Oracle would like you to jump into
“the grid” and buy into all that it offers. What is not clear immediately is the
complexity of these systems, nor what they offer. The central piece of “the
grid” in this new version of database software is failed technology,
reworked, repackaged, definitely renamed from many years of earlier
versions. It requires database
administrators to have experience that most do not have, and most cannot
get. It remains expensive.
The cost of the required Real Application
Clusters (RAC) license is brushed aside in general conversation. More than casual conversation with an
Oracle representative on the subject and you will find offers of “standard
edition” vs. “enterprise edition” with RAC “included
free”.
There are also a couple of offerings from
Dell that include hardware with up to four CPUs, software, licenses and
consulting services. Beyond that
you must negotiate for an expensive RAC license.
A practical “grid” would not be
restricted to four CPUs. The
biggest benefits in the scientific/computational area using Oracle’s
solution have been shown using 300 +/- nodes doing heavy parallel jobs on
otherwise largely under-utilized hosts.
So what is “the grid” anyway?
Carl
Kesselman1 and Ian Foster2 literally wrote the book on
the grid as it is envisioned in most research circles, including the San Diego
Supercomputer Center (SDSC).
What Is
the Grid? See Ian Foster’s
short definition from 2002,
here.
Who is using “the
grid” as defined by Kesselman and Foster?
Universities,
facilities like the SDSC whose mission statements include “to develop and
use technology to advance science”3 are using the grid.
There is
also development taking place where money is being spent on research. In
There is a convoluted interconnection
between these disciplines and the universities and businesses, and related legal
and security requirements.
Complex rules about who may do what, at
which location, and with which data have to be in place to allow access without
hindering resources within the grid.
Rules for sharing are probably not much
different than those required by law enforcement, government, military…
and even private organizations today considering the current threats of
terrorism.
The academically defined grid (“the
grid”) consists of environments custom built to meet specific goals of
the target organizations.
"It takes a degree in computer science
to understand today's parallel computers and several more years of training to
write code to use them efficiently."4
If you are considering “the
grid” do your homework. Know
why you are doing it.
But my vendor says its product is “the grid”.
The big
industry players seized on “the grid” as a new marketing
opportunity and turned it into a new term for “actively managed
clusters”.
Oracle
has RAC; IBM has On-Demand Computing, and HP has Utility Computing. In fact, HP
is reworking their entire storage line to provide Smart Cells of grid storage using the Lustre parallel file system.
Regardless
of the fancy bells and whistles involved these guys are all just talking about
clusters on a LAN. That's why
Oracle started that breakaway GT'04 conference from the Global Grid Forum
(GGF).
IBM does
understand WAN grids and is working actively on helping merge standards for
grid services and web services.
However, at the end of the day, only Sun truly embraces the WAN version
of grids and provides tools for it.
Back to “the
Grid”… is it for me?
All of
this may sound like “fun” to a techie. There’s a place for models and
standards battles, beta software, and all that. If you’re reading this I presume that
isn’t what you want to be doing with your business application or
development project.
There
are two cases to be made for grid-computing in the cluster sense of Oracle and
their competitors - uptime and dynamic
capacity balancing.
Uptime
is pretty obvious since you're just talking about moving the sessions from a
dead box to a live one in the cluster.
24/7 operation comes at a cost.
How much uptime does the user need?
How much can the user afford?
Dynamic capacity balancing is far more
interesting since you can take a pool of ten servers and during your monthly
sales campaign convert an underutilized accounting server to be an extra web or
transactional server to handle that load.
When it's time to run payroll, you redeploy a web server into accounting.
It's a powerful concept and they actually have some very slick tools for this.
A company would need to invest in a capacity
survey and planning exercise to really get full benefit from this level of
balancing. Most firms have no idea
what utilization their servers are at during various periods. Many firms intentionally overbuy
hardware only to later push to take it out of service because of power and
cooling requirements in the data center.
Just as the licensing fees need to be
adjusted for “the grid” the fees charged to cost centers by the IT
department will have to be considered.
There’s got to be some trick in the software to track costs, and
in an environment that already includes so many security issues and access
rules this should not be a problem to calculate.
Do we need Oracle10g?
As a rule the response should be “it
depends”. Are you currently
experiencing problems that appear to be fixed in a new release of Oracle? Are there other problems that might be
fixed by installing, removing, or re-installing software? Is your product about to expire? Are there upgrades to your applications
that need to be or can’t be upgraded? The list goes on.
If you are already creating, customizing,
and using software against an Oracle database your database administration team
should always be looking at new versions.
There are new features, tools, controls, performance and maintenance
improvements in 10g that do not require RAC.
An article to follow will discuss what you
can do with Oracle10g without “the grid”, and when your
requirements are for actively managed clusters, there are some new cost savings to
be had with Oracle10g features and aftermarket storage hardware.
1
Carl Kesselman, director,
Center for Grid Technologies, USC,
2
Ian Foster, professor of
computer science at Argonne National Laboratory and the
3
Fran Berman, director of the
4
Kennedy, director
Rice’s Center for High Performance Software Research
Last Revised: April 2007