Hi Muthukrishnan,

right - I was referring to 'Maximum outbound queue depth'.

25600 is not maximum amount of messages in outbound queue - it is just a 
threshold value and when crossed then automatically a barrier message is added 
to queue. But the weird thing about this is that the limit for barrier in queue 
(25.6k) should be actually smaller than rpc limit (20k). So if you are 
observing one switch throughput then I believe that by increasing rpc-limit to 
30k or decreasing the barrier in queue limit to 15k will improve your 
performance.


See:

openFlowPlugin/openflowplugin-api/src/main/yang/openflow-provider-config.yang

openFlowPlugin/openflowplugin-blueprint-config/src/main/resources/initial/openflowplugin.cfg

rpc-requests-quota


Regards,

Michal



________________________________
From: Muthukrishnan Thangasamy <muthukrishna...@tataelxsi.co.in>
Sent: Wednesday, February 22, 2017 08:35
To: Michal Rehak -X (mirehak - PANTHEON TECHNOLOGIES at Cisco); 
openflowplugin-dev@lists.opendaylight.org
Cc: openflowjava-...@lists.opendaylight.org
Subject: Re: [openflowjava-dev] Barrier Message Timeout - 500ms - Rational 
behind the value


Hi Michal ,


Thank you for your response , I agree with you , I need some clarity on your 
response


Also there is a complementary parameter - amount of messages in outbound queue 
without barrier inside. The more messages kept in queue the higher throughput 
you shall get.


Question : My understanding is that you are referring to Maximum outbound queue 
depth which is 25600 by default , here if we are sending the flow via SAL-RPC  
and barrier timer is 500 ms

how to achieve flow throughput of 1000 flows/sec ( total number of non-barrier 
message kept in queue as 1000) , since RPC call form northbound takes ( 500ms + 
20 ms ) for writing single flow (assuming writing each flow sequentially in 
loop), throughput is less even you are sending 1000 flows/sec. If am sending 
via RPC how to keep 1000 openflow messages in queue without barrier ? is there 
any other parameter which is controlling the same ?



Regards

Muthukrishnan

9952012433






________________________________
From: Michal Rehak -X (mirehak - PANTHEON TECHNOLOGIES at Cisco) 
<mire...@cisco.com>
Sent: Thursday, February 16, 2017 2:21 PM
To: Muthukrishnan Thangasamy; openflowplugin-dev@lists.opendaylight.org
Cc: openflowjava-...@lists.opendaylight.org
Subject: Re: [openflowjava-dev] Barrier Message Timeout - 500ms - Rational 
behind the value


Hi Muthukrishnan,

there is no golden standard behind this value. Setting it to 100ms or even 
lower value will result into low latency system but this also decreases 
throughput of system.


If you have few switches and little no big requirement on flows per second 
throughput then 100ms might suit better to your setup. On the other hand if 
there are hundreds of devices and throughput of thousands of flows per second 
required then 500ms or even higher will improve overall performance.


Also there is a complementary parameter - amount of messages in outbound queue 
without barrier inside. The more messages kept in queue the higher throughput 
you shall get. But there is memory cost and also by extending size of those 
queues (per device) there comes overload risk of md-sal notification queue.


True - this value might be configurable. But that would solve the situation 
only for static requirements.



Regards,

Michal



________________________________
From: openflowjava-dev-boun...@lists.opendaylight.org 
<openflowjava-dev-boun...@lists.opendaylight.org> on behalf of Muthukrishnan 
Thangasamy <muthukrishna...@tataelxsi.co.in>
Sent: Wednesday, February 15, 2017 08:13
To: openflowplugin-dev@lists.opendaylight.org
Cc: openflowjava-...@lists.opendaylight.org
Subject: [openflowjava-dev] Barrier Message Timeout - 500ms - Rational behind 
the value



Dear Team ,


I have been working in Lithium and Boron Stable Version .


Lets take Boron , In openflowplugin.cfg default timer for barrier message is 
500ms is there any

rational behind the same ? why not 100ms ?


If I push via SAL-FLOW RPC , total time taken for flow push = Time taken for 
flow push + barrier delay  = ~ 30 ms + 520 ms (barrier ).


If I change the value from 500ms to 100 ms I am getting drastic performance 
improvement in flow push processing . Can anyone tell why we have gone for 
500ms as default timer value ?

How we have arrived this golden value of 500ms ?



Reference.


File : openflowplugin.cfg

"#Timeout interval in milliseconds between each barrier message.

#Default value is set to 500 milliseconds
barrier-interval-timeout-limit=500"

Regards
Muthukrishnan
9952012433

_______________________________________________
openflowplugin-dev mailing list
openflowplugin-dev@lists.opendaylight.org
https://lists.opendaylight.org/mailman/listinfo/openflowplugin-dev
  • [openflowplugin... Muthukrishnan Thangasamy
    • Re: [openf... Michal Rehak -X (mirehak - PANTHEON TECHNOLOGIES at Cisco)
      • Re: [o... Muthukrishnan Thangasamy
        • Re... Michal Rehak -X (mirehak - PANTHEON TECHNOLOGIES at Cisco)

Reply via email to