April 16, 2009
If there is any information technology problem more intractable than parallel programming, it's the data deluge problem. In most cases, demand for storage is growing even faster than demand for more compute cycles, and storage boxes are increasingly taking up a larger share of the budget and real estate in the datacenter. Good news for storage vendors.
Or maybe not. Data by itself is worthless. Here I'm not talking about content destined for entertainment, like movies or music; I'm referring to those special bytes destined to be processed by computers. For the latter category, the value is in the knowledge that can be extracted from the data. The only reason to store it in the first place is that you're not prepared to process it yet, either because you don't have the time or resources in place or because other information needs to arrive before the original data can be digested.
But embracing the data deluge in the moment may be answer. Stream computing, which processes data on the fly, short-circuits the traditional workflow of storing data, reading it, processing it, and delivering the results. It's basically data processing without the database. This model is especially useful for complex event processing (CEP), which Neal Leavitt wrote about (PDF) in the April issue of IEEE's Computer. CEP software pipes one or more data sources to an analysis engine, so it intrinsically relies on the stream model to do its job. Writes Leavitt:
Vendors build CEP systems so that they initiate processing in response to inbound events, which enables them to function in real time. Traditional event-processing systems wait for a user action to initiate processing.
In fact, it is the real-time response, rather than the simplification of the storage layer, that is usually cited as the most valuable feature of these systems. That's why CEP turns out to be a very attractive platform for applications like options pricing or terrorist threat detection. Although the idea of CEP has been around for a while, general-purpose platforms are just starting to emerge. Software providers like Truviso, Aleri, Realtime Monitoring GmbH, and Event Zero are some of the early players.
Leavitt's research led him to Aite Group, which predicted that CEP products will "more than double from $180 million in 2008 to $370 million this year and increase to $460 million in 2010." He believes two advancements that are helping push CEP into the mainstream: the proliferation of real-time feeds and the modularization of IT infrastructure. To that I would also add the increase in computational performance of hardware. And this is where high performance computing could turn out to be a key enabling technology for such systems.
IBM recently unveiled a prototype of its stream computing technology by porting the company's InfoSphere Streams analytics software to the Blue Gene/P supercomputer. IBM has been working on a stream technology for several years, and started talking about it publicly just a couple of years ago under the name of System S (PDF). According to IBM, "It has some similarity to CEP systems, but it is built to support higher data rates and a broader spectrum of input data modalities. It also provides significant infrastructure support to address needs for scalability and dynamic adaptability such as scheduling, load balancing, and high availability."
The IBM prototype announced this month was done in conjunction with TD Securities to demonstrate what Big Blue thinks could be the model for next-generation trading systems. The press release quotes Nagui Halim, chief scientist of the Stream Computing Project at IBM: "TD Securities could potentially use the new system to analyze and act on information before their competitors can finish ingesting and analyzing, effectively blinding the competition to its actions. We're not talking about 20 percent faster here. We're talking about 20 times faster."
Of course, if this works as advertised, all the big financial firms will buy such a system. The real advantage to organizations comes from combining lots of streams of data in novel ways and asking unique questions. For example, pouring news feeds, weather reports, and maybe even proprietary data streams into the analytic soup could offer a way for companies to a gain a real edge on their competition.
In the meantime, it will be interesting to see how this technology makes its way into the world. As much as everyone complains about being inundated with information 24/7, the average person has come to accept it, and the movers and shakers have come to exploit it. In such an environment, buying more hardware to analyze yesterday's data may be a losing proposition.
Posted by Michael Feldman - April 16, 2009 @ 4:20 PM, Pacific Daylight Time
Michael Feldman is the editor of HPCwire.
No Recent Blog Comments
10/30/2013 | Cray, DDN, Mellanox, NetApp, ScaleMP, Supermicro, Xyratex | Creating data is easy… the challenge is getting it to the right place to make use of it. This paper discusses fresh solutions that can directly increase I/O efficiency, and the applications of these solutions to current, and new technology infrastructures.
10/01/2013 | IBM | A new trend is developing in the HPC space that is also affecting enterprise computing productivity with the arrival of “ultra-dense” hyper-scale servers.
Ken Claffey, SVP and General Manager at Xyratex, presents ClusterStor at the Vendor Showdown at ISC13 in Leipzig, Germany.
Join HPCwire Editor Nicole Hemsoth and Dr. David Bader from Georgia Tech as they take center stage on opening night at Atlanta's first Big Data Kick Off Week, filmed in front of a live audience. Nicole and David look at the evolution of HPC, today's big data challenges, discuss real world solutions, and reveal their predictions. Exactly what does the future holds for HPC?