Mad Dog 21/21: Vulture Culture
October 18, 2004 Hesh Wiener
Of all birds, vultures may have the worst public relations. Their unglamorous image, coupled with looks only another vulture could love, threatens to kill off a few Asian varieties, and to take along with them plenty of people, very likely through the painful means of rabies. Computing environments and application suites perform similar tasks, picking legacy applications and data from the bones of obsolete systems so users can move ahead. The value of backward compatibility is widely underestimated until it is withdrawn, trapping those who had counted on continuity. In both natural environments and computing environments, the potential for systemic disaster can lie hidden for a considerable time. Then, all of a sudden, there is a crisis. The problem and its consequences are exposed. For the vultures of South Asia, life was grim business-as-usual until a few years ago. That was when an anti-inflammatory drug called Diclofenac (which is also sold under the trade names Cataflam and Voltaren) gained popularity among Asians who wanted to give their working cattle and bullocks relief from arthritis and other inflammatory ailments. The drug worked as expected, but when cattle fed the pharmaceutical eventually cast off their bovine coils, an entire ecosystem was pushed to the breaking point. In South Asia, dead cattle are often left for the vultures. Once the birds get to work, they can strip a carcass to bones very quickly. If there are enough vultures around, that can mean less than half an hour. Unfortunately, Diclofenac is deadly to vultures, even in the small residual amounts present in the bodies of dead cattle treated with the drug. The immediate result has been a catastrophic decline in the population of a number of species of vulture formerly common in South Asia. Moreover, what has been a disaster for the birds has turned into a disaster for people, too. Without vultures to perform their job, the dead cattle are left to other creatures, particularly feral dogs. Rabies is common and, perhaps, endemic among these canines. Once infected, the dogs become a vector for the disease, putting not only their own species but also other animals and humans at risk.
Cattle in South Asia are plentiful, and dead cattle have become sufficiently abundant to nourish feral dogs so well that there has been an explosion in the population of the canines. This has amplified the risk of contracting rabies, particularly in rural areas where cattle and feral dogs are more numerous and where medical care is scarce. When IBM brought out its current generation of mainframes, the two families called the z890 and z990, it told customers that the new hardware would provide a number of additional features but that it also had limitations. Customers who moved to these machines would no longer be able to run certain older programs. Even though IBM mainframes can concurrently support multiple operating systems and even multiple versions of the same operating system, the z890 and z990 product lines will not run software that is more than a couple of generations old. In mainframe country, customers in some cases have applications that are several years or even decades old. These applications work just fine, even if they were built to run in an old operating environment. The source code for many of these programs has long since been lost. Documentation, if it ever existed, is often missing, too. Making matters even more complicated, programs often harbor patches written in assembly language, because it’s efficient; mainframes have always been so expensive that every instruction counts. Until the z890 and z990 came out, IBM mainframes did a very good job of supporting code dating from the 1970s or 1980s, including all the patches that helped the mainframes deliver better value. Now mainframe shops, which support the largest and arguably the most heterogeneous collections of software in computing, are having flashbacks of the year-2000 nightmare. Some old instructions no longer work, and some old environments are no longer acceptable. In some ways, the mainframe situation is worse than the Y2K mess. As 2000 approached, everyone in computing had an idea of what might go wrong. There were tools to help spot software or data that might not clear the century hurdle. Vendors and consultants were primed to help users who didn’t think they could manage the situation on their own. One thing the reduction in backward compatibility has in common with the Y2K issue is that it can have the biggest impact on the shops that are least able to cope. The big shops, with glass houses full of mainframes and deep pools of technical talent, can deal with problems as they crop up. The smaller shops running old machines that are at the low end of the mainframe power range no longer have the big budgets for personnel they might have had in the past, when their staffers wrote those assembly code patches and got that extra few percent out of the costly big iron. All these shops will find solutions, of course, but they might not always be the ones IBM’s mainframe marketing people would hope for. Some may defer upgrades and look for ways to keep older machines running. Some may decide it’s time to move some work (or possibly all the work) to a less costly platform, one that doesn’t press users to worry about getting 99 percent utilization and then some. Some will just pull the plug and go to service bureaus, and not necessarily to IBM’s own Global Services. On the other hand, some shops will buy or write applications to replace the ones that can’t be brought forward, and this time they might be a lot more careful about documentation and source code libraries. Or, human nature being what it is, the shops that catch up may well resume their old habits: few of the staffers are going to worry whether the same kind of problem will recur a few years from now. The people running the mainframe shops that face big, if temporary, bulges in their budgets will, for the most part, ride out the story. The ones that have incurred the wrath of corporate managers and top-level bean counters won’t know for sure how much trouble they are in until the forward migration is complete. Corporations that come to believe their computing departments are not being run by the right people will eat their revenge cold in the interest of continuity and to obtain the long-term benefits of having all employees living with perpetual fear of what their bosses are really thinking. In some corporations, the vultures are always circling. For the most part, vultures are welcome in South Asia, or ought to be. The possibility that some species of South Asian vultures will become extinct is of interest not only to ornithologists and villagers who don’t want their kids bitten by rabid dogs. It is a major issue among the Parsees. These people, with roots in ancient and approximately eponymous Persia, are Zoroastrians with a significant population in and around Mumbai. If they didn’t invent monotheism, they are at least in the running, having swapped ideas with the Jews for quite a while a few thousand years ago. Their faith centers on the struggle between good and evil and in a symbolic way Parsees equate good with light. The force of good is called Ahura Mazda, and in the West, Mazda persists as the name for a particular type of light bulb socket, while in the East it is the name of a family of cars. Go figure.
Anyway, in Parsee tradition you don’t bury the dead but instead place the departed at the top of special towers where nature, in the form of vultures, will do the undertaking, except for the bones. As the vultures of South Asia die out, so, too, must the Parsee funeral ceremony. It’s no wonder, then, that Parsees are one voting block in India who want the government to put more effort into vulture breeding projects and to ban the use of Diclofenac. Arthritic Parsees medicated with dicofenac would be, when dead, as toxic to the vultures as cattle that have been fed the drug. It’s hard to say how things will turn out, but there is definitely more hope for the vultures now than there was only a little while ago. Quite a few people around the world are aware of the Diclofenac issue and while it’s possible the rescue effort will be too little, too late, until pretty recently the vultures didn’t stand any chance at all. The birds probably have a better chance of surviving than do mainframe shops with less than 200 MIPS of processing power. They might also have better prospects than the Intel Itanium. The 64-bit Itanium does a terrible job of supporting 32-bit software. That’s the way Intel planned it. The chipmaker believed that a new processor family would gain more adherents if it did more 64-bit work for less money, something that would not have been the case if the chip included powerful 32-bit capability along with its 64-bit technology. For this reason and others, industry wags quickly dubbed the circuit Itanic, and the joke has turned out to be sadly prophetic. (To be specific, the X86 emulation layer in the Itanium processor has a 50 percent overhead. That means a 1.5 GHz Itanium, which on ported code has about the same performance as a 32-bit version of the software running on a 3 GHz Xeon processors, runs that 32-bit code as if it were running on a 1.5 GHz Xeon. This is an unacceptable performance penalty for all but the smallest workloads.)
When Intel stumbled getting Itaniums out the door and into servers, Advanced Micro Devices developed its Opteron line, which might not have all the 64-bit capabilities of the Itanium but which does a bang-up job running IA-32 code and provides good performance on code extended to 64-bits. Opteron is killing Itanium in the workstation segment, where users run a mix of 32-bit and 64-bit applications and Intel has been forced to follow AMD by cooking up the Xeon Nocona, a 32/64-bit circuit. If Intel wants to stay on top of the game, it is going to have to merge the Nocona and Itanium lines or, perhaps, just let the Itanium go the way the vultures will if Asian cattle don’t get some new anti-inflammatory medicines. Because IBM has been very careful with binary compatibility with OS/400 servers–made possible through a sophisticated hardware abstraction layer that allows IBM to radically change technologies underneath OS/400–AS/400 and iSeries shops are not facing these problems as they contemplate porting their applications to i5/OS and the new Power5-based eServer i5 machines. IBM did the right thing in 1988, when it decided to not let OS/400 get too close to the iron. But IBM had not done this with its mainframes. However, IBM may get away with diminished legacy support in the mainframe business because it is the only company making mainframes. If Amdahl or Hitachi had stayed in the mainframe game, they would now have the kind of opening Intel foolishly gave AMD. But that’s not the case. In computing, if not in nature, the vultures are doing just fine. |