Automatic Failure Recovery for Software-Defined Networks

Automatic Failure Recovery for Software-Defined Networks
Automatic Failure Recovery for Software-Defined Networks
Maciej Kuźniar† ∗
Peter Perešíni† ∗
TU Berlin / T-Labs
Nedeljko Vasić†
Dejan Kostić‡
Institute IMDEA Networks
[email protected][email protected]
These authors contributed equally to this work
Conceptually, this requires computing the new forwarding
state in response to a failure event, similarly to how a linkstate routing protocol re-converges after detecting a failure.
In an SDN however, this computation takes place at the
SDN controller, based on its current view of the network.
To achieve higher availability, this computation could be
done in advance to determine backup paths for a range of
failure scenarios. With the appropriate hardware support1 ,
the backup paths can be preinstalled to enable switches to
quickly adapt to failures based just on local state.
Regardless of whether the recovered forwarding state is
computed in response to a failure or precomputed in advance, failure recovery requires the presence of extra software logic at the controller. In particular, in today’s modular controller platforms such as POX, Floodlight, etc., each
individual module (e.g., access control, routing) potentially
needs to include its own failure recovery logic [6]. This leads
to controller modules that are more difficult to develop, and
potentially increases the chance of bugs [1]. Avoiding this
problem by relying on simple recovery logic like timeouts or
deleting rules forwarding to an inactive port is inefficient [6]
and may introduce forwarding loops in the network [4].
In this paper, we propose AFRO (Automatic Failure Recovery for OpenFlow), a system that provides automatic failure recovery on behalf of simpler, failure-agnostic controller
modules. Inspired by the successful implementation of such
a model in other domains (e.g., in the large-scale computing framework MapReduce [2]), we argue that applicationspecific functionality should be separated from the failure
recovery mechanisms. Instead, a runtime system should be
responsible for transparently recovering the network from
failures. Doing so is critical for dramatically reducing the
chance of introducing bugs and insidious corner cases, and
increasing the overall chance of SDN’s success.
AFRO extends the functionality of a basic controller program that is only capable of correctly operating over a network without failures and allows it to react to changes in
the network topology. The intuition behind our approach is
based on a straightforward recovery mechanism, or rather
a solution that sidesteps the recovery problem altogether:
After any topology change, we could wipe clean the entire
network forwarding state and restart the controller. Then,
the forwarding state gets recovered as the controller starts
to install forwarding rules according to its control logic, initial configuration and the external events that influence its
Tolerating and recovering from link and switch failures
are fundamental requirements of most networks, including Software-Defined Networks (SDNs). However, instead
of traditional behaviors such as network-wide routing reconvergence, failure recovery in an SDN is determined by
the specific software logic running at the controller. While
this admits more freedom to respond to a failure event, it
ultimately means that each controller application must include its own recovery logic, which makes the code more
difficult to write and potentially more error-prone.
In this paper, we propose a runtime system that automates failure recovery and enables network developers to
write simpler, failure-agnostic code. To this end, upon detecting a failure, our approach first spawns a new controller
instance that runs in an emulated environment consisting of
the network topology excluding the failed elements. Then, it
quickly replays inputs observed by the controller before the
failure occurred, leading the emulated network into the forwarding state that accounts for the failed elements. Finally,
it recovers the network by installing the difference ruleset
between emulated and current forwarding states.
Categories and Subject Descriptors
C.2.4 [Distributed Systems]: Network operating systems;
C.4 [Performance of Systems]: Reliability, availability,
and serviceability
Software Defined Network, Fault tolerance
Marco Canini♯
To ensure uninterrupted service, Software-Defined Networks (SDNs) must continue to forward traffic in the face of
link and switch failures. Therefore, as with traditional networks, SDNs necessitate failure recovery—adapting to failures by directing traffic over functioning alternate paths.
Permission to make digital or hard copies of part or all of this work for
personal or classroom use is granted without fee provided that copies are
not made or distributed for profit or commercial advantage and that copies
bear this notice and the full citation on the first page. Copyrights for thirdparty components of this work must be honored. For all other uses, contact
the owner/author(s).
HotSDN’13, August 16, 2013, Hong Kong, China.
ACM 978-1-4503-2178-5/13/08.
In OpenFlow, the FastFailover group type is available since
version 1.1 of the specification.
However, because all rules in the network need to be reinstalled, such a simple method is inefficient and has significant disadvantages. First, a large number of packets may be
dropped or redirected inside the PacketIn messages to the
controller, which could be overwhelmed. Second, the recovery speed will be adversely affected by several factors, including the time to reset the network and controller, as well
as the overhead of computing, transmitting and installing
the new rules.
To overcome these issues, we take advantage of the “software” part in SDN. Since all the control decisions are made
in software running at the controller, AFRO repeats (or predicts) these decisions by simply running the same software
logic in an emulated environment of the underlying physical network [3, 7]. During normal execution, AFRO simply
logs a trace of the events observed at the controller that influence its execution (for example, PacketIn messages). In
case of a network failure, AFRO replays at accelerated speed
the events prior to the failure within an isolated instance of
the controller with its environment—except it pretends that
the network topology never changed and all failed elements
never existed since the beginning. Finally, AFRO recovers
the actual network forwarding state by transitioning it to the
recovered forwarding state of the emulated environment.
formance is suboptimal even if there is fast failover implemented. Further, consecutive failures may cause issues that
even failover can not handle. To improve the replay performance, we use two orthogonal optimizations. First, we
filter PacketIn messages and choose only those necessary to
reach the recovered forwarding state. Then, we parallelize
the replay based on packet independence.
2.2 Network Reconfiguration
After the replay ends, AFRO needs to efficiently apply
the changes without putting the network in an inconsistent
state. The reconfiguration requires two steps: (i) modifying
rules in the switches, and (ii) migrating the controller to a
new state.
For rules in the switches we need a consistent and proportional update mechanism. If the size of an update is much
bigger than the size of a required configuration change, it
offsets the advantages of AFRO. Therefore we propose using a two-phase commit mechanism of per-packet consistent
updates that is very similar to the one presented as optimization in [5]. We start modifying rules that will not be
used and use a barrier to wait until switches apply a new
configuration. At this point we install rules that direct traffic to the previously installed, new rules.
The transition between the old and new controller can
be done either by exposing an interface to pass the entire
state between the two controllers, or by replacing the old
one altogether. In the second case, the Shadow Controller
simply takes over the connections to real switches and takes
responsibility of the original controller.
AFRO works in two operation modes: record mode, when
the network is not experiencing failures, and recovery mode,
which activates after a failure is detected. The network and
controller start in a combined clean forwarding state that
later is gradually modified by the controller in response to
PacketIn events. In record mode, AFRO records all PacketIn arriving at the controller and keeps track of currently
installed rules by logging FlowMod and FlowRem messages.
When a network failure is detected, AFRO enters the recovery mode. Recovery involves two main phases: replay and
During replay, AFRO creates a clean copy of the original
controller, which we call Shadow Controller. Shadow Controller has the same functionality as the original one, but
it starts with a clean state and from the beginning works
on a Shadow Network—an emulated copy of the actual network that is modified by removing failed elements from its
topology. Then, AFRO feeds the Shadow Controller with
recorded PacketIn messages processed by the original controller before the failure occurred. When replay ends, the
Shadow Controller and Shadow Network contain a new forwarding state.
At this point, reconfiguration transitions the actual network from the current state to the new one. This transition
requires making modifications to both the controller internal state as well as forwarding rules in the switches. First,
AFRO computes a minimal set of rule changes. Then, it uses
an efficient, yet consistent method to update all switches. Finally, it replaces the controller state with the state of Shadow
Controller and ends the recovery procedure. In case another
failure is detected during recovery, the system needs to interrupt the above procedure and restart with another Shadow
Network representing the current topology.
2.3 Prototype
We implemented a prototype of AFRO working with a
POX controller platform. POX allows programmers to extend the controller functionality by adding new modules.
Therefore, it is possible to introduce AFRO requiring no
modifications to controller logic.
[1] M. Canini, D. Venzano, P. Perešı́ni, D. Kostić, and
J. Rexford. A NICE Way to Test OpenFlow
Applications. In NSDI, 2012.
[2] J. Dean and S. Ghemawat. MapReduce: Simplified
Data Processing on Large Clusters. Communications of
the ACM, 51(1), 2008.
[3] N. Handigol, B. Heller, V. Jeyakumar, B. Lantz, and
N. McKeown. Reproducible Network Experiments
Using Container-Based Emulation. In CoNEXT, 2012.
[4] P. Perešı́ni, M. Kuźniar, N. Vasić, M. Canini, and
D. Kostić. OF.CPP: Consistent Packet Processing for
OpenFlow. In HotSDN, 2013.
[5] M. Reitblatt, N. Foster, J. Rexford, C. Schlesinger, and
D. Walker. Abstractions for Network Update. In
SIGCOMM, 2012.
[6] S. Sharma, D. Staessens, D. Colle, M. Pickavet, and
P. Demeester. Enabling Fast Failure Recovery in
OpenFlow Networks. In DRCN, 2011.
[7] A. Wundsam, D. Levin, S. Seetharaman, and
A. Feldmann. OFRewind: Enabling Record and Replay
Troubleshooting for Networks. In USENIX ATC, 2011.
2.1 Replay
Replay needs to be quick and efficient. While AFRO is
reacting to the failure, the network is vulnerable and its per-
Was this manual useful for you? yes no
Thank you for your participation!

* Your assessment is very important for improving the work of artificial intelligence, which forms the content of this project

Download PDF