HI,
l2_multi uses of.ofp_match.from_packet(packet), to get the match for the
folw table entry. This will be an exact match right. Now i have to add one
more flow entry(for backup path) for the same kind of packets. Hence i used
the same match and different priority. But as u have already mentioned,
this priority is defaulted to highest priority as it is an exact match and
i am getting an error as this entry is duplicate of working path entry.
What is the solution for this? can i make the backup entry with
wildcards? If so which fields? and how to set wildcards in match?
Or can v change the behavior of openflow protocol and make it to alow the
exact match entry with different priorities. In that case where should i
make the change? when i go thru libopenflow_01 , i couldn't find anything
which overwrites priority?
Please provide some input.
Thanks and Regards,
Padma V
On Mon, Sep 8, 2014 at 2:39 AM, Murphy McCauley <murphy.mccau...@gmail.com>
wrote:
> Why use exact matches?
>
> P.S., please consider not dropping the list.
>
> On Sep 7, 2014, at 3:23 AM, Padma Jayasankar <padmaj...@gmail.com> wrote:
>
> Hi,
> Thanks for the clarification. Mine is exact match only.
> But to enable backup path installation i have to enter two flow entries
> for the same match with different priorities.
> Is this possible? Hope i have to change the switch part for it.Or is it
> not possible ?
>
> Thanks and Regards,
> Padma V
>
>
>
> On Sun, Sep 7, 2014 at 12:06 PM, Murphy McCauley <
> murphy.mccau...@gmail.com> wrote:
>
>> On Sep 6, 2014, at 11:19 PM, Padma Jayasankar <padmaj...@gmail.com>
>> wrote:
>>
>> Hi,
>> I am trying to modify forwarding.l2_multi for backup path computation.
>> For this as a first step , i wanted to put some flow entries with the
>> priority 0x0001. But though i am setting the msg.priority to this value ,
>> the flows which are installed in the switch have the priority 65535.
>> Ia m setting the msg.priority of ofp_flow_mod message before
>> switch.connection.send(msg) statment
>>
>> def _binstall (self, switch, in_port, out_port, match, buf = None):
>> out_port_temp =3
>> msg = of.ofp_flow_mod()
>> msg.match = match
>> msg.match.in_port = in_port
>> msg.idle_timeout = FLOW_IDLE_TIMEOUT
>> msg.hard_timeout = FLOW_HARD_TIMEOUT
>> msg.priority = BACKUP_PRIORITY
>> log.debug("Installing backup path with priority %s",msg.priority)
>> msg.actions.append(of.ofp_action_output(port = out_port))
>> msg.buffer_id = buf
>> switch.connection.send(msg)
>>
>> 1)The log.debug statement shows the priority as 1. But the flow entries
>> are with the priority 65535.
>> What is the reason for it?
>>
>>
>> From Section 3.4 of the OpenFlow 1.0 specification:
>> Packets are matched against flow entries based on prioritization. An
>> entry that
>> specifies an exact match (i.e., it has no wildcards) is always the
>> highest priority.
>> All wildcard entries have a priority associated with them. Higher
>> priority entries
>> must match before lower priority ones.
>>
>> So... is your match an exact match?
>>
>> 2) What will be the effect of switch.connection.send(msg).
>> (ie) After the execution of this statement where will the control go.I
>> guess the control goes to the switch. If so which module of openvswitch
>> will handle this?
>>
>>
>> The effect is that it takes msg, serializes it into an OpenFlow command,
>> and tells the OS to send it out the TCP connection to the switch. Then
>> control returns to the statement immediately following the call to send().
>>
>> Elsewhere, the command goes through the OS's TCP stack, is sent over a
>> link, is received by the switch, and the switch processes it. I wouldn't
>> call any of that a transfer of control, though. It's all asynchronous.
>> And it's all outside the scope of POX. If you want to understand how OVS
>> works, you might try reading the OVS code or searching the OVS mailing
>> lists.
>>
>>
>> Good luck!
>>
>> -- Murphy
>>
>> On Sat, Sep 6, 2014 at 8:07 AM, Padma Jayasankar <padmaj...@gmail.com>
>> wrote:
>>
>>> Hi,
>>> I waited for complete discovery cycle..even though it is not working
>>> for the single,2 topology. But it is working for the custom topology which
>>> i created with loop.
>>>
>>> Regarding openvswitch, i followed the below instructions to install it.
>>>
>>> root@mininet-vm# cd /root
>>> root@mininet-vm# wget
>>> http://openvswitch.org/releases/openvswitch-1.10.0.tar.gz
>>> root@mininet-vm# tar zxvf openvswitch-1.10.0.tar.gz
>>>
>>> Build debian packages and install
>>> ---------------------------------
>>> root@mininet-vm# cd openvswitch-1.10.0/
>>> root@mininet-vm# apt-get install build-essential fakeroot
>>> root@mininet-vm# apt-get install debhelper autoconf automake libssl-dev
>>> pkg-config bzip2 openssl python-all procps python-qt4 python-zopeinterface
>>> python-twisted-conch
>>> root@mininet-vm# fakeroot debian/rules binary
>>> root@mininet-vm# cd ..
>>> root@mininet-vm# dpkg -i openvswitch-common*.deb
>>> openvswitch-datapath-dkms*.deb openvswitch-controller*.deb
>>> openvswitch-pki*.deb openvswitch-switch*.deb
>>>
>>> But before following debian package install mehods, i tried to install
>>> with -
>>>
>>> *Binary build and Install*
>>> ------------------------------------
>>>
>>> root@mininet-vm# cd openvswitch-1.10.0/
>>> root@mininet-vm# ./configure --prefix=/usr --with-linux=/lib/modules/`uname
>>> -r`/build
>>> root@mininet-vm# make
>>> root@mininet-vm# make install
>>> root@mininet-vm# make modules_install
>>> root@mininet-vm# rmmod openvswitch
>>> root@mininet-vm# depmod -a
>>>
>>> But i got some error while executing make..
>>>
>>> *My doubt is*
>>> 1. with debian packages install method, will i be able to change the source
>>> code
>>> and build again.
>>>
>>> 2. As per this installation guidelines openvswitch1.10.0 folder is created
>>> under root folder
>>> and i am not able to see this folder contents(permission denied).. So how
>>> to change the
>>> source code of openvswitch
>>>
>>>
>>>
>>>
>>>
>>>
>>> On Sat, Sep 6, 2014 at 2:33 AM, Murphy McCauley <
>>> murphy.mccau...@gmail.com> wrote:
>>>
>>>>
>>>> On Sep 5, 2014, at 11:08 AM, Padma Jayasankar <padmaj...@gmail.com>
>>>> wrote:
>>>>
>>>> Hi,
>>>> 1)I have removed ovs that come along with Mininet and installed
>>>> openvswitch 1.10.0. After this i tested mininet, with the following command
>>>> sudo mn --topo single,2 --switch ovsk --controller
>>>> remote,ip=192.168.56.1,port=6633
>>>> its working and pingall runs successfully.
>>>>
>>>> Does this sufficient to check whether the installed openvswitch works
>>>> fine? Please clarify.
>>>> Here i checked with POX controller with following 3 commands.
>>>>
>>>>
>>>> Probably, though I am not an authoritative source on the subject.
>>>>
>>>> ./pox.py log.level forwarding.l2_learning
>>>> ./pox.py log.level openflow.discovery forwarding.l2_multi
>>>> ./pox.py log.level openflow.discovery openflow.spanning_tree
>>>> forwarding.l2_multi
>>>>
>>>> For all these pingall test was successful
>>>>
>>>> 2) But when i run the pox controller with the below options
>>>> ./pox.py log.level openflow.discovery openflow.spanning_tree
>>>> --no-flood --hold-down forwarding.l2_multi
>>>> the ping test fails for the sudo mn command given above.
>>>>
>>>> But for another topology with loop, pingall test was successful
>>>> sudo mn --custom ~/mininet/custom/topo-3sw-2host.py --topo mytopo
>>>> --controller remote,ip=192.168.56.1,port=6633
>>>>
>>>> i have attached the custom topology file.
>>>>
>>>> Why this happens?
>>>>
>>>>
>>>> Did you wait a while before running the ping test? With no-flood and
>>>> hold-down, you have to wait for a complete discovery cycle before you can
>>>> expect things to work.
>>>>
>>>> Thanks,
>>>> Padma V
>>>>
>>>>
>>>>
>>>> On Thu, Sep 4, 2014 at 10:54 AM, Murphy McCauley <
>>>> murphy.mccau...@gmail.com> wrote:
>>>>
>>>>> Many features of OVS are available via specific OVS extensions to
>>>>> OpenFlow 1.0. They may also be available as features in higher OpenFlow
>>>>> versions.
>>>>>
>>>>> POX currently only supports OpenFlow 1.0, but it does support a number
>>>>> of OVS extensions (and others can be added).
>>>>>
>>>>> As far as I know, there are no plans to have OVS stop supporting any
>>>>> lower versions of OpenFlow. Certainly, the latest versions still support
>>>>> 1.0.
>>>>>
>>>>> -- Murphy
>>>>>
>>>>> On Sep 3, 2014, at 10:17 PM, Padma Jayasankar <padmaj...@gmail.com>
>>>>> wrote:
>>>>>
>>>>> HI,
>>>>> Regarding OVS extensions to support back up path, do u mean the fast
>>>>> fail over group type..
>>>>> If that is the case then openflow controller(i am using POX), have
>>>>> to support atleast OpenFlow 1.1 know..But it supports only OpenFlow 1.0
>>>>> right.
>>>>> Further, can i use the new OVS versions to support OpenFlow 1.0. Or
>>>>> will they support only OpenFlow 1.3.
>>>>>
>>>>> Thanks ,
>>>>> Padma V
>>>>>
>>>>>
>>>>> On Thu, Sep 4, 2014 at 10:27 AM, Murphy McCauley <
>>>>> murphy.mccau...@gmail.com> wrote:
>>>>>
>>>>>> Right, you'd have to build your own version. I think you have the
>>>>>> right idea of asking on the mininet list for more guidance here.
>>>>>>
>>>>>> Though again... OVS has lots of extensions already for doing
>>>>>> failover/backup-paths (even without controller involvement), which are
>>>>>> already used extensively. I would not be surprised if it already had all
>>>>>> the capabilities you need.
>>>>>>
>>>>>> Good luck.
>>>>>>
>>>>>> -- Murphy
>>>>>>
>>>>>> On Sep 3, 2014, at 6:56 AM, Padma Jayasankar <padmaj...@gmail.com>
>>>>>> wrote:
>>>>>>
>>>>>> Hi,
>>>>>> If i have to modify the openvswitch,then do i have to remove the
>>>>>> one which is installed by mininet and download and install openvswitch.
>>>>>>
>>>>>> Thanks & Regards,
>>>>>> Padma V
>>>>>>
>>>>>>
>>>>>> On Fri, Aug 29, 2014 at 9:51 AM, Murphy McCauley <
>>>>>> murphy.mccau...@gmail.com> wrote:
>>>>>>
>>>>>>> On Aug 28, 2014, at 1:12 AM, Padma Jayasankar <padmaj...@gmail.com>
>>>>>>> wrote:
>>>>>>>
>>>>>>> Thanks a lot.
>>>>>>> I am able to see the flow tables of other switches by changing the
>>>>>>> port like 6635,6637..
>>>>>>> One more doubt.
>>>>>>> * If i have to add a new type of openflow message then where should
>>>>>>> i do it.*
>>>>>>> In the server where i have installed minnet, Openflow folder is
>>>>>>> also there.
>>>>>>> I am able to find Openflow.h file in openflow/include/openflow
>>>>>>> folder. This file has the declaration and structure definition of all
>>>>>>> openflow messages like port_status,Packet_in etc. So here i have to add
>>>>>>> the
>>>>>>> new message name and format i guess..
>>>>>>>
>>>>>>>
>>>>>>> Probably not. By default, Mininet uses Open vSwitch. So you'll
>>>>>>> want to modify Open vSwitch. Of course, Open vSwitch already contains
>>>>>>> features for backup paths...
>>>>>>>
>>>>>>> But the controller also will be using openflow protocol to listen
>>>>>>> to switch..So in the server in which i ghave installed the controller,
>>>>>>> where will it be..i believe, i have to make changes in that part too..
>>>>>>> is
>>>>>>> it correct?
>>>>>>>
>>>>>>>
>>>>>>> Sure. In POX, this is mostly in libopenflow_01.py. You can look at
>>>>>>> nicira.py also to see how some extensions have been added. (It's not
>>>>>>> currently as clean/modular as would be desirable... cleaning it up will
>>>>>>> hopefully happen in a later POX version.)
>>>>>>>
>>>>>>> On Tue, Aug 26, 2014 at 11:57 PM, Murphy McCauley <
>>>>>>> murphy.mccau...@gmail.com> wrote:
>>>>>>>
>>>>>>>>
>>>>>>>> On Aug 26, 2014, at 9:40 AM, Padma Jayasankar <padmaj...@gmail.com>
>>>>>>>> wrote:
>>>>>>>>
>>>>>>>> HI,
>>>>>>>>
>>>>>>>> 1. You might use l2_multi as a starting point, but you might want
>>>>>>>> to use something like NetworkX to base your path computation upon.
>>>>>>>> *what is this NetworkX..please explain*
>>>>>>>>
>>>>>>>>
>>>>>>>> It's a graph library for Python which you can learn a lot about by
>>>>>>>> doing a web search for "NetworkX".
>>>>>>>>
>>>>>>>> In the below command 'switch ' is specified as tcp:127.0.0.1:6634.
>>>>>>>>
>>>>>>>> *dpctl add-flow tcp:127.0.0.1:6634 <http://127.0.0.1:6634/>
>>>>>>>> in_port=1,actions=output:2*
>>>>>>>>
>>>>>>>> How is this mapped to a particular switch. In my topology if i have
>>>>>>>> 3 switches s3,s4,s5 then how to specify one of these switches in the
>>>>>>>> command. Please clarify.
>>>>>>>>
>>>>>>>>
>>>>>>>> This is a Mininet question not a POX one. Furthermore, as I
>>>>>>>> already mentioned, you are probably better off using the OVS tools
>>>>>>>> (ovs-dpctl, ovs-vsctl, ovs-ofctl, etc.). Among other things, you can
>>>>>>>> usually specify a datapath by name when using the OVS tools (e.g.,
>>>>>>>> ovs-vsctl s3 ...). That all said, I believe each switch in Mininet
>>>>>>>> listens
>>>>>>>> on a different port. So maybe port 6634 is s3, and 6635 is s4, etc.
>>>>>>>>
>>>>>>>> Good luck.
>>>>>>>>
>>>>>>>> -- Murphy
>>>>>>>>
>>>>>>>> On Tue, Aug 26, 2014 at 5:27 AM, Murphy McCauley <
>>>>>>>> murphy.mccau...@gmail.com> wrote:
>>>>>>>>
>>>>>>>>> On Aug 25, 2014, at 7:25 AM, Padma Jayasankar <padmaj...@gmail.com>
>>>>>>>>> wrote:
>>>>>>>>>
>>>>>>>>> HI,
>>>>>>>>> I feel, it will be better if i explain my requirement clearly. My
>>>>>>>>> aim is to reduce the failure recovery time. For this, i am trying to
>>>>>>>>> avoid
>>>>>>>>> controller intervention at the time of failure. For that,in response
>>>>>>>>> to
>>>>>>>>> traffic the controller should find primary and backup path and
>>>>>>>>> install it
>>>>>>>>> in the switches.Back up path will be installed with a different(lower)
>>>>>>>>> priority.
>>>>>>>>> When a failure occurs, the switches connected to the failed link
>>>>>>>>> should do the following.
>>>>>>>>> 1) Delete the flow entries that use the failed link(in/out
>>>>>>>>> port)
>>>>>>>>> By doing this, now the traffic will be redirected through the
>>>>>>>>> backup path automatically as the switch has the backup path entries
>>>>>>>>> (which
>>>>>>>>> is installed at the beginning) itself.
>>>>>>>>> 2) Renew this backup path entries:
>>>>>>>>> When there is no failure, as the switches use the working path
>>>>>>>>> only, the back up paths will be removed automatically. But this
>>>>>>>>> should not
>>>>>>>>> happen. Hence the switches have to send some packets through this path
>>>>>>>>> periodically to avoid expiration. These packets should be different
>>>>>>>>> from
>>>>>>>>> normal data packets. I am referring these packets as renewal packets.
>>>>>>>>> These
>>>>>>>>> packets has to be identified uniquely, so that egress switches doesn't
>>>>>>>>> deliver them to hosts.
>>>>>>>>>
>>>>>>>>> 3) Restoration of working path:
>>>>>>>>> Once a link fails, the switches starts using the back up path.
>>>>>>>>> But these backup paths are less efficient than primary path. Hence
>>>>>>>>> once the
>>>>>>>>> failure is rectified and the link is up, again we should reroute the
>>>>>>>>> traffic through the primary path.
>>>>>>>>> But as the switches has backup entries, it won't direct the
>>>>>>>>> packets to the controller. They will route the traffic through backup
>>>>>>>>> path
>>>>>>>>> only.hence when a link is up, the corresponding switch has to send
>>>>>>>>> restoration packets(a special type of packet) to the controller.Then
>>>>>>>>> the
>>>>>>>>> controller will recompute the path for that flow and install the
>>>>>>>>> primary
>>>>>>>>> and backup entries.
>>>>>>>>>
>>>>>>>>> Now in implementation perspective, please clarify the below points.
>>>>>>>>> 1. For the (1) requirement can i use, l2_multi with changes to
>>>>>>>>> compute backup path.
>>>>>>>>> 2. How to make the switches to generate renewal packets..where
>>>>>>>>> should i place the code for this??
>>>>>>>>>
>>>>>>>>>
>>>>>>>>> 1. You might use l2_multi as a starting point, but you might want
>>>>>>>>> to use something like NetworkX to base your path computation upon.
>>>>>>>>>
>>>>>>>>> 2. Assuming you want to do it from the controller, maybe you want
>>>>>>>>> to use callDelayed() or a Timer (see the POX manual) to generate a
>>>>>>>>> packet-out message to send to a switch.
>>>>>>>>>
>>>>>>>>> Beyond that, I'm not sure if I'll be much help to you. Good luck!
>>>>>>>>>
>>>>>>>>> -- Murphy
>>>>>>>>>
>>>>>>>>> On Mon, Aug 25, 2014 at 10:12 AM, Murphy McCauley <
>>>>>>>>> murphy.mccau...@gmail.com> wrote:
>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>> On Aug 24, 2014, at 9:08 PM, Padma Jayasankar <
>>>>>>>>>> padmaj...@gmail.com> wrote:
>>>>>>>>>>
>>>>>>>>>> Hi,
>>>>>>>>>> In my previous mails, i asked two things
>>>>>>>>>> 1) I want to install back up paths proactively. According to my
>>>>>>>>>> understanding l2_multi does this reactively(once failure occurs,it
>>>>>>>>>> finds
>>>>>>>>>> the new path). Did u mean the same abt l2_multi.
>>>>>>>>>> *2)ofp_flow_mod_flags* of *ofp_flow_mod* has to be set to
>>>>>>>>>> * OFPFF_SEND_FLOW_REM* to invoke flow removal notification
>>>>>>>>>> messages.
>>>>>>>>>> Is it correct? please confirm
>>>>>>>>>>
>>>>>>>>>> *Does your 'Yes' holds for both*.
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>> My yes was just for the second one.
>>>>>>>>>>
>>>>>>>>>> For the first one... l2_multi is reactive. It doesn't create any
>>>>>>>>>> switch state except in response to traffic. In general, this is a
>>>>>>>>>> bad
>>>>>>>>>> design. topo_proactive is a somewhat more reasonable variation.
>>>>>>>>>> But you
>>>>>>>>>> seem to be talking about creating both primary and backup paths in
>>>>>>>>>> response
>>>>>>>>>> to traffic. I'd say this is still reactive. I mean... I guess you
>>>>>>>>>> could
>>>>>>>>>> install just the primary and then install the backup in response to
>>>>>>>>>> failure, but in that case I wouldn't call it a backup -- it's just a
>>>>>>>>>> new
>>>>>>>>>> path.
>>>>>>>>>>
>>>>>>>>>> Further regarding back up path renewal,
>>>>>>>>>> - For fast recovery, i have to enable the switches to delete the
>>>>>>>>>> working path entries , when the corresponding link fails(though
>>>>>>>>>> timeout
>>>>>>>>>> constraint not met). Hence i can not make the switches to delete the
>>>>>>>>>> backup
>>>>>>>>>> paths on flow removal notification.
>>>>>>>>>> Hence i have to use renewal packets.In which part this has to
>>>>>>>>>> be done? It will be helpful, if i get a starting point.
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>> I'm not sure I know what you mean by renewal packets. I was
>>>>>>>>>> assuming you just meant packets that kept the flows from expiring.
>>>>>>>>>>
>>>>>>>>>> Regardless, I believe the way I mentioned still works. Yes, you
>>>>>>>>>> delete them manually when you get a flow removal message. This is
>>>>>>>>>> in order
>>>>>>>>>> to "clean them up". This has nothing to do with how you react on
>>>>>>>>>> failure;
>>>>>>>>>> this has to do with removing useless entries.
>>>>>>>>>>
>>>>>>>>>> I don't actually know what mechanism you plan to use to detect a
>>>>>>>>>> failure and switch to an alternate path, but I think it probably
>>>>>>>>>> diverges
>>>>>>>>>> heavily from how I'd do it, so I'm not sure how helpful I can be.
>>>>>>>>>>
>>>>>>>>>> Thanks and Regards,
>>>>>>>>>> Padma V
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>> On Mon, Aug 25, 2014 at 2:27 AM, Murphy McCauley <
>>>>>>>>>> murphy.mccau...@gmail.com> wrote:
>>>>>>>>>>
>>>>>>>>>>> Yes.
>>>>>>>>>>>
>>>>>>>>>>> On Aug 24, 2014, at 7:15 AM, Padma Jayasankar <
>>>>>>>>>>> padmaj...@gmail.com> wrote:
>>>>>>>>>>>
>>>>>>>>>>> Hi,
>>>>>>>>>>> 2)When i checked the openflow specification , came to know
>>>>>>>>>>> that
>>>>>>>>>>> *ofp_flow_mod_flags* of *ofp_flow_mod* has to be set to
>>>>>>>>>>> * OFPFF_SEND_FLOW_REM* to invoke flow removal notification
>>>>>>>>>>> messages.
>>>>>>>>>>> Is it correct? please confirm
>>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>> On Sun, Aug 24, 2014 at 7:42 PM, Padma Jayasankar <
>>>>>>>>>>> padmaj...@gmail.com> wrote:
>>>>>>>>>>>
>>>>>>>>>>>> Hi,
>>>>>>>>>>>> 2)When i checked the openflow specification , came to know
>>>>>>>>>>>> that
>>>>>>>>>>>> ofp_flow_mod_flags
>>>>>>>>>>>>
>>>>>>>>>>>>
>>>>>>>>>>>> On Sun, Aug 24, 2014 at 6:25 PM, Padma Jayasankar <
>>>>>>>>>>>> padmaj...@gmail.com> wrote:
>>>>>>>>>>>>
>>>>>>>>>>>>> HI,
>>>>>>>>>>>>>
>>>>>>>>>>>>> Thanks for quick reply.
>>>>>>>>>>>>>
>>>>>>>>>>>>> 1) I want to install back up paths pro actively. According to
>>>>>>>>>>>>> my understanding l2_multi does this reactively(once failure
>>>>>>>>>>>>> occurs,it finds
>>>>>>>>>>>>> the new path). Did u mean the same abt l2_multi.
>>>>>>>>>>>>>
>>>>>>>>>>>>> 2) I will implement back up path entries as per ur guidence.
>>>>>>>>>>>>> But to install flow entries with flow removal notification, what
>>>>>>>>>>>>> should be
>>>>>>>>>>>>> done? please provide a starting point.
>>>>>>>>>>>>>
>>>>>>>>>>>>> Tahnks and Regards,
>>>>>>>>>>>>> Padma V
>>>>>>>>>>>>>
>>>>>>>>>>>>>
>>>>>>>>>>>>>
>>>>>>>>>>>>> On Sun, Aug 24, 2014 at 3:40 PM, Murphy McCauley <
>>>>>>>>>>>>> murphy.mccau...@gmail.com> wrote:
>>>>>>>>>>>>>
>>>>>>>>>>>>>> On Aug 24, 2014, at 2:54 AM, Padma Jayasankar <
>>>>>>>>>>>>>> padmaj...@gmail.com> wrote:
>>>>>>>>>>>>>>
>>>>>>>>>>>>>> Hi,
>>>>>>>>>>>>>> Thanks for the clarifications..
>>>>>>>>>>>>>> 1) For the first query, as u have said the iCMP type differs
>>>>>>>>>>>>>> in the entries.
>>>>>>>>>>>>>> 2) Regarding the back up path, i can not install them without
>>>>>>>>>>>>>> a timeout..Let me explain my requirement.
>>>>>>>>>>>>>> My aim is to improve the recovery time.
>>>>>>>>>>>>>> Hence i have to install backup paths along with working
>>>>>>>>>>>>>> paths(both have to be computed and installed).
>>>>>>>>>>>>>> when a link fails, the corresponding switch has to remove
>>>>>>>>>>>>>> the corresponding entry, involving that link.
>>>>>>>>>>>>>> As back up flow is already installed, the switch will use
>>>>>>>>>>>>>> that entry immediately.
>>>>>>>>>>>>>> But though we install backup path entries in the switch,
>>>>>>>>>>>>>> when there is no failure , it won't be used Hence because of
>>>>>>>>>>>>>> idle time out,
>>>>>>>>>>>>>> they will get expired.If we install it as permenant entries ,
>>>>>>>>>>>>>> then this
>>>>>>>>>>>>>> will result in large no of flow entries in the flow tables(in
>>>>>>>>>>>>>> largel
>>>>>>>>>>>>>> networks).
>>>>>>>>>>>>>> Hence ,the switches has to generate some renew packets(of
>>>>>>>>>>>>>> small size), and send it through the back up path
>>>>>>>>>>>>>> periodically(to avoid
>>>>>>>>>>>>>> expiration). How to accomplish this.
>>>>>>>>>>>>>>
>>>>>>>>>>>>>>
>>>>>>>>>>>>>> Are you writing this reactively and using fine-grained table
>>>>>>>>>>>>>> entries? Despite the fact that this is how l2_multi, for
>>>>>>>>>>>>>> example, works...
>>>>>>>>>>>>>> I strongly caution against this. This introduces scaling
>>>>>>>>>>>>>> problems which are
>>>>>>>>>>>>>> generally avoidable.
>>>>>>>>>>>>>>
>>>>>>>>>>>>>> More to the point, though... I can think of several ways of
>>>>>>>>>>>>>> keeping the backup paths for active flows. One way (which
>>>>>>>>>>>>>> doesn't involve
>>>>>>>>>>>>>> any "renew packets" is... Install the primary table entries with
>>>>>>>>>>>>>> timeouts
>>>>>>>>>>>>>> and with flow-removal notification. Install the backup entries
>>>>>>>>>>>>>> with no
>>>>>>>>>>>>>> timeouts. If you get an expiration notification from the
>>>>>>>>>>>>>> primary path,
>>>>>>>>>>>>>> remove the backup path (unless you've failed over and are using
>>>>>>>>>>>>>> the backup
>>>>>>>>>>>>>> path; in this case, reinstall the backup path with expirations).
>>>>>>>>>>>>>>
>>>>>>>>>>>>>> 3) To start with, i created a custom topology, with 3
>>>>>>>>>>>>>> switches and 2 hosts in mininet. i wrote a python script to
>>>>>>>>>>>>>> start the
>>>>>>>>>>>>>> network and test pingall. I have attached the python script.
>>>>>>>>>>>>>> I tried to connect that with remote POX controller
>>>>>>>>>>>>>> ./pox.py log.level openflow.discovery openflow.spanning_tree
>>>>>>>>>>>>>> --no-flood --hold-down forwarding.l2_multi
>>>>>>>>>>>>>>
>>>>>>>>>>>>>> ping was successful.
>>>>>>>>>>>>>> But i want to down a link and test the behaviour of ping
>>>>>>>>>>>>>> after the link down. How to include it in the python script.
>>>>>>>>>>>>>> i know the mininet command
>>>>>>>>>>>>>> h1 ping -c10 h2
>>>>>>>>>>>>>> link s3 s4 down
>>>>>>>>>>>>>> h1 ping -c10 h2
>>>>>>>>>>>>>> link s3 s4 down
>>>>>>>>>>>>>> h1 ping -c10 h2
>>>>>>>>>>>>>> But how to invoke this in the script. Is there any way? Then
>>>>>>>>>>>>>> only i can identify that how much time the controller takes to
>>>>>>>>>>>>>> recover from
>>>>>>>>>>>>>> failure.
>>>>>>>>>>>>>>
>>>>>>>>>>>>>>
>>>>>>>>>>>>>> This is a Mininet concern, and you should look into the
>>>>>>>>>>>>>> Mininet resources (the FAQ, mailing list archive, etc.).
>>>>>>>>>>>>>>
>>>>>>>>>>>>>> Sorry, if it is a very silly doubt..I am new to SDN,
>>>>>>>>>>>>>> Python,mininet and POX and i have a very little time to finish
>>>>>>>>>>>>>> my project.
>>>>>>>>>>>>>>
>>>>>>>>>>>>>>
>>>>>>>>>>>>>>
>>>>>>>>>>>>>>
>>>>>>>>>>>>>>
>>>>>>>>>>>>>> On Sun, Aug 24, 2014 at 4:05 AM, Murphy McCauley <
>>>>>>>>>>>>>> murphy.mccau...@gmail.com> wrote:
>>>>>>>>>>>>>>
>>>>>>>>>>>>>>> On Aug 23, 2014, at 5:42 AM, Padma Jayasankar <
>>>>>>>>>>>>>>> padmaj...@gmail.com> wrote:
>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>> Hi,
>>>>>>>>>>>>>>> I am new to POX and Openflow. In my project.I have to
>>>>>>>>>>>>>>> compute working path and backup path and install them in
>>>>>>>>>>>>>>> switches.I am
>>>>>>>>>>>>>>> going to use Mininet and POX..Please guide me where to start
>>>>>>>>>>>>>>> with..
>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>> In POX, i tried with
>>>>>>>>>>>>>>> python ./pox.py openflow.discovery forwarding.l2_multi
>>>>>>>>>>>>>>> and in Mininet,
>>>>>>>>>>>>>>> sudo mn --topo single,2 --controller
>>>>>>>>>>>>>>> remote,ip=192.168.1.4,6633
>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>> The topology was built successfully.
>>>>>>>>>>>>>>> 1)I wanted to check the flow entries in the flow table of
>>>>>>>>>>>>>>> switch s1.
>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>> i executed
>>>>>>>>>>>>>>> sudo dpctl dump-flows tcp:127.0.0.1:6634
>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>> Initially it showed a single entry which directed the flows
>>>>>>>>>>>>>>> to the controller.
>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>> This is probably an entry to direct the discovery packets to
>>>>>>>>>>>>>>> the controller.
>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>> Then i executed h1 ping -c10 h2
>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>> mean while checked the flow entries using dpctl command
>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>> Now it listed some 5 entries including the previous one. For
>>>>>>>>>>>>>>> the same input and poutput pairs there were 2 entries..why is
>>>>>>>>>>>>>>> it so?
>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>> l2_multi installs exact-match flows. There's probably a
>>>>>>>>>>>>>>> pair for both directions of ARP and another pair for both
>>>>>>>>>>>>>>> directions of
>>>>>>>>>>>>>>> ICMP. Check the Ethertype and IP protocol numbers for each
>>>>>>>>>>>>>>> table entry.
>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>> Also, you should probably use the OVS tools (ovs-dpctl,
>>>>>>>>>>>>>>> ovs-ofctl, etc.) rather than the reference distribution's dpctl.
>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>> 2) According to my requirements, some of the switches have
>>>>>>>>>>>>>>> to generate renewal packets peroiodically, to keep the back up
>>>>>>>>>>>>>>> paths alive.
>>>>>>>>>>>>>>> Which part has to be modified for this?
>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>> You haven't really explained your strategy for backup paths,
>>>>>>>>>>>>>>> so I don't have any idea what you're talking about here.
>>>>>>>>>>>>>>> Offhand, it seems
>>>>>>>>>>>>>>> like you could just install the entries for your backup paths
>>>>>>>>>>>>>>> without a
>>>>>>>>>>>>>>> timeout.
>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>> Please clarify the doubts..
>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>> Thanks & Regards,
>>>>>>>>>>>>>>> Padma V
>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>>
>>>>>>>>>>>>>> <mytestping.py>
>>>>>>>>>>>>>>
>>>>>>>>>>>>>>
>>>>>>>>>>>>>>
>>>>>>>>>>>>>
>>>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>
>>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>
>>>>>>
>>>>>
>>>>>
>>>> <topo-3sw-2host.py>
>>>>
>>>>
>>>>
>>>
>>
>>
>
>
