Roy Schestowitz wrote:
> Smaller businesses make most use of open source
> ,----[ Quote ]
> | Six out of 10 Oracle database sites are now running open source operating
> | systems, but adoption is lower among big businesses, new research has
> | revealed.
> `----
>http://www.computerweekly.com/Articles/2006/08/14/217700/Smaller+businesses+make+most+use+of+open+source.htm
This makes perfect sense. If I only have a few hundred gigabytes of
storage to manage, and I can use fireware, direct-attached SCSI, or
USB-2 drives, or a small SAN array, then Linux handles the job pretty
well, and there isn't much need for anything much bigger. Oracle can
deliver the load from Linux and get very good peformance numbers.
On the other hand, if I have a few hundred terabytes (a terabyte is
1024 gigabytes) or a few petabytes (a petabyte is 1 million gigabytes),
then it's quite likely that you will want a high end AIX, Solaris, or
HP_UX machine such as a Regatta, StarFire, or Superdome which can not
only handle the load, but has layers of built-in redundancy and load
balancing capabilities which are critical to SQL database servers.
A common technique these days is to have middleware manage clusters of
database servers, but again, there are distinct advantages to having a
ultra-high-speed SAN on a machine which has a huge number of
ultra-high-speed FiberChannel or 10 Gigabyte ethernet ports directly
and exclusively to the back-end storage arrays, such as most high end
UNIX machines. In addition, the ability to dynamically allocate
virtual machines such as AIX LPARs, (Solaris and HP_UX have their own
terminoligy for VMs), tends to make purchasing one or two co-located
machines more desirable than purchasing dozens of individual
4-processor SMP servers.
Microsoft isn't even in the ball-bark in these entirprise database
server markets.
Ironically, as the UNIX vendors increase the number of processors and
cores per server (a P590 has up to 128 cores at 2.5 Ghz each), and the
clock speeds, directly addressable memory, and interconnection speeds
increase, the UNIX server counts drop, along with the costs of
implementation and production. As a result, UNIX appears to be losing
market share when measured by number of servers and revenue, but in
reality, it's just that UNIX TCO has been dropping radically due to
"On-Demand" or other "Pay for what you use" approaches have become
increasingly popular.
I remember back in the old days when we needed 3-4 separate servers,
one for development, one for test, one for production, and one for
fail-over. These days, we can often just configure separate LPARs for
dev, test, and production, and fail-over is simply a matter of having
redundant copies of the image (because the processors and remote sites
cover each other).
Linux is frequently used to "Front-End" this information. The apache
servers, the gateways, the middleware, and the applicaton logic, are
now often stored directly on Linux servers, frequently on blade arrays,
and the applications are designed to be clustered using cluster
technology such as that used in Beowulf, and/or SOA technology.
Linux supports MPI, PVM, CORBA, SOAP, Java RMI, WSDL, and SOA.
|
|
