Dq36708711

Kiran Kumar Reddi et al Int. Journal of Engineering Research and Applications
ISSN : 2248-9622, Vol. 3, Issue 6, Nov-Dec 2013, pp.708-711

RESEARCH ARTICLE

www.ijera.com

OPEN ACCESS

Different Techniques to Transfer Big data: a Survey
Kiran Kumar Reddi1 Dnvsls Indira2
1
2

Department of Computer Science, Krishna University, Machilipatnam, Krishna Dt., AP- 521001
Research Scholar, Department of CS, Krishna University, Machilipatnam, Krishna Dt., AP- 521001

ABSTRACT
Now a days the world is moving around the social networks, i.e most of the people are interacting with each
other via internet only. Transmitting data via the Internet is a routine and common task for users today.
Transferring a gigabyte of data in an entire day was normal, however users are now transmitting multiple
gigabytes in a single hour. The Big Data is the combination of structured, semi-structured, unstructured,
homogeneous and heterogeneous data. With the influx of big data and massive scientific data sets that are
measured in tens of petabytes, a user has the propensity to transfer even larger amounts of data. When
transferring data sets of this magnitude on public or shared networks, the performance of all workloads in the
system will be impacted. This paper addresses the issues and challenges inherent with transferring big data
over networks. A survey of current transfer techniques is provided in this paper.
Key words: Big Data, Grid, Parallel Transfer, NICE
is a complicated and time-consuming process.
These long duration transfers could take tens of hours
I.
INTRODUCTION
to several days and a normal "one click and wait"
Over the past several years there has been
method will not suffice. During the course of the
a tremendous increase in the amount of data being
transfer, servers may go off-line and network
transferred between Internet users. Escalating usage
conditions may change that either hinder or stop the
of streaming multimedia and other Internet based
transfer completely. The user needs to know how to
applications has contributed to this surge in data
maintain the data transmission until completion. This
transmissions. Another facet of the increase is
paper gives an overview of previously existing
due to the expansion of Big Data, which refers
techniques to transfer files around the world.
to data sets that are an order of magnitude larger
This section concentrates on grid computing
than the standard file transmitted via the Internet.
to distribute data and parallel transfer techniques.
Big Data can range in size from hundreds of
gigabytes to petabytes. Today everything is being
2.1 GRIDS
stored digitally.
Within the past decade,
Grid computing has emerged as a
everything from banking transactions to medical
framework
for
aggregating
geographically
history has migrated to digital storage. This
distributed, heterogeneous resources that enables
change from physical documents to digital files has
secure and unified access to computing, storage and
necessitated the creation of large data sets and
networking resources (1). Grid applications have
consequently the transfer of large amounts of data.
vast datasets and/or complex computations that
There is no sign that the amount of data being stored
require
secure
resource
sharing
among
or transmitted by users is steady or even decreasing.
geographically distributed systems. The term "Grid"
Every year average Internet users are moving
was inspired by the electrical grid system, where a
more and more data through their Internet
user can plug in an appliance to a universal socket
connections. Depending on the bandwidth of these
and have instant access to power without knowing
connections and the size of the data sets being
exactly where that power was generated or how
transmitted, the duration of transfers could
it came to reach the socket (1). The vision for
potentially be measured in days or even weeks.
grids was similar.
A user could simply access as
There exists a need for an efficient transfer technique
much computing power as required through a
that can move large amounts of data quickly and
common interface without concern for who was
easily without impacting other users or
providing the resources. Currently, grids have not
applications.
This paper presents different
yet reached that level of simplicity. Grids offer
techniques to transfer the data across the globe. This
coordinated resource sharing and problem solving in
paper concentrated on survey on grids, parallel
dynamic, multi institutional virtual organizations (2).
techniques to transfer data on internet and a new
A virtual organization (VO) comprises a set of
model to transfer big data efficiently across the globe.
individuals and/or institutions having access to
computers, software, data, and other resources for
II.
RELATED WORK
collaborative problem-solving or other purposes (3).
Retrieving large data files (GB, TB, PB)
www.ijera.com

708 | P a g e

A grid can also be defined as a system that
coordinates resources that are not subject to
centralized control, using standard, open, generalpurpose protocols and interfaces in order to
deliver nontrivial qualities of service (4). Data grids,
a specialized extension of grid computing, are
responsible for providing the infrastructure and
services to access, transfer, and modify massive
datasets stored in distributed storage resources (5).
They allow users to access computational and storage
resources in order to execute data-intensive
applications on remote data. The objective of a data
grid system is to integrate heterogeneous data files
stored in a large number of geographically distributed
sites into a single virtual data management system
and to provide diverse services to fit the needs of
high-performance distributed and data intensive
computing (5).
2.2 GRID MIDDLEWARE
The sharing of resources in a grid is facilitated
and controlled by a set of services that allow
resources to be discovered, accessed, allocated,
monitored, and accounted for, regardless of the
their physical location (6). Since these services
create a layer between physical resources and
applications, they are often referred to as Grid
Middleware.
Every grid has different service
requirements, therefore the architecture and grid
middleware implementation of every grid can vary.
The middleware of many grids is based on the
software architecture called the Globus Toolkit
(7).
The toolkit is a set of libraries and
programs that address common problems that
occur when building distributed system services
and applications (8). It provides a set of
infrastructure services that implement interfaces for
managing computational, storage,
and other
resources. The Globus Toolkit provides all of these
services and it is left to grid administrators to
determine whether or not to include certain services
in their grid implementation. These are a few of the
well known and widely used grids that deploy
the Globus Toolkit: TeraGrid, Open Science Grid,
EGEE, Worldwide LHC Computing Grid (WLCG)
(8), China National Grid, UK National Grid Service
and NAREGI (9). The architecture of the Globus
Toolkit contains several components, each of which
is responsible for different grid functions. A few of
these services are (7):
• Grid Resource Allocation and management
(GRAM) - This service initiates, monitors, and
manages the execution of computations on
remote computers. It allows a user to specify:
the quantity and type of resources needed, the
data sets required for their computation, the
executable application to be run, the
necessary security credentials, and the job
persistence requirements.
• Data access and movement - The reliable file
www.ijera.com

www.ijera.com

transfer (RFT) service is provided to ensure that
data is successfully transferred from one location
to another.
• Replica Management - This service keeps track
of all replicas and their content using a replica
location service (RLS) and a data replication
service (DRS).
• Monitoring and Discovery - Multiple services
collect and process information about the
configuration and state of all resources to enable
monitoring of system status.
• Security - Services establish the identity of
users or services (authentication), protect
communications, and determine who is
allowed
to
perform
what
actions
(authorization), as well as manage user
credentials and maintain group membership
information.
There are several grid applications
available for moving data from one location in a
grid to another. The most widely-used data
movement tool, which is also a component of the
Globus Toolkit, is called GridFTP (10; 11). It is
an extension of the File Transfer Protocol (FTP) and
was designed specifically for grid environments.
Several data retrieval techniques developed
specifically for retrieving large files in grid
computing environments. The sizes of data files
requested in grids are much larger than normal web
data requests. It is not uncommon for a grid data
file size to be measured in gigabytes or terabytes.
Users want to be able to download these files as
quickly as possible, by any means necessary.
Since utilizing a single server can be limiting,
retrieving data from multiple servers in a parallel
(also known as data co-allocation) has been
suggested as an alternative.

III.

PARALLEL TRANSMISSION
TECHNIQUES USED ON THE
INTERNET

Rodriguez
and
Biersack
present
mechanisms for parallel access to data on the
Internet (11). They develop two different parallelaccess schemes: history-based and dynamic. The
goal for all of their schemes is to balance the load
amongst all available servers by allocating a
workload to each replica that is proportional to its
service rate. The authors state that parallel access has
additional overhead in comparison to a single access.
The additional overhead occurs when multiple
connections are opened and extra traffic is generated
to perform block requests. In order to minimize
these overhead costs, these techniques should only
be utilized for larger files. Their history-based
technique utilizes a database with information about
previous rates from the different servers to the
receiver in order to estimate future transfers. Using
these estimates the algorithm assigns varying
portions of the file to each replica with the goal that
709 | P a g e

all servers will finish transferring the portions at the
same time. The authors evaluate their history-based
technique using live webservers on the Internet
distributed across the world. Due to the presence of
other Internet traffic, the authors found that the
performance of their technique varied at different
times of the day. They found that network
conditions rapidly change and estimating the
transfer rate to every server using past histories
results in poor estimates. Their results show that
during peak traffic times when transfer rates vary
dramatically and historical information is not a
good indicator of future performance, their historybased parallel
access technique
has higher
download times than clients accessing a single
server. In response to the performance of their
history-based technique, the authors develop a
dynamic technique that adjusts to changing
network conditions. Their dynamic technique
divides the desired file into a fixed number of
equal sized blocks. The client requests one block
from every replica. When a server completes a
request, another block is assigned. When there are a
small number of blocks outstanding, idle servers
are requested to deliver blocks that have already
been assigned to another server, but have yet to
be received. There will then be multiple servers
working on the same requests. The authors state that
the bandwidth wasted on overlapping these requests
is smaller than the worst-case scenario of waiting
for the slowest server to deliver the last block. To
further enhance the performance of their
technique, they utilize TCP-persistent connections
between the client and every server to minimize
the overhead of opening multiple TCP connections.
They also propose pipelining the requests to each
server in order to decrease inter block idle times.
With pipelining, a new block request is sent to a
server before the previous block request is
completely received. In the evaluations of their
dynamic technique, the authors find that there is
a significant speedup in comparison to a single
server access. Since the dynamic technique is not
relying on historical information and can adapt to
changing
network conditions, it
has greater
performance than requesting data from a single
server even under peak traffic conditions. They also
observe that the transfer time of a dynamic parallel
access is very close to the optimum transfer time.
Utilizing request pipelining, the authors demonstrate
that their technique would be almost equal to the
optimal transfer time. These advanced retrieval
techniques allow users to utilize multiple resources
simultaneously.
These
advanced
techniques
provide improved performance for users, however
they are quite complicated to implement and use.
They require significant user involvement and
require multiple user decisions that can
dramatically affect the performance of the transfer.
A user needs know how in order to make these
www.ijera.com

www.ijera.com

techniques function properly and efficiently. Another
configuration option that is frequently left for the
user to determine is segment size.
In some
advanced techniques, the data file is divided into
small portions called segments. The segment size is
often left for the user to decide and the size chosen
can affect the performance of the transfer.
Determining the appropriate segment size is not a
simple task. If the size is too small, a server may
receive hundreds to thousands of requests for
portions of a single file. This will result in longer
disk service times at a server, as the number of
users increases. A server's storage system can best
service requests if it has greater knowledge of a user's
workload. It can better schedule reading from the
hard disks, as well as take advantage of prefetching and caching strategies.

IV.

NICE MODEL

A new, NICE Model for Big Data
transfers, which is based on a store-and-forward
model instead of an end-to-end approach, is
presented. This nice model ensures that Big Data
transfers only occur during low demand periods when
there is idle bandwidth that can be repurposed for
these large transfers. Under this model, Big Data are
transmitted when the Internet traffic at the senders
LAN is low. If the Internet traffic at the receivers
LAN is high at this time, then the data are stored at a
staging server and later transmitted to the receiver.
Similar to the nice command in Linux, a transfer
tool based on the nice algorithm, gives itself low
priority and is nice to other applications using the
Internet. The overall goal is to develop an
application that can transmit big data via the
Internet for all users with in any campus.

Fig 1: Nice Model architecture
In order to avail of maximum installed
bandwidth without impacting other users, the key is
to open multiple transmission streams during low
traffic. If the sender and receiver are in the
same time zone then a direct transmission from
sender to receiver is feasible. If the sender and
receiver are in different time zones, then the low
traffic periods at the two end points do not
coincide. In this instance, the file is transmitted
from the sender to one or more staging server(s),
placed in the Internet zone. Depending on the
Internet configuration between the sender and
710 | P a g e

receiver, the file may be transmitted to a single
staging server via multiple streams or the file
may be divided and parts of the file are transmitted
concurrently to multiple staging servers. When the
receiver's LAN traffic is low, the file can be
transmitted from the staging server(s) to the
receiver. A file transmission tool such as GridFTP
could be used for the transmission from the sender to
the staging server(s) and from the staging server(s) to
the receiver.

V.

CONCLUSION & FUTURE WORK

Big data transfers via the Internet are
not a commonplace task for most users today.
Currently, there are no tools to facilitate these kinds
of transmissions. The task of transferring massive
amounts of data across the country or even the globe
is a challenging and daunting undertaking for any
user.
As the popularity of distributed storage
propagates and the amount of scientific data
continues to surge, the demand for big data transfers
will grow at a tremendous rate. The existing tools
for moving large amounts of data are based on
the parallel model, which is designed to grab as
much bandwidth as possible by opening
concurrent data streams. This greedy approach
may be good for a single user's transfer,
however it is not scalable for multiple users on a
shared network. The entire system suffers when
users attempt to grab bandwidth. Parallel system
supports multiple servers to transfer data via internet
but implementation of that system is very difficult.
The Nice model can be used for big data transfers.
Under this model, these transfers are relegated to low
demand periods when there is ample, idle bandwidth
available.
This bandwidth can then be
repurposed for big data transmissions without
impacting other users in the system. Since the
nice model uses a store-and-forward approach by
utilizing staging servers, the model is able to
accommodate differences in
time zones and
variations in bandwidth. There is a lot of challenging
issues like network protocols, security, compression
techniques, routing algorithms, Traffic Monitoring
and bandwidth management devices. New algorithms
are required to transfer big data around the globe by
concentrating all the above issues.

[3]

[4]
[5]

[6]

[7]

[8]
[9]
[10]

[11]

REFERENCES
[1]

[2]

FOSTER, I., AND KESSELMAN, C. The
Grid 2: Blueprint for a New Computing
Infrastructure.
Morgan
Kaufmann
Publishers Inc., San Francisco, CA, USA,
2003.
FOSTER, I., KESSELMAN, C., AND
TUECKE, S. The anatomy of the Grid:
Enabling scalable virtual organizations. The
International Journal of High Performance
Computing Applications 15, 3 (Fall 2001),
200-222.

www.ijera.com

[12]

www.ijera.com

LAURE,
E.,
FISHER,
S.
M.,
FROHNER, A., GRANDI, C., KUNSZT,
P. Z., KRENEK, A., MULMO, O.,
PACINI, F., PRELZ, F., WHITE, J.,
BARROSO, M., BUNCIC, P., HEMMER,
F., DI MEGLIO, A., AND EDLUND, A.
Programming the grid with glite.
Computational Methods in Science and
Technology 12, 1 (2006), 33-45.
FOSTER, I. What is the grid? a three point
checklist. GRIDToday (July 2002).
XLAO QIN, H. J. Data Grids: Supporting
Data-Intensive Applications in Wide Area
Networks. 2006, pp. 481-494.
LAURE, E., HEMMER, F., PRELZ, F.,
BECO, S., FISHER, S., LIVNY, M.,
GUY, L., BARROSO, M., BUNCIC,
P., KUNSZT, P. Z., DI MEGLIO, A.,
AIMAR, A., EDLUND, A., GROEP, D.,
PACINI, F., SGARAVATTO, M., AND
MULMO, O. Middleware for the next
generation grid infrastructure. Computing in
High Energy Physics and Nuclear
Physics (October 2004), 826.
FOSTER, I. Globus toolkit version 4:
Software for service-oriented systems. In
IFIP International Conference on
Network and Parallel Computing (2006),
pp. 2-13.
Worldwide lhc computing grid (wlcg).
http://lcg.web.cern.ch/LCG/.
Naregi: National research grid initiative,
http://www.naregi.org, 2009.
ALLCOCK,
W.,
BESTEE,
J.,
BEESNAHAN, J., CHEEVENAK, A.
L.,
FOSTEE, I., KESSELMAN, C.,
MEDEE, S., NEFEDOVA, V., AND
STEVEN, D. Q. Secure, efficient data
transport and replica management for
high-performance data-intensive computing.
In IEEE Mass Storage Conference (2001).
ALLCOCK,
W.,
BESTEE,
J.,
BEESNAHAN, J., CHEEVENAK, A.
L.,
FOSTEE, I., KESSELMAN, C.,
MEDEE,
S.,
NEFEDOVA,
V.,
QUESNEL, D., AND TUECKE, S. Data
management and transfer in high
performance
computational
grid
environments. Parallel Computing Journal
28, 5 (May 2002), 749-771.
RODRIGUEZ, P., AND BIERSACK, E.W.
Dynamic parallel access to replicated
content in the internet. IEEE/ACM Trans.
Netw. 10, 4 (2002), 455-465.

711 | P a g e

Dq36708711

More Related Content

What's hot (20)

Viewers also liked (20)

Similar to Dq36708711 (20)

Recently uploaded (20)

Dq36708711