Re: [vpp-dev] adding large number of route prefixes

2018-02-01 Thread Konrad Gutkowski 

Hi,

Thanks for the info, that seems to brought 18.01 up to 1M, although it  
stops there.

With heap-size manipulation i can get 17.10 much higher (>3M).
While it seems to be purely academical at this point 18.01 showed  
preference towards sequential prefixes.


For 18.01 there seems to be no difference in maximum amount between 1G -  
4G setting, is that expected?


(only tested it with cli)

Konrad

W dniu 01.02.2018 o 10:11 Neale Ranns (nranns) <nra...@cisco.com> pisze:


Hi Konrad,

As of 18.01 the amount of memory available for IP routes is controlled  
in the startup config with:

  ip {heap-size }

/neale

-Original Message-
From: <vpp-dev-boun...@lists.fd.io> on behalf of Konrad Gutkowski  
<kgutkowski-...@ffs.pl>

Date: Thursday, 1 February 2018 at 03:01
To: vpp-dev <vpp-dev@lists.fd.io>
Subject: [vpp-dev] adding large number of route prefixes

Hi,
   While testing ipv4 routing I found that vpp crashes when supplied with
large sets of prefixes, the exact point when vpp goes down changes
depending on prefix set and vpp version tested.
   Tried 17.10 and 18.01, first consistently crashes at around 1m (though
tested only with sequential /24 and /32 prefixes), for the latter it
depends on the set, for sequential it's around 390k, while public bgp
table does crash pretty randomly around 100-300k, (my file has  
around 750k

routes).
   For 17.10, adding less than 1m prefixes works for both scenarios and
continues working - adding and deleting it multiple times does not  
cause

any stability issues.
It doesn't seem dependent on heap-size parameter value.
   I used both cli "ip route add count" as well as modified router  
plugin to

pull the table from netlink.
Any suggestions are welcome.
   --
Konrad Gutkowski
___
vpp-dev mailing list
vpp-dev@lists.fd.io
https://lists.fd.io/mailman/listinfo/vpp-dev




--
--
Konrad Gutkowski
___
vpp-dev mailing list
vpp-dev@lists.fd.io
https://lists.fd.io/mailman/listinfo/vpp-dev

Re: [vpp-dev] adding large number of route prefixes

2018-02-01 Thread Neale Ranns (nranns) 
Hi Konrad,

As of 18.01 the amount of memory available for IP routes is controlled in the 
startup config with:
  ip {heap-size }

/neale

-Original Message-
From: <vpp-dev-boun...@lists.fd.io> on behalf of Konrad Gutkowski 
<kgutkowski-...@ffs.pl>
Date: Thursday, 1 February 2018 at 03:01
To: vpp-dev <vpp-dev@lists.fd.io>
Subject: [vpp-dev] adding large number of route prefixes

Hi,

While testing ipv4 routing I found that vpp crashes when supplied with  
large sets of prefixes, the exact point when vpp goes down changes  
depending on prefix set and vpp version tested.

Tried 17.10 and 18.01, first consistently crashes at around 1m (though  
tested only with sequential /24 and /32 prefixes), for the latter it  
depends on the set, for sequential it's around 390k, while public bgp  
table does crash pretty randomly around 100-300k, (my file has around 750k  
routes).

For 17.10, adding less than 1m prefixes works for both scenarios and  
continues working - adding and deleting it multiple times does not cause  
any stability issues.
It doesn't seem dependent on heap-size parameter value.

I used both cli "ip route add count" as well as modified router plugin to  
pull the table from netlink.
Any suggestions are welcome.

--
Konrad Gutkowski
___
vpp-dev mailing list
vpp-dev@lists.fd.io
https://lists.fd.io/mailman/listinfo/vpp-dev


___
vpp-dev mailing list
vpp-dev@lists.fd.io
https://lists.fd.io/mailman/listinfo/vpp-dev

[vpp-dev] adding large number of route prefixes

2018-01-31 Thread Konrad Gutkowski 

Hi,

While testing ipv4 routing I found that vpp crashes when supplied with  
large sets of prefixes, the exact point when vpp goes down changes  
depending on prefix set and vpp version tested.


Tried 17.10 and 18.01, first consistently crashes at around 1m (though  
tested only with sequential /24 and /32 prefixes), for the latter it  
depends on the set, for sequential it's around 390k, while public bgp  
table does crash pretty randomly around 100-300k, (my file has around 750k  
routes).


For 17.10, adding less than 1m prefixes works for both scenarios and  
continues working - adding and deleting it multiple times does not cause  
any stability issues.

It doesn't seem dependent on heap-size parameter value.

I used both cli "ip route add count" as well as modified router plugin to  
pull the table from netlink.

Any suggestions are welcome.

--
Konrad Gutkowski
___
vpp-dev mailing list
vpp-dev@lists.fd.io
https://lists.fd.io/mailman/listinfo/vpp-dev