![](bg2.png)
networking approaches offer up to an entire
wavelength for some time, through which GigaByte
(GB) files can be delivered. Optical burst (OBS)
(Jue and Vokkrane, 2005) or flow switching (OFS)
(Chan, 2010), or user-controlled end-to-end
lightpaths (e.g. CAnet4, MONET, CORONET and
GRIPhoN (Mahimkar et al., 2011)) have been
explored fully. For example, an OBS approach
(Yong et al., 2010) uses concatenated data bursts
where the data units are organized as non-contiguous
and non-periodic series of concatenated timeslots
(bursts), which are then handled as a whole in an all-
optical network infrastructure. These optical
approaches are not embraced by industry, in part
because the power efficiency of optical switching is
questionable (Tucker et al., 2009), (Tucker, 2006)
and optical buffers, widely used in most OBS
proposals, have not yet offered a commercially
viable alternative to electronic buffers. Also, while
capable of supporting large bandwidth, targeted
implementations are in the interconnection of
specialized nodes rather than broadly distributed
Internet users.
Hybrid architectures have been studied in which
both electronic and optical switching are combined
(Aleksic et al., 2011) to simultaneously handle
packets, bursts and TDM circuits. A large reduction
in the power consumption is achieved by selecting
adaptively which part of the node to activate based
on a per-flow evaluation of the data to be routed
while the other blocks are put in sleep or low-power
mode. While potentially powerful, this approach
requires the complex integration of disparate
switching and control elements, some of which (like
OBS and optical delay lines) have not proven
compelling individually.
A more incremental overlay network approach is
electronic burst switching (EBS) (Peng et al., 2010).
Following the OBS model, bursts are assembled at
edge burst switches and switched electronically at
core switches. It was concluded that using large
bursts (> 1 Mb) may lead to reduction in header-
related power consumption in core switches, but the
power consumed by burst assembly negates much of
the advantage gained in core switching.
In this paper we continue along the path of EBS.
Users share the bandwidth of an overlay network,
which we presume to be statically provisioned, using
electronic switches or routers specifically designed
to handle large file transactions. Unlike (Peng et al.,
2010), we eliminate burst assembly at edge switches
and consider direct end-to-end delivery of large
“media” frames (MF) (roughly 1-10Mb) to users
through an overlay to next-generation optical access
networks. Free from the constraints of coexisting
with highly granular and dynamic IP traffic, this
EBS overlay network can be designed specifically
for the efficient delivery of the large data
transactions that did not exist when the Internet was
conceived. Compared to traditional IP routers,
switch reconfiguration can be far less dynamic since
only very large packets are supported. Unlike
proposed optical alternatives, this can be
accomplished using available electronic buffers in a
form that is entirely compatible with today’s highly
efficient cross-point switch arrays.
Our objective is to enable a significant reduction
in power consumption of network hardware while
optimizing the use of resources. We first explore
routing MFs using a standard UDP protocol (MF-
UDP). UDP is selected for this study, rather than
TCP, as this avoids numerous complexities that add
little insight to a comparison with conventional IP
and, as discussed later, gives the best case scenario
for IP. Based on simple hardware considerations,
network performance simulations and comparison
with traditional UDP, we arrive at the anticipated
conclusion that router power consumption is reduced
dramatically, but performance is otherwise
unaffected and larger buffers are needed.
We then consider using concatenations of MFs
into periodic semi-transparent chains (MFCs) and
the scheduled transmission of these MFCs using a
two-way reservation mechanism. While such a
scheduling mechanism would be inappropriate for
traditional IP traffic, the large size of each MFC
(e.g. 1 GB) makes scheduling both manageable and
worthwhile. Also, the structure of an MFC makes it
easy to condense information on its configuration
with minimal control plane information, minimizing
the amount of information to be processed at each
node to schedule the chain and reducing the
probability of control plane collisions. Simulation
results show increased utilization efficiency and
decreasing buffer requirements in comparison to
MF-UDP as well as standard UDP. An MFC router
is designed based on a commercially available cross-
point switch array and power consumption is
estimated to be roughly 20% of that of a standard IP
router.
The paper is organized as follows: Section 2
provides an overview of the reference network
architecture in the context of transactions of large
files. Section 3 compares, using OMNeT++
simulation, traditional UDP to MF-UDP in
supporting representative large transactions. In
Section 4, chains of media frames (MFCs) are
introduced, along with exemplary admission control
SMARTGREENS2013-2ndInternationalConferenceonSmartGridsandGreenITSystems
44