Data Center Networking in the Cloud Era Technical
Seminar
November 28, 2010
Organized by IBM R&D Labs in Israel
IBM Research Lab Auditorium, Haifa
Abstracts
Efficient Live Migration in the Cloud
Prof. Danny Raz, Technion
One of the big promises of cloud computing is the ability to
provide high quality services at a low cost, which is a
result of sharing the resources among many users (economy of
scale). However, the fact that resources can be shared is by
itself not sufficient to guarantee high availability of
services with the needed QoS level at a reasonable cost. One
has also to make sure that the use of resources is
effective. A key building block in ensuring cost effective
use of resources is the ability to move both applications
and user data between different nodes in the cloud in an
efficient way.
This talk addresses this point by studying efficient ways of
application and data migration in the cloud. While live
migration provides extreme versatility of management, it
comes at a price of degraded service performance during
migration. The bulk of studies devoted to live migration of
virtual machines, focus on the duration of the copy phase as
a primary metric of migration performance. While shorter
down times are clearly desirable, the pre-copy phase imposes
an overhead on the infrastructure that may result in severe
performance degradation. We observe that there is a
non-trivial trade-off between minimizing the copy phase
duration and maintaining an acceptable quality of service
during the pre-copy phase, and we introduce a new model to
quantify this tradeoff. We then show that using our model,
an optimal migration algorithm can be calculated
efficiently. Finally, we simulate, using real traces, live
migrations of a virtual machines running a search engine and
compare the migration cost using our algorithm and commonly
used live-migration methods.
This talk is based on joint paper with David Breitgand and
Gilad Kutiel.
Network Virtualization
David Hadas, IBM
Virtual machines are of very little use if they cannot
access the underlying physical network. Virtualizing the
network has traditionally been considered a challenge best
met by such network-centric measures as VLANs, implemented
by switches. In this work, we present the benefits of adding
network abstraction to hypervisors, moving the
responsibility for network virtualization from the network
to the server. Modern hypervisors do a poor job in
virtualizing the network, leaking details of the physical
network into virtual machines. For example, IP addresses
used across the host's physical network, are exposed to
guest virtual machines. We propose a method for plugging the
network-related leaks by ensuring that the virtual network
traffic is encapsulated inside a host envelope prior to
transmission across the underlying physical network.
IBM's PowerEN Developer Cloud: Fertile Ground for
Academic Research
Amit Golander, IBM
The Power Edge of Network(PowerEN) processor, recently
announced by IBM, merges network and server attributes to
create a wire-speed processor. PowerEN is a hybrid computer
that employs: massive multithreading capabilities,
integrated I/O and unique special-purpose accelerators for
compression, cryptography, pattern matching, XML and
Ethernet.
As a novel architecture, the PowerEN processor offers
fertile ground for research. It can facilitate the
development of faster applications in many fields of
computer science such as Networking, Cryptography,
Virtualization, Bioinformatics, SOA, and Systems. IBM
encourages academic research and has set up the "PowerEN
Developer Cloud" infrastructure to allow it. Israeli
universities are of the first to perform research on this
global infrastructure.
Optimizing Application Delivery for Cloud Services
Amir Peles, Radware
Amir Peles is the Chief Technology Officer of Radware, the
leading Application Delivery and Application Security
company. As virtualization and cloud computing take off as
the leading trends in enterprise and Telco IT departments,
they also bring challenges in maintaining user quality of
experience. Amir will be presenting the unique advantages of
smart application delivery for optimizing the Virtual Data
Center and Cloud Services. A special focus will be given to
the management of the Application Delivery functions as part
of the overall data center and cloud management solution.
Distributed Oblivious Load Balancing Using Prioritized
Job Replication
Amir Nahir, Technion
Distributed Oblivious Load Balancing Using Prioritized Job
Replication" Effective load balancing in large distributed
server systems is a highly complex optimization problem,
which often involves multiple objectives and many
parameters.
Irrespective of the precise optimization criteria, any
attempt to address such an optimization problem would incur
significant overhead due to the need to collect the required
(state-dependent) information from the various servers.
In this study we propose to tackle the problem through an
oblivious approach, i.e., a distributed load-sharing scheme
that does not use any state information. Our scheme is based
on creating, in addition to the regular job requests that
are assigned to a randomly chosen server, also low priority
job request replicas that are sent to a different server. We
show that, when servers can coordinate the removal of
redundant copies upon completion of a job, the performance
of the system exhibits significant improvement of up to a
factor of $2.2$, even under high load conditions, if job
lengths are exponentially distributed, and a more dramatic
improvement of over a factor of $5$ for typical loads, when
job lengths adhere to a heavy-tailed distribution.
For cases where such coordination is not feasible, we
propose simple, coordination-free schemes, which still yield
a significant improvement of up to a factor of $1.15$ for
exponentially distributed job lenghts and up to a factor of
over $1.5$ for the heavy-tailed case.
We then design a hybrid scheme, which, while requiring just
limited coordination, retains the main benefits of the
full-coordination scheme. Finally, we demonstrate the
benefit of our prioritized-replicas approach also within the
realm of non-oblivious and centralized schemes.
This talk is based on joint paper with Danny Raz and Ariel
Orda.
