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

List:       openvswitch-discuss
Subject:    [ovs-discuss] prio scheduler
From:       lucabianchi91 () hotmail ! it (Luca Bianchi)
Date:       2015-06-25 22:59:10
Message-ID: DUB127-W7356D3235566B8BED569F3C6AE0 () phx ! gbl
[Download RAW message or body]

Hi,

(sorry for dropping the mailing list, I had troubles with a new mail client)


I think that the problem is in the translation of queue names between OF, ovs and \
linux. 


I explain briefly:


I have 3 iperf client, an ovs switch and an iperf server (linked to switch's eth4) 





h1---link 10M---|-s1:eth1---queue1 or 123---\


h2---link 10M---|-s1:eth2---queue2 or 234---|-qdisc---s1:eth4-|---link 10M---h4


h3---link 10M---|-s1:eth3---queue3 or 345---/





I direct the traffic of each client on a separate queue to see how 
bandwidth sharing reflects scheduler's setting (observed from h4)





---------HTB-------------


In both htb cases, the controller install rules that direct flows to queues 123, 234, \
345.


---case1, ovs commands, working---


With ovs commands I create queue 123, 234, 345 (that in exa correspond to 7b, ea, \
159).


When I check with tc commands, I notice that Linux kernel automatically 
creates an htb qdisc 1:, to which is attached class 1:ffe,to which are 
attached classes 1:7c, 1:eb, 1:15a (notice that it increments the number
 of 1). Everything works.


---case2, tc commands, working---


I can also execute directly tc commands (creating classes ffe,7c,eb,15a)
 and obtain the same htb scheduler, and everything works.


 


(bandwidth is divided as expected, wheter I configure with ovs commands or tc \
commands)





---------PRIO-------------------------


When I use the command "tc qdisc add dev s1-eth4 root handle 1: prio"


linux kernel creates the qdisc prio 1:, to which are attached classes 1:1, 1:2, 1:3.


Using the same logic of htb (linux increments queue id of 1) I thought 
that the openflow rules should use queues 0,1,2. But every flows get 
simultaneously the whole link capacity.


(I tried also with 1,2,3)





I don't know how to catch the queues' name used by Openflow, since if 
classes are created with tc commands (in both cases, htb and prio) 
"ovs-vsctl list qos" and "ovs-vsctl list queue" gives no answer.





Thank you very much



Luca Bianchi

> Date: Thu, 11 Jun 2015 15:21:55 -0700
> From: blp at nicira.com
> To: lucabianchi91 at hotmail.it
> CC: discuss at openvswitch.org
> Subject: Re: [ovs-discuss] prio scheduler
> 
> Please don't drop the mailing list.
> 
> Are you sure Mininet isn't interfering?  It also sets up QoS.
> 
> Are you directing packets to queues?
> 
> ### Q: I configured Quality of Service (QoS) in my OpenFlow network by
> adding records to the QoS and Queue table, but the results aren't
> what I expect.
> 
> A: Did you install OpenFlow flows that use your queues?  This is the
> primary way to tell Open vSwitch which queues you want to use.  If
> you don't do this, then the default queue will be used, which will
> probably not have the effect you want.
> 
> Refer to the previous question for an example.
> 
> On Thu, Jun 11, 2015 at 08:17:09PM +0200, Luca Bianchi wrote:
> > Hi,I am trying to configure a prio qdisc on an OVS switch (all the network is \
> > emulated on Mininet).Since officially OVS does not support this type of qdisc, I \
> > used directly tc commands to configure it.Then, when making packets pass through \
> > the interface with prio qdisc,I noticed this strange behaviour, and I think it's \
> > an interaction problem between OVS and prio.Can you please help me to do this \
> > configuration?Thank youBest Regards Luca Bianchi
> > 
> > > Date: Thu, 11 Jun 2015 10:53:59 -0700
> > > From: blp at nicira.com
> > > To: lucabianchi91 at hotmail.it
> > > CC: bugs at openvswitch.org
> > > Subject: Re: [ovs-discuss] prio scheduler
> > > 
> > > On Fri, Jun 05, 2015 at 07:19:14PM +0200, Luca Bianchi wrote:
> > > > I am trying to configure QoS on OVS, running on mininet network.
> > > > 
> > > > I install a PRIO qdisc on an interface with the command
> > > > 
> > > > tc qdisc add dev s1-eth4 root handle 1: prio
> > > > 
> > > > and then the classes 1:1, 1:2, 1:3 are automatically created.
> > > > 
> > > > The problem is that every user flow passing from that interface now has the \
> > > > whole interface capacity. 
> > > > E.g. 3 user TCP flows pass from this interface (link of capacity 10Mbps), \
> > > > everyone is served with 10Mbps...that is impossible. With htb qdisc (also if \
> > > > configured with tc commands) there are no problems.
> > > 
> > > Are you just reporting how the prio qdisc works, or are you talking
> > > about some interaction between OVS and the prio qdisc?
> > 		 	   		  
 		 	   		  
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://openvswitch.org/pipermail/discuss/attachments/20150626/de15e03d/attachment-0001.html>



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