Speaker
Christos Papadopoulos
(Colorado State University)
Description
Introduction
------------
The Computing Models of the LHC experiments continue to evolve from
the simple hierarchical MONARC model towards more agile models where
data is exchanged among many Tier2 and Tier3 sites, relying on both
strategic data placement, and an increased use of remote
access with caching through CMS's AAA and ATLAS' FAX projects, for example.
The challenges presented by expanding needs for CPU, storage and network capacity
have pointed the way towards future more agile pervasive models that make
best use of highly distributed heterogeneous resources.
In this paper, we explore the use of **Named Data Networking (NDN)** [1],
a new Internet architecture focusing on content rather than the location of the
data collections. As NDN has shown considerable promise in another data intensive field,
Climate Science, we discuss the similarities
and differences between the Climate and HEP use cases, along with specific
issues HEP faces and will face during LHC Run2 and beyond, which NDN
could address.
NDN
---------------------
NDN, an instance of Information Centric Networking (ICN), is a new Internet
architecture which focuses on the content of the data collections themselves,
rather than on where the data resides. The end host addresses are
replaced with content names, which, similar to URLs,
are hierarchical, unique and human readable. Thus, NDN imposes minimal
structure on applications, which can choose their own naming schemes.
The hierarchical structure of NDN names has several advantages:
1. it is an intuitive, common organizational structure (e.g., file systems, URLs, etc.),
2. it is scalable (similar to hierarchical IP addresses), and
3. coupled with longest prefix matching, it allows for data discovery and enumeration.
NDN has a wide range of potential benefits such as in-network
content caching with request deduplication to reduce congestion and improve delivery
speed, simpler application configuration, and security built into the
network at the data level.
The NDN concepts, structure and initial applications have been developed
through an NSF Future Internet Architecture project in its second round of funding,
involving eight universities.
NDN has attracted significant interest from industry, including Cisco, Intel,
Alcatel, Huawei, and Panasonic, and involves many of these companies
through an industry consortium.
NDN and Climate Applications
----------------------------
We have successfully begun to test NDN in the climate application domain [2].
To handle the various naming schemes used in climate applications, we
have designed and implemented translators that take existing names with
arbitrary structure (produced by climate models, or home-grown) and
translated them into NDN-compliant names. Depending on the original name
structure, the translation can be fairly direct (e.g., data that complies
with the "Data Reference Syntax" from the Coupled Model Intercomparison Project),
or complex (from home-grown naming schemes that require the analysis of
metadata embedded in the dataset or even user feedback in order to construct
proper NDN names).
We have deployed a dedicated 6-node testbed for climate applications that
reaches locations such as the Atmospheric and Computer Science Departments at
Colorado State University, LBNL and NWCS. The testbed is connected via
10G links by ESnet and is composed of high-end machines each with 40 core CPUs,
128GB RAM and 48TB diskspace. The machines cumulatively host over
50TB of climate data and are used for research, experimentation and development
of climate applications.
NDN Support for HEP Applications
------------------------------------------------------
Several features of NDN can be beneficial to the HEP computing use case.
Data sources publish new
content to the network following an agreed upon naming scheme. Data delivery
is always performed in a pull mode, driven by the consumer issuing interest
packets. Intermediate nodes in the network dynamically cache data based on
content popularity, ready to satisfy subsequent interests directly from
the cache, thus lowering the load on servers with popular content.
Combining this with the pull-mode results in a multicast-like data delivery,
possibly optimizing both the network utilisation as well as server load.
The use of multiple data sources simultaneously, as well as the
native use of multiple paths between client and data source, provide for
robust failover in case of network segment, node, or end-site failure.
All these are active research areas today. Caching as well as forwarding
strategies, naming schemes, multi-sourcing and multi-path forwarding need
to be investigated not only from the network but also the application perspective.
HEP experiments using the World-wide LHC Computing Grid (WLCG)
have well-developed, hierarchical naming schemes in use, which already fit the NDN
approach well. We take this logical file name structure as a starting
point for investigating the benefits of using NDN as the data distribution and
access network for HEP data processing. For this, we use the testbed described above.
We further target simultaneous optimization of storage and bandwidth resource utilization
through dynamic caching using the VIP framework in [3].
For the scalability study, we complement the testbed with the use of a simulation
environment with a representative topology including network nodes and end-sites.
Summary
-------
In this paper, we study data access over an NDN testbed developed
for Climate research. We study the behaviour using HEP-like data structures based
on the CMS naming scheme, showing data publishing, discovery and retrieval in
an NDN network. We demonstrate the benefits of caching, speeding up data delivery
in multi-job access from a single source, with jobs executing at multiple sites.
We also show the results of the simulation studies of remote data access over
an NDN network demonstrating the scalability of the system.
References
----------
1. V. Jacobson, et al.; "Networking Named Content", 2009
2. C. Olschanowsky, et al., "Supporting Climate Research using Named Data Networking", LANMAN, 2014
3. E. Yeh, et al.; "VIP: A Framework for Joint Dynamic Forwarding and Caching in Named Data Networks", Proc. ACM Conf. on Information-Centric Networking, 2014
Primary authors
Artur Jerzy Barczyk
(California Institute of Technology (US))
Susmit Shannigrahi
(Colorado State University)
Co-authors
Alex Sim
(LAWRENCE BERKELEY NATIONAL LABORATORY)
Christos Papadopoulos
(Colorado State University)
Edmund Yeh
(Northeastern University)
Harvey Newman
(California Institute of Technology (US))
Inder Monga
(ESNET)
John Wu
(LAWRENCE BERKELEY NATIONAL LABORATORY)
