Automated and Scalable QoS Control for Network Convergence Back to Program
Network convergence is becoming increasingly important
for cost reduction and management simplification.
However, this convergence requires strict performance
isolation while keeping fine-grained control of
each service (e.g. VoIP, video conference etc.). It is difficult
to guarantee the performance requirements for various
serviceswith manual configuration of the Quality-of-
Service (QoS) knobs on a per-device basis as is prevalent
today. We propose a network QoS control framework
for converged fabrics that automatically and flexibly programs
a network of devices with the necessary QoS parameters,
derived from a high level set of application requirements.
The controller leverages our QoS extensions
of OpenFlow APIs, including per-flow rate-limiters and
dynamic priority assignment. We also present some results
from a testbed implementation to validate the performance
of our controller.
The Case for Fine-Grained Traffic Engineering in Data Centers Back to Program
In recent years, several techniques have been suggested
for routing and traffic engineering in data centers. However,
not much is known about how these techniques performrelative
to each other under realistic data center traffic
patterns. Our preliminary study reveals that existing
techniques can only achieve 80% to 85% of the ideal solution
in terms of the number of bytes delivered. We find
that these techniques suffer due to their inability to utilize
global knowledge of the properties of traffic flows and
their inability to make coordinated decision for scheduling
flows at fine time-scales. Even recent traffic engineering
techniques such as COPE fail in data centers despite
their proven ability to adapt to dynamic variations,
because they are designed to operate at longer time scales
(on the order of hours, at least). In contrast, data centers,
due to the bursty nature inherent to their traffic, require
adaptation at much finer times scales. To this end, we define
a set of requirements that a data center-oriented traffic
engineering technique must possess in order to successfully
mitigate congestion. In this paper, we present
the design for a strawman framework that fulfills these
requirements.
HyperFlow: A Distributed Control Plane for OpenFlow Back to Program
OpenFlow assumes a logically centralized controller, which
ideally can be physically distributed. However, current
deployments rely on a single controller which has major drawbacks including lack of scalability. We present
HyperFlow, a distributed event-based control plane for
OpenFlow. HyperFlow is logically centralized but physically distributed: it provides scalability while keeping
the benefits of network control centralization. By passively synchronizing network-wide views of OpenFlow
controllers, HyperFlow localizes decision making to individual controllers, thus minimizing the control plane
response time to data plane requests. HyperFlow is resilient to network partitioning and component failures.
It also enables interconnecting independently managed
OpenFlow networks, an essential feature missing in current OpenFlow deployments. We have implemented
HyperFlow as an application for NOX. Our implementation requires minimal changes to NOX, and allows
reuse of existing NOX applications with minor modifications. Our preliminary evaluation shows that, assuming sufficient control bandwidth, to bound the window of inconsistency among controllers by a factor of
the delay between the farthest controllers, the network
changes must occur at a rate lower than 1000 events per
second across the network.
The "Platform as a Service" Model for Networking Back to Program
Decoupling infrastructure management from service
management can lead to innovation, new business models,
and a reduction in the complexity of running services.
It is happening in the world of computing, and is
poised to happen in networking. While many have considered
this in the context of network virtualization, they
all focus on one model—overlaying a virtual network
of multiple virtual routers on top of a shared physical
infrastructure, each completely isolated from the others
through the use of virtualization. In this paper we argue
for a different approach, where those running the
service are presented with the abstraction of a single
router in order to enable them to focus solely on their
service rather than worrying about managing a virtual
network as well. We discuss the abstraction of a single
router, and the challenges of mapping the collection of
abstract routers (from different parties) to the distributed
and shared physical infrastructure.
vDC: Virtual Data Center Powered with AS Alliance for Enabling Cost-Effective Business Continuity and Coverage Back to Program
In the cloud computing era, cloud providers must design data
centers that satisfy the requirements such as business continuity,
coverage and performance, and cost-effectiveness for offering
application providers the competitive hosting services.
However, it is hard for even elephant cloud providers to satisfy
these requirements all together because of the cost problem.
In this paper, we propose the concept of virtual data center
(vDC) of multiple geographically distributed data centers over
the Internet to extend the coverage of hosting services in a cost-effective
manner and apply the concept of AS alliance to ensure
resilient connectivity of vDC to achieve high business continuity.
We also introduce the detail design of AS alliance tailored
for the vDC concept and conduct a feasibility study for making
vDC connectivity robust.
Europa: Efficient User Mode Packet Forwarding in Network Virtualization Back to Program
Network virtualization provides the ability to run concurrent
virtual networks over a shared substrate. However,
it is challenging to design such a platform to host multiple
heterogeneous and often highly customized virtual
networks. Not only minimal interference among different
virtual networks is desired, high-speed packet forwarding
is also required. This paper presents Europa, a virtual network
platform uses Efficient UseR mOde Packet forwArding,
which supports high-speed and highly customizable
virtual networks. Our platform adopts lightweight OS-level
virtualization to slice a physical server into virtual
machines. The data plane of a virtual router runs in an
isolated virtual machine so as to safe for customization.
We design a new user mode packet processing scheme for
virtual routers hosted in Europa to achieve high speed forwarding.
Experiments show that an Europa virtual router
can be four times faster than conventional user mode software
router.
A Preliminary Analysis of TCP Performance in an Enterprise Network Back to Program
Although TCP behavior is one of the most studied aspects
of Internet traffic, little is known about TCP performance
within modern enterprise networks. In this paper
we analyze aspects of TCP performance observed in
packet traces taken over four months from a medium-sized
enterprise. We assess the prevalence of broken
TCP transactions, applications used, throughput of
TCP connections, and phenomena that influence performance,
such as retransmissions, out-of-order delivery,
and packet corruption. While much remains to explore,
this work represents a first step towards understanding
TCP performance in the under-studied environment.
Extensible and Scalable Network Monitoring Using OpenSAFE Back to Program
Administrators of today's networks are highly interested
in monitoring traffic for purposes of collecting statistics,
detecting intrusions, and providing forensic evidence.
Unfortunately, network size and complexity can make
this a daunting task. Aside from the problems in analyzing
network traffic for this information—an extremely
difficult task itself—a more fundamental problem exists:
how to route the traffic for network analysis in a robust,
high performance manner that does not impact normal
network traffic.
Current solutions fail to address these problems in a
manner that allows high performance and easy management.
In this paper, we propose OpenSAFE, a system
for enabling the arbitrary direction of traffic for security
monitoring applications at line rates. Additionally, we
describe ALARMS, a flow specification language that
greatly simplifies management of network monitoring
appliances. Finally, we describe a proof-of-concept
implementation that we are currently undertaking to
monitor traffic across our network.
Beyond the Best: Real-Time Non-Invasive Collection of BGP Messages Back to Program
Interdomain routing in the Internet has a large impact
on network traffic and related economic issues. For this
reason, BGP monitoring attracts both academic and industrial
research interest. The most common solution for
collecting BGP routing data is to establish BGP peerings
between border routers and a route collector.
The downside of this approach is that it only allows
us to trace changes of routes selected as best by routers:
this drawback hinders a wide range of analyses that need
access to all BGP messages received by border routers.
In this paper, we present an effective technique enabling
fast, non-invasive and scalable collection of all
BGP messages received by border routers. By selectively
cloning BGP traffic and sending it to a remote monitor,
we are able to collect BGP messages without establishing
additional BGP peerings. Our technique does not require
any new feature to be implemented by routers and we experimentally
show that our approach incurs a negligible
processing overhead at the border routers. Our prototype
implementation is able to process and archive all BGP
messages in near real-time on commodity hardware.
Experiences with Tracing Causality in Networked Services Back to Program
Unlike device-centric monitoring, task-centric tracing
enables an operator to causally trace the complete execution
of a networked system across the boundaries of applications,
protocols, and administrative domains. In this
paper, we argue that causal, end-to-end tracing should be
an integral part of network services. Moreover, it is not
fundamentally difficult to achieve, given a primitive that
propagates task metadata alongside logical execution and
communication paths.
X-Trace is a framework that relies on such propagation
to provide comprehensive causal tracing. We report
on our experience integrating X-Trace into several production
networked services—including 802.1X authentication,
Web content distribution, and DNS-based replica
selection—to illustrate benefits of causal tracing, and to
discuss the instrumentation of different protocols and
component architectures. We highlight the challenges we
encountered and techniques we developed to better integrate
causal tracing into network services.
Proactive Network Management of IPTV Networks Back to Program
Consumer communications and
entertainment services, including broadcast TV and
VoIP, require service providers to meet stringent
availability and latency constraints. When a packet
technology, such as IP, is used to transport these
services, this also poses stringent packet loss
requirement on the network. This aspect of IPTV,
where impairments have consumer-visible impact and
potential public relations consequences, creates new
challenges in protocol design, as well as network
management. The key to operating an effective network
is to expand to a more comprehensive network
management approach to be able to better anticipate
and manage potential network problems. This paper
describes network management techniques deployed in
a production IPTV network with over 2 million
customers.
|