10
SEPTEMBER,
2004
SOA hardware – does Intel's goal hold up?
It goes without saying that the most important elements of an SOA environment
will be the software – both the applications and the management systems
through which they run – but that does not necessarily mean that any
old hardware platform will do. So it is apposite that two recent events
have occurred in the hardware domain that are directly relevant to how SOA’s
grow and are utilised. They are also important because they will, indirectly
at least, play a role in how developers think through building their applications
and how they select the greater range of options that will be available
to them.
The first event was the Intel Developers Forum, at which Prasad Rampalli,
Intel’s director and chief architect, set out the hardware platform
steps that enterprises need to undertake in order to be SOA-ready. The first
step is server consolidation, something that many enterprises have already
started on as a way of containing costs, particularly in support and management
areas. Next up is creating a more modular hardware environment through the
use of blade servers. This we will return to shortly. Step three is the
virtualization of servers and storage, which was one of this year’s
IDF themes, and lastly will come the building of an autonomic infrastructure
which IT managers will handle with what he called `cockpit analytics’.
This now all goes under the name of Intel’s Service Oriented Enterprise
(SOE) initiative, and the company said it is combining elements of mobility,
grid computing and system management into the framework. The company said
the silicon-level controls also provide a basis for new capabilities and
services such as RFID and Voice over IP telephony.
" The way we view SOE is a new generation for compute platforms,"
Deborah Conrad, vice president and co-general manager of Intel's Solutions
Market Development Group, said during an IDF keynote on the subject. "This
is going to require a broad ecosystem leadership of which, Intel is a key
leader."
`Ecosystem’ has, of course, become the buzzword of the day in SOA-land
and it is a goal that actually has some sense and merit attached to it.
But simply claiming rights to assign the word to oneself is not always going
to be enough to make it `true’. One of the key factors about SOA is
that it moves the goalposts quite a bit. A lot of what we assume to be a
`given fact’ about computers, systems design and their relationship
to business management is now coming open to question. In Intel’s
case, the assumption that there is a need for an ecosystem implicitly based
upon the company’s processor and chipset technologies is fundamentally
flawed. In a short sentence, with a service-based approach the choice of
processor technology is no longer relevant. Indeed, it can be said that
the choice of operating system is no longer relevant, either. The point
where the ecosystem needs to be built has moved up a level in systems abstraction.
In the same way that, within an SOA, applications will increasingly need
to be written to integrate with a systems management infrastructure rather
than an operating system, so hardware will need to be designed to integrate
with an interconnect system, and as long as they do that the choice of processor
family will no longer be an important question. This is why the other event,
the announcement from IBM that it is opening up the specification for Blade
servers that fit into its BladeCenter systems, is probably more important
than it seems. The BladeCenter is the racking system into which compatible
Blade servers can slot. But it is more than just a rack, as it contains
power supplies, interconnect management, storage if necessary and a whole
raft of supporting infrastructure that servers require. This has been developed
in collaboration with Intel, and both are maintaining close control over
the technology. Both, however, are more than happy to OEM the system to
other vendors, including deadly rivals such as Dell and HP (OK, in practice,
it’ll probably be Intel that cuts those particular deals).
What they are doing instead is making an open standard of the specification
for building Blades that fit into the BladeCenters. Vendors will be able
to obtain a royalty-free licence and then build any type of Blade server
for which they can see a market. This opens up an interesting possibility
– the chance to build functionally specific servers. If you think
about it there are many available already, with communications tools such
as routers being an obvious example, so why not extend it. The Conformative
XML accelerator (see issue dated 23-4-04) is another case in point. But
there is no reason why this approach could be extended to produce Blade
servers that are dedicated to performing a specific business process. As
with the Conformative system (which uses a purpose-designed ASIC-based parallel
processor that is effectively a `grid’ of 16 processors on a single
chip) they could then use whatever processor technology is most appropriate
to performing the required business process, so long as the system integrated
with both the hardware and the systems and applications management infrastructure.
So is this a blatant attempt at profiteering and ripping-off customers by
IBM and Intel? Well, there is possibly an element of that – they’re
in business, after all – but there is also some sense to IBM’s
argument that making it an open standard rather than open source means that
there is some on-going certainty about the nature of the infrastructure.
That is why the pair are going open standard rather than open source –
where anyone could start changing the design of the system. The one thing
SOA is going to depend on is some standards and some certainty, and that
means working with fixed standards whether they come from a vendor or an
industry committee.
www.ibm.com
www.intel.com
|
|
