Cluster and Grid Computing Come
of Age
Cluster and grid computing has long been the
domain of scientific computing, but as the technologies mature,
enterprises of all sizes and industries are reaping the benefits of
lower-cost, higher performance computing.
Five years ago, clusteringóan OS-level technology designed to
aggregate the processing power of multiple serversómeant little to
managers of Intel-based server systems beyond fault tolerance. Sun,
Microsoft, and many Linux vendors each added basic failover
capabilities to their respective systems, but those features merely
allowed IT managers to link two servers together to ensure a higher
level of availability. Today, OS vendors are expanding their
clustering capabilities to make it possible to link four or more
servers together, manage clusters as a single system, and most
significantly, share the processing power among all the servers in
the cluster to vastly increase application scalability.
For example, Microsoft's Cluster Services is evolving from a
separate utility for linking pairs of Exchange or SQL Servers
together to an integral part of the OS (see
related story). Sun's Sun Clusters 3.0, released in October
2001, now lets IT managers cluster up to eight Solaris servers
together, and distributeóor load balanceóapplication processing
tasks among those servers. RedHat's Linux Advanced Server, released
in May 2002, comes with a copy of its still under-construction Load
Balancer for distributing workload among clustered Linux servers,
giving IT managers significantly more flexibility in choosing
applications that can take advantage of the technology.
Likewise, grid computing used to be considered an esoteric
technology for creating government supercomputers. Now, IT managers
are seriously considering buying processing power as they need it
from IBM, Gateway, and an expanding array of others. Grid
computingówhich in effect lets companies borrow processing power
from groups of computersógives companies the option to forgo
purchasing and maintaining their own server clusters or expensive,
high-end machines and pay for processing power on a weekly or
monthly basis instead (see
related story). Additionally, vendors including IBM, Sun and
Microsoft are releasing software for companies looking to build
their own grids internally.
What's driving the development of mainstream cluster and grid
computing technologies? Simple: economics. Demand for low-cost
servers has never been stronger, according to IDC. Not only are
companies trimming their overall IT budgets, they're buying smaller
increments of infrastructure capacity. IDC predicts that sale of
servers running Linux will triple to $6.5 billion by 2006; sale of
servers running Windows is projected to reach $19 billion. While
revenues may have remained flat, sales of low-end servers surged in
2002. U.S. server shipments increased by 14%, with HP, Dell and IBM
accounting for more than 60% of the 1.9 million servers shipped,
according to the Gartner Group.
Put simply, IT managers are doing more with less. That means that
low-cost Intel-based servers that can be clustered together to
aggregate computing power are becoming attractive options to
expensive, high-end RISC-based machines. Similarly, renting
computing power from a large vendor has the potential to save
companies thousands in server and maintenance costs. Still,
drawbacks remain. For one thing, managing server clusters is still
complex, and applications not optimized for clustering may actually
drain performance. A big issue for grid computing is security, for
both grid customers and providers. But vendors and IT managers alike
are investing in cluster and grid technologies now, knowing their
payback will be substantial.
The Future of Shared Computing
 |
The idea of
tying computers together via the Internet to take advantage of
spare CPU cycles is so appealing, analysts expect that 10% of
U.S. companies will make use of grid computing by 2005; 40% of
U.S. companies with more than 10,000 employees are expected to
make use of the technology by 2005. |
| As businesses
grow, so does the demand for increased scalability and
availability. At the same time, budget-conscious businesses
need to do more with less. From an IT cost perspective,
clustering servers together to aggregate processing power is
an attractive option to mid-range and high-end server
systems. |
 |
Microsoft Advances Clustering in Windows Server 2003
IT managers were first given the chance to experiment with
clustering two Windows servers together with the release of
Microsoft's NT Server 4.0, but the add-on utility remained more or
less a novelty until the release of Windows 2000, which gave IT
managers a viable utility for improving server fault tolerance. Now,
with the release of Windows Server 2003, Microsoft is finally
getting serious about clustering.
Windows Server 2003, due for release in April, significantly
advances the OS' clustering capabilities, and also makes the
utility-now called Microsoft Cluster Service-an integral part of the
OS for the first time. The integration means IT Managers can now
control security settings for Windows server clusters via Active
Directory, and allows IT managers to view server clusters as single
directory object. Microsoft Cluster Service also increases the
number of nodes that can be linked for failover from two to eight,
and includes a new management utility for simplifying cluster
configuration.
Load balancing, a key component of clustering technology, is also
addressed in the updated version of the OS, but Microsoft's
definition of load balancing differs from that of other vendors.
Typically, load balancing (sometimes referred to as parallel
processing) refers to distributing an application's processing load
among several servers in a cluster, thereby increasing the
application's overall performance. But Microsoft's version of load
balancingóadded as a new utility in Windows Server 2003órefers to
redirecting IP traffic from server to server in a particular order
in response to failure notifications. While useful for maintaining
high availability for overloaded Web servers, for example,
Microsoft's Load Balancer does nothing in the way of distributing
processing tasks among multiple servers.
Still fairly application-specific, Microsoft's Cluster Service
can only be used to ensure failover operations for Exchange and SQL
Server. However, IT managers can use scripting languages such as
Visual Basic and Jscript to program 64-bit applications to take
advantage of Microsoft's load balancing and basic failover
capabilities.
IBM, Gateway Offer Supercomputing On-Demand
Beginning this month, businesses of all sizes will be able to
make use of grid technology with just a phone call.
The first of IBM's "e-Business on Demand" hosting facilities,
built to provide supercomputing services to select industries, will
open in Poughkeepsie, New York. Initially, the grid facility-which
will house massive clusters of IBM servers running Linux-will target
businesses that need access to large-scale computing power on a
short-term basis. Digital movie animators and pharmaceutical
researchers are examples of two such businesses IBM hopes will take
advantage of its computing on-demand services.
While IBM has yet to announce pricing for its grid services, the
company claims companies of all sizes will be able to realize
significant savings in leaving the installation, management and
maintenance of server clusters to it. IBM's first customer is PGS
Data Processing, a division of Petroleum Geo Services which needs
the extra CPU cycles to complete a seismic imaging project. But IBM,
which says it plans to open several more U.S.-based and
international grid facilities-will also target companies in the
financial services, automotive, aerospace, life sciences and
government sectors. IBM is also working with grid middleware
developers Platform Computing and DataSynapse to create software
solutions for enterprises looking to set up their own internal
grids.
By contrast, Gateway is targeting businesses on the smaller end
of the spectrum, offering rentable computing power borrowed from the
thousands of high-performance PCs that sit idle in its stores much
of the time. For a per-hour fee, companies can now tap the combined
computing power of Gateway's 8,000 high-end PCs via the Web. The
service, designed with help from grid developer United Devices, lets
clients submit job requests, get status updates, and pick up results
when job is complete.
Linux Clusters Gain Credibility with Corporate
America
Without a doubt, corporate America is turning to
Linux as an alternative to Unix, and entrusting their
mission-critical operations to clusters of Intel-based systems.
Tapping grid computingówhich makes use of clustering technologyóis
the next step.
More often than not, it's a question of cost. Mid-range and
high-end server systems are not only significantly more expensive
than PC-based servers, they're more costly to manage and maintain.
Faced with IT budgets that have been severely downsized, CIOs are
cutting back on hardware purchases, opting to invest in low-end
systems that typically cost thousands of dollars less than their
RISC-based counterparts and then clustering those low-end servers
together to boost scalability and availability.
E*Trade, Google, and Dell Computer are a few examples of
companies that have replaced proprietary, high-end RISC-based
servers with clusters of Intel-based servers. Last year, E*Trade
replaced 60 Sun Sparc servers with 80 Intel-based Dell servers
running Linux, and expects the move to save the financial services
firm upwards of $13 million. Google claims it runs its search
engines on the world's largest commercial cluster-more than 10,000
servers running Linux. Even Dell Computer has switched 14 of its
internal Sun systems to a cluster of Linux servers for running its
databases.
Research firm Gartner Group projects that Intel servers in 2003
will for the first time generate more revenue than Unix servers, $20
billion, compared with $18 billion. Linux now runs 10% of all
servers, according to IDC. IBM, a big proponent of Linux, says the
open-source OS represents the future of grid computing, which works
on the principle of parallel processingóa form of clustering. Grids
allow companies to either tap external computing power as a utility,
or create their own internal source for balancing processing loads
among server clusters and dramatically decreasing application
processing time.
Charles Schwab is one company that is already combining
clustering and grid computing to reduce the processing time for its
financial apps. Other companies implementing internal grids include
Ford Motor Company, Saab Automobile, Motorola and Sony
Semiconductor. But while closely related, grid computing won't
replace clustering. "Grid computing is great if you're looking for
raw processing power and you don't need the overhead of high
availability features," says Dan Kusnetsky, Vice President of
Systems Software for IDC.
Give us feedback on
this newsletter so we can make it better
Please forward a copy of this newsletter to your colleagues.
To SUBSCRIBE, click here
For ADVERTISING INFORMATION, click here
To view online, click here
President & Publisher
Vince Ridley
vridley@extensionmedia.com
Karen Murray
Vice President, Product Development
kmurray@extensionmedia.com
Editorial Director
Claudia Graziano
cgraziano@extensionmedia.com
Design and Production Manager
Brent Hill
bhill@extensionmedia.com
Sales Director
Shari Weelborg
sweelborg@extensionmedia.com
Shared Computing is an Extension Media service (c)Copyright
2002-2003, Extension Media LLC
.gif){kind=small}
.gif){kind=small}
