The Changing world of InfiniBand
The Changing world of InfiniBand
From STORAGE Magazine
Vol 7, Issue 8 - November/December 2007
infiniband is a low-latency, high-bandwidth cluster interconnect - and it’s time
to shine has come, says LAURA MANNING, PR EXECUTIVE, HAMMER PLC
InfiniBand is a buzzword that has been circulating within the channel a great
deal in recent years.
A low-latency, high-bandwidth cluster interconnect - as well as a mature and
field-proven technology - it allows for greater bandwidth and the elimination of
bottlenecks. Infiniband requires a low processing overhead and is seen as the
ideal format to carry multiple traffic formats. The hype surrounding Infiniband
is seemingly justified, but where does its future lie?
Established in high-performance niche markets, InfiniBand's reputation has
recently been nothing, if not high profile. Its origins can be traced back to
1999, where two competing input/output standards merged to create a unified
standard that became known as 'InfiniBand'.
After a brief dip in the economy and the 9/11 disaster, among other aspects,
InfiniBand seems to have regained its footing after a lot of its suppliers
seemed to abandon their efforts. Offering throughput of up to 2.5GB per second,
InfiniBand has unlimited expandability, which is expected gradually to replace
existing shared-format approaches for servers and computers of varying levels.
In today's market, its uses include aiding complex seabed mapping activities,
detailed financial dealings and meticulous biochemical research, as well as in
the intricate functioning and precise control of Formula One motorcars. This
calibre of technological capabilities can only spell positive things for the
technologies of the future that are yet to benefit from InfiniBand's strengths
and advantages. Its systems have been proving they can perform at levels
exceeding expensive proprietary embedded systems and even the traditional
supercomputers - and at a fraction of the cost.
So where is InfiniBand likely to progress in the coming years? From high-end
servers, it has been predicted it will move into interprocessor communication,
as an interconnection for blade-servers and possibly even as a storage connect
interface, which all have potential InfinBand uses.
"Any industry where the response time to get results from IT infrastructure
outweighs the cost of the infrastructure itself is an opportunity for Infiniband,"
says Paul Hickingbotham, solutions sales manager at dedicated storage
distributor Hammer. "Whereas competing technologies, such as Ethernet, can
initially provide a lower cost outlay for clustering, InfiniBand is uniquely
scalable, due to its low latency, enabling clusters to stretch beyond previously
defined scalability barriers."
Phil Murphy, senior director of platform engineering, QLogic Network Solutions
Group, agrees. "InfiniBand has established a leadership position in HPC cluster
interconnects, as evidenced by the rapid growth of the technology in the
industry's closely monitored Top500.org, and we are now seeing an increase in
the adoption of InfiniBand-based solutions by enterprise customers for a variety
of applications, including Oracle RAC, media and entertainment, manufact- uring,
and oil and gas.
“Much of InfiniBand's success in these environments,” he adds, “can be
attributed to the mature management software that now complements the
technology's excellent latency and bandwidth characteristics."
Why then is such an innovative and ground-breaking technology still slowly
emerging? "InfiniBand is evolving into different markets at various stages of
its life," confirms Hickingbotham.
"The future for InfiniBand is not necessarily constrained by its previous market
positioning within HPC. Changes can already be tracked in its progression into
mainstream applications and this will likely be true to a greater extent in the
future.”
By addressing key technological demands that not only the channel, but IT in
general, face - and will be facing in increasing amounts in the future -
InfiniBand undoubtedly has the answers for any issue that arises.
Or does it? Does its high-end components and experience really fit the bill for
everyday computer and server specifications? Its gradual initiation into such
markets may mean that it is overqualified and practised in the wrong areas for
such public and ubiquitous use.
"Whilst it is true that the use of Infiniband as an interconnect technology will
decrease latency, offer greater scalability of the Cluster (which is of even
greater interest with the explosion of virtualised server farms) and simplify
cabling complexities, these benefits alone do not mean Infiniband will be a
prolific technology and is certainly not for everyone. It requires calculated
design considerations to make sure the benefits it provides offer a real return
on investment and justify its use."
One thing's for certain: the many advantages of InfiniBand's config- uration are
argument enough to fight its corner. With undeniable performance capabilities -
such as reliable inter-fabric formats that allow for continued performance, in
the event of a fault; efficient scalability of multiple systems; and end-to-end
data connections which are extremely reliable - InfiniBand seems to have so much
to offer.
With its lengthy introduction to the market, InfiniBand seems to be waiting for
its time to shine in the multiplicity of applications and contexts that this
technology would seem to merit.
Whether that time really is now remains to be seen, but one thing is clear: when
it does fully break into the commercial market, it will do so in a way that is
likely to change the manner in which our computers, servers and IT in general
have ever been used to functioning before. ST
|