Thursday, December 31, 2009

TOP500 v. Gartner (just another opinion)

Top500's 2nd list for 2009 was released on 17th of November, 2009. The Cray Jaguar beat IBM Roadrunner to the top first rank. Jaguar was upgraded to AMD x86_64 Opteron Six Core 2600 MHz (10.4 GFlops) processors for the feat. It runs the Linux kernel. Eka, a machine from India, has gradually dropped to #26 now, from its June 2007 rank of #4. The list is closely followed, and is one of the most critical analyzers of computer hardware and operating system software. The Top500 is being published twice a year since 1993 and the data accumulated over the years reveals many patterns and points to future trends as well. "An Overview of High Performance and Challenges for the Future" talk by Dr. Jack Dongarra, one of the maintainers of the Top500 list, is available at The interesting talk concludes with an indictment of the software development process in vogue:

  • For the last decade or more, the research investment strategy has been overwhelmingly in favour of hardware
  • This strategy needs to be rebalanced - barrier to progress are increasingly on the software side
  • Moreover, the return on investment is more favourable to software
    • Hardware has a half-life measured in years while software has a half-life measured in decades

  • High Performance Ecosystem out of balance
    • Hardware, OS, Compilers, Software, Algorithms, Applications
    • No Moore's Law for software, algorithms and applications

Current software is woefully unable to utilize the hardware developed and available. Software has failed to keep pace with hardware growth and progress.

Why is the software domain lagging behind?

The answers most probably lie in ZL v. Gartner, and more specifically in Gartner. It is not easy to believe that a single corporation could hold a tight grip over a domain as large as the software industry, but it seems to be the most important reason why the field of software has lagged behind while hardware has progressed in predictable ways.

A number of blogs list articles on how Gartner has been feeding news and opinions promoting fud about potential products. A lengthy list is available at Apparently, no serious challenge to the nonchalant and frivolous views of Gartner appear to have been made so far, and ZL v. Gartner might come to save the whole software industry from collapse, shame and disgust.

It is hard to dismiss the complaint made by Dr. Jack Dongarra. Dual core 64 bit machines will turn into multi core multi bit machines in future. Large scale parallelizations are suddenly feasible and economical in terms of power consumption. The disruptive and fast emerging processor technology throws serious challenges to software developers. Whole kernels and applications would have to be re-written, if the new features of the processors are to be utilized. If Gartner is continued to be taken seriously, much of the revenues would only get piped into useless products and would only ultimately leave the field barren and lagging woefully behind the gaint strides in hardware.

Ideally, software should be ready 14 months before the hardware becomes available, based on emerging trends in hardware development. But thanks to Gartner and its chums, software is always at least one decade behind hardware on a regular basis now. This ought to change, and I hope the American legal system takes ZL v. Gartner to its logical end, and saves the software industry from further doom.

The motion to dismiss ZL's plaint also raises some interesting questions with regard to the US Constitution as well. Under the VII Amendment to the US Constitution:

"In suits at common law, where the value in conrtoversy shall exceed twenty dollars, the right of trial by jury shall be preserved, and no fact tried by a jury shall be otherwise re-examined in any court of the United States, than according to the rules of the common law."

Long before the US Constitution and the VII Amendment, the right to trial by jury was well known and recognized:

In Holmes v. Walton (1780), New Jersy employed six man juries in cases of small amounts (six pounds) and twelve man juries in all other cases. Because the property involved had substantial value, Holmes was entitled to a trial by a jury of twlve; but the trial judge, for reasons unknown, allowed him only a six man jury. He contended that he was denied a right to trial by a twelve-member jury. The high court upheld his claim. (The case is best remembered for another reason though. The constitutionality of the State's trading-with-the-enemy Act which provided for trial by a six-man jury was not at issue, but the disaffected citizens of the locality alleged in a petition to the state assembly that the states' supreme court had held the seizure act void. The legislature supported the court, by enacting a new measure providing that in any suit exceeding six pounds, trial by jury meant a trial by a jury of twelve.

In the interesting case of Trevett v. Weeden (1786), John Weeden, a Rhode Island butcher, refused to accept paper bills rather than gold or silver for the meat sold in his market. Under a recent Rhode Island law, the penalty for such refusal was harsh: anyone accused of the cime was given only three days to prepare for trial, no jury was allowed, no appeal from conviction was allowed, and the fine was fixed steeply at 100 pounds. As Weeden's lawyer could not reasonably deny that Weeden had refused the paper scrip, he confronted the law itself. The law under which Weeden stood accused, he argued, was unconstitutional; in enacting it the legislature had overreached its legitimate powers and violated the recognized principle of trial by jury and hence the Superior Court of Rhode Island should nullify it. The judges held that the case was "not congnizable" before them and dismissed the complaint filed by Trevett. Weeden was freed and suffered no penalty despite the fact that he had openly disregarded Rhode Island's paper money law. The legislature summoned the high court judges, as if they were offenders, and demanded an explanation of their reasons for hold an act "unconstitutional", and "unprecedented" judgement that tended to "abolish the legislative authority". Judge David Howell defended judicial review and judicial independence before the legislature, and insisted that the legislature had confused the argument with the judgment. The legislature initiated a motion to remove the judges, but ultimately did not succeed in doing so.

The Ten Pound Act Cases (1787) of that era again point to the vigour with which the right to trial by jury was affirmed. The Ten Pound Act of 1785 allowed justices of peace to try minor civil cases, involving sums less than ten pounds, without juries. The state constutitional guarantee of trial by jury extended to all civil cases except those which juries customarily did not try. New Hampshire practice had previously allowed a justice of the peace to try a case without a jury if the sum amounted to less than two pounds. After the court ruled that the act conflicted with the right to trial by jury, petitions to the state legislature demanded impeachment of the judges. The house, by a 3-1 majority, voted that the act was constitutional, but the judges courageously stood by their initial decision and reaffirmed it in another case. Following the failure of a motion to impeach the judges, the house capitulated and repealed the Ten Pound Act.

Now, after the Constitution of the United States of America came into being and the VII Amendment, the right to trial by jury in cases where the value in controversy shall exceed twenty dollars, shall be preserved. In my humble opinion, the right to trial by jury was hard won, and enshrined in the Constitution by no accident. A motion to dismiss could invariably lie only if a light view of the VII Amendment is allowed to prevail.

Dr. Jack Dongarra predicted the arrival of petaflop supercomputers by about 2010, way back in 1999, on the basis of benchmarking of the top500 since 1993. His concerns regarding the serious crisis in software development is to be taken quite seriously, and the first most valuable opportunity to bring in corrections may start with ZL v. Gartner.

I am happy about ZL v. Gartner for another reason too. My interest in computers could be traced to Dr. Arvind Srinivasan, who is now CTO of ZL Technologies. He was sitting to my left in high school. Maths became fun simply because of Arvind's contagious enthusiasm for it. Solving a problem was important but doing it with speed was often more significant and that also brought in an element of race and fun into equations. Speed defines elegance. He was the fastest in our maths class. I am least surprised that ZL took Gartner to court and would not be surprised if the American legal system, as I learnt about it acts fairly, reasonably and justly either. That should pave way for a more exciting decade down 2010 when software releases are ahead of hardware in the making.