Doctor Recommended
100% Natural

Botnets are basically an automated collection of zombie computers. The term zombie computer comes from the fact that a virus, trojan, or some kind of malware has compromised the security of the system and it in turn can be used remotely for nefarious purposes (like sending massive amounts of spam).

Unfortunately the impression most non-technical people have of infected computers is that diseased computers are isolated islands that only ruin your TPS reports and latest budget outlays.

The truth of the matter is that millions of these zombied computers are actually controlled by a select few hacker groups. And interestingly enough, a gigantic multi-billion dollar black market has emerged from this “organized chaos.” The perpetrators sell and siphon off resources to the highest bidder, who typically utilize the systems to hawk a slew of wares across the unsuspecting interweb. Worse yet, some of the organizations that manage the botnets use their collective bandwidth capacity to blackmail organizations into submission, or else the target(s) will suffer a massive DoS.

If you are familiar with distributed or grid computing, this next info nugget is not at all surprising: the aggregate power of these botnets dwarfs the fastest of all supercomputers and then some.

And to add more drama to this situation, botnet owners not only battle one another for more systems (by hacking and infecting one another), but their ever-evolving strains of viruses are designed to counterattack anti-cracker organizations and endeavors.

For instance, if I used my computer to sniff and trace suspected botnets, upon being detected the botnet would retaliate with a massive DoS that would effectively end my investigation. Thus, it is typically very difficult to find and prosecute those directly responsible for the torrential flood of spam that propagates across the internet (the latest Storm war alone has resulted in a double-digit growth in spam this past summer).

So, how exactly do GPGPUs fit into this equation?

Well, the common thread is that botnets and GPGPUs are both effective at what they do due in large part because of parallelization.

Again, the prevailing non-technical view of video cards and processors is that there is one (1) core underneath all of the glossing packaging. However, this is not the reality of the situation.

One of the reasons why supercomputing firms such as Cray were able to calculate, transform and manipulate data at mind blowing rates is that their innovative systems employed the use of parallelization to maximize throughput. Systems developed by Thinking Machines and Cray used thousands of processors to accomplish feats that one computer or hundreds of independent computers could not do alone. It is this collective cooperation that efficiency is achieved, and is in many ways similar to how the human brain works (massive parallelization).

And for a number of reasons – chief of which is energy consumption and heat dissipation – many semiconductor firms have begun cramming multiple CPU cores into each physical processor. For instance, the Core Duo from Intel has two CPU cores attached at their silicon hips; and for nearly a year Intel has sold quad-core versions as well. AMD, IBM and others have also sold dual and quad-core varieties of their CPUs (the last PowerPC used by Apple was a dual-core G5 from IBM and as of last month Sun began selling an octo-core).

As a result many programs have to be rewritten to take advantage of this mutli-brain innovation — it is a detail oriented task that few programmers are proficient at (although help is coming in the form of toolkits from both AMD and Intel as well as the HPC community).

Putting the U back into Bungholio Marks

With the advent of stream computing and unified shaders, GPGPU is now the name of the brains found on newer video cards. And while the tasks of these video cards have traditionally been pigeon-holed to a select few tasks, they are now capable of sequencing proteins, accelerate anti-virus software, and a slew of other resource intensive tasks that can be quickly surmounted if processed in parallel.

Over the past five years, in addition to the traditional “main” core, consumer-grade video cards have all come with multiple shader “cores,” and by cores I don’t mean anything nearly as big as a fruitfly. For instance, nVidia’s latest and greatest 8800 Ultra card, comes with 128 shader “cores” that can be utilized to perform calculations in parallel. ATI’s newest 2900XT comes with 320. Even Intel’s latest integrated chipset (GMA X3500) includes 16 shader units.

And the world of desktops and laptops is not the only place you will find these products. In the latest battle between video game consoles, the much ballyhooed Cell processor found in the Playstation 3 has numerous “cores” including 8 SPEs; the XBox 360 has many different types as well, including 48 shader units.

And for the same evolutionary reasons all mammalian brains remained parallel in nature (redundancy, multi-taskable, scalable) this trend will probably never abate any time in our lifetime. (Note: SLI and SIMD are different topics).

No one snowflake is responsible for the entire avalanche

So while, one ALU, APU, or even CPU may seem relatively powerless in the scheme of things, hundreds and thousands of them working in tandem can provide a powerful toolset for programmers to manipulate. The same can be said for a compromised computer system. Its loner status is nothing compared to the aggregated power of thousands working in conjunction.

See also:
The Commercial Malware Industry
What’s wrong with Moore’s Law?
FLOPS, MIPS, Watts and the Human Brain
Seth Lloyd’s Million Megahertz CPU
Specialization, Centralization, and the Future of Chip Integration