fib...
>> (we're a tier3 isp mostly default routing to the geant, the only exception 
>> is bix (with
>> 120k prefixes, mostly from he.net) and some private peerings to local 
>> bigger isps....
>> br,
>> cs
>> 
>> On 1/4/23 18:42, mc36 wrote:
>>> not again! i forgot the commit url:
>>> https://github.com/rare-freertr/freeRtr/commit/d76a255b26b2ad3ade8de3a1124e21a1c32c2bcf
>>> the whole algo fits the screen... :)
>>> 
>>> On 1/4/23 18:04, mc36 wrote:
>>>> hi,
>>>> i've just implemented fib compression based on the following two ideas:
>>>> -two consecutive prefixes with the same nexthop...
>>>> -subnets of a supernet with the same nexthop...
>>>> it reduced the niif v6 fib from 160968 to 72298 (44%) and the v4 fib 
>>>> from 901199 to482847 (53%)
>>>> for the 
>>>> https://bitbucket.software.geant.org/projects/RARE/repos/rare/browse/profiles/9.10.0/tofino2/profile-rawip.p4
>>>> profile we have a plenty of headroom for v6 and some headroom for v4...
>>>> for now only show commands are there but nothing holds back from adding 
>>>> auto-summary knob
>>>> to the routing protocols and/or export-compress knob to the p4lang 
>>>> server... or both...:)
>>>> br,
>>>> cs
>>> 
>>> 
>>> 
>>> 
>>> 
>> 
>> 
>> -=-=-=-=-=-=-=-=-=-=-=-
>> Groups.io Links: You receive all messages sent to this group.
>> View/Reply Online (#919): https://groups.io/g/freertr/message/919
>> Mute This Topic: https://groups.io/mt/96066813/6006518
>> Group Owner: 
>> Unsubscribe: https://groups.io/g/freertr/unsub []
>> -=-=-=-=-=-=-=-=-=-=-=-
>> 
>>