[prev in list] [next in list] [prev in thread] [next in thread] 

List:       openvswitch-discuss
Subject:    [ovs-discuss] OVS Switch startup time
From:       sukhbir.in () gmail ! com (Sukhbir Singh)
Date:       2014-03-31 23:00:52
Message-ID: 20140331230052.GB22224 () gmail ! com
[Download RAW message or body]

* Ben Pfaff:

> On Mon, Mar 31, 2014 at 06:39:52PM -0400, Sukhbir Singh wrote:
> > * Ben Pfaff:
> > 
> > > On Mon, Mar 31, 2014 at 06:35:14PM -0400, Sukhbir Singh wrote:
> > > > In Mininet, I am using OVSSwitch for creating the switches in a Mininet
> > > > topology. I am using OVS 2.1.0 and there is no controller attached as I
> > > > am setting up the flows manually.
> > > > 
> > > > I have a question about the switch startup time, which gets
> > > > progressively slower as the number of switches increase. As a rough
> > > > estimation, after about 200 switches, it takes 1-1.5 seconds to create a
> > > > switch which is not horrible but I was hoping this would be parallized?
> > > 
> > > How are you doing it?
> > 
> > Nothing fancy, just using addSwitch() [0].
> > 
> > [0] - https://github.com/mininet/mininet/blob/master/mininet/topo.py#L82
> 
> OK, I don't know what that does underneath.  Does it block waiting
> until the operation is complete?  If so then you should be able to get
> better performance by creating several switches in the database in
> parallel and waiting for all of them to be created.

Looking at the code in node.py (which does the actual setup), it does
seem to block. I guess that is where I should be looking at then. 

Out of pure curiosity, does it make sense to try any of the other
options, for example using multiple ovsdb instances?

-- 
Sukhbir

[prev in list] [next in list] [prev in thread] [next in thread]