Slashdot Mirror
IT Snake Oil — Six Tech Cure-Alls That Went Bunk
snydeq writes "InfoWorld's Dan Tynan surveys six 'transformational' tech-panacea sales pitches that have left egg on at least some IT department faces. Billed with legendary promises, each of the six technologies — five old, one new — has earned the dubious distinction of being the hype king of its respective era, falling far short of legendary promises. Consultant greed, analyst oversight, dirty vendor tricks — 'the one thing you can count on in the land of IT is a slick vendor presentation and a whole lot of hype. Eras shift, technologies change, but the sales pitch always sounds eerily familiar. In virtually every decade there's at least one transformational technology that promises to revolutionize the enterprise, slash operational costs, reduce capital expenditures, align your IT initiatives with your core business practices, boost employee productivity, and leave your breath clean and minty fresh.' Today, cloud computing, virtualization, and tablet PCs are vying for the hype crown." What other horrible hype stories do some of our seasoned vets have?
IT snake oil: Six tech cure-alls that went bunk
By Dan Tynan
Created 2009-11-02 03:00AM
Today, cloud computing [4], virtualization [5], and tablet PCs [6] are vying for the hype crown. At this point it's impossible to tell which claims will bear fruit, and which will fall to the earth and rot.
[...]
1. Artificial intelligence
2. Computer-aided software engineering (CASE)
3. Thin clients
4. ERP systems
5. B-to-b marketplaces
6. Enterprise social media
1. AI: Has to have existed before it can be "bunk"
2. CASE: Regarding Wikipedia, it seems to be alive and kicking.
3. Thin Clients: Tell that to the guys over at TiVo that thin-client set-top-boxes are bunk.
4. ERP Systems: For low complexity companies, I don't see why ERP software isn't possible.
5. Web B2B: He is right about this one.
6. Social media: Big companies like IBM have been doing "social media" within their organization for quite some time.It's just a new name for an old practice
And as far as his first comment,
"Today, cloud computing [4], virtualization [5], and tablet PCs [6] are vying for the hype crown. At this point it's impossible to tell which claims will bear fruit, and which will fall to the earth and rot."
[4] Google.
[5] Data Servers.
[6] eBooks and medical applications.
The people actually putting artificial intelligence into practice knew that AI, like so many other things, would benefit us in small steps.
Actually, there was a period very early on ('50s) when it was naively thought that "we'll have thinking machines within five years!" That's a paraphrase from a now-hilarious film reel interview with an MIT prof from the early 1950's. A film reel which was shown as the first thing in my graduate level AI class, I might add. Sadly, I no longer have the reference to this clip.
One major lesson was that there's an error in thinking "surely solving hard problem X must mean we've achieved artificial intelligence." As each of these problems fell (a computer passing the freshman calc exam at MIT, a computer beating a chess grandmaster, and many others), we realized that the solutions were simply due to understanding the problem and designing appropriate algorithms and/or hardware.
The other lesson from that first day of AI class was that the above properties made AI into the incredible shrinking discipline: each of its successes weren't recognized as "intelligence", but often did spawn entire new disciplines of powerful problem solving that are used everywhere today. So "AI" research gets no credit, even though its researchers have made great strides for computing in general.
This is a bit OT but I wanted to say that snydeq deserves a cookie for linking to the print version. I can only imagine that the regular version is at least seven pages. I hope slashdot finds a way to reward considerate contributors such as him or her for making things easy for the rest of us.
Cory Doctorow talking about cloud computing makes as much sense as George W Bush talking about electrical engineering.
I don't know of a single IT department that hasn't been helped by virtualization of servers. It makes more efficient use of purchased hardware, keeps businesses from some of the manipulations to which their hardware and OS vendors can subject them, and is (in the long term) cheaper to operate than a traditional datacenter. IT departments have wondered for a long time: "if I have all this processing power, memory, and storage, why can't I use all of it?" Virtualization answers that question, and does it in an elegant way, so I don't consider it snake oil.
---don't make me break out my red pen.
I kind of miss the crazy hotties that used to pervade the network sales arena. I won't even name the worst offenders, although the worst started with the word cable. They would go to job fairs and hire the hottest birds, put them in the shortest shirts and low cut blouses, usually white with black push-up bras - and send them in to sell you switches.
It was like watching the cast of a porn film come visit. Complete with the sleazebag regional manager, some of them even had gold chains on. Pimps up, big daddy!
They would laugh at whatever the customer said wildly, even if it wasn't really funny. The girls would bat their eyelashes and drop pencils. It was so ridiculous it was funny, it was like a real life comedy show skit.
I wonder how much skimming went on in those days. Bogus purchase orders, fake invoices. Slap and tickle. The WORST was if your company had no money to afford any of the infratsructure and the networking company would get their "capital finance" team involved. Some really seedy slimy stuff went down in the dot-com boom. And not just down pantlegs, either.
Having taken several courses on AI, I never found a contributor to the field that promised it to be the silver bullet -- or even remotely comparable to the human mind.
Not today, after the "AI Winter". But when I went through Stanford CS in the 1980s, there were indeed faculty members proclaiming in print that strong AI was going to result from expert systems Real Soon Now. Feigenbaum was probably the worst offender. His 1984 book, The Fifth Generation (available for $0.01 through Amazon.com) is particularly embarrassing. Expert systems don't really do all that much. They're basically a way to encode troubleshooting books in a machine-processable way. What you put in is what you get out.
Machine learning, though, has made progress in recent years. There's now some decent theory underneath. Neural nets, simulated annealing, and similar ad-hoc algorithms have been subsumed into machine learning algorithms with solid statistics underneath. Strong AI remains a long way off.
Compute power doesn't seem to be the problem. Moravec's classic chart indicates that today, enough compute power to do a brain should only cost about $1 million. There are plenty of server farms with more compute power and far more storage than the human brain. A terabyte drive is now only $199, after all.
you've got a straight road to expensive failure.
Sing it brother (or sister)! As one who is currently helping to support an Oracle-based ERP project, expensive doesn't begin to describe how much it's costing us. Original estimated cost: $20 million. Last known official number I heard for current cost: $46 million. I'm sure that number is over $50 million by now.
But wait, there's more. We bought an off-the-shelf portion of their product and of course have to shoe-horn it to do what we want. There are portions of our home-grown process that aren't yet implemented and probably won't be implemented for several more months even though those portions are a critical part of our operations.
But hey, the people who are "managing" the project get to put it on their résumé and act like they know what they're doing, which is all that matters.
an aggressive sales force that would sell ice to eskimos
I see you've read my column.
We will bankrupt ourselves in the vain search for absolute security. -- Dwight D. Eisenhower
Yes, it helps, but it really only helps with under-utilized hardware (and this is really only a problem in Microsoft shops). It doesn't help at all with OS creep; in fact, it makes it worse by making the upfront costs of allocating new "machines" very low; however, it has been and continues to be marketed a cure all which is where the snake-oil comes in. VMware's solution to OS creep: run tiny stripped down VMs with a RPC like management interface (that will naturally only work with vSphere) so that the VM instances essentially become just really heavy weight processes. We are basically coming full circle back to ESX just being yet another general purpose operating system where applications are written specifically for it and thereby defeats the entire purpose of using "virtualization" in the first place.
A film reel which was shown as the first thing in my graduate level AI class, I might add. Sadly, I no longer have the reference to this clip.
Heh. Day 1 of my AI class, the lecture was titled: "It's 2001 -- where's HAL?"
The other lesson from that first day of AI class was that the above properties made AI into the incredible shrinking discipline: each of its successes weren't recognized as "intelligence", but often did spawn entire new disciplines of powerful problem solving that are used everywhere today. So "AI" research gets no credit, even though its researchers have made great strides for computing in general.
Yeah that's when the prof introduced the concept of "Strong AI" (HAL) and "Weak AI" (expert systems, computer learning, chess algorithms etc). "Strong" AI hasn't achieved its goals, but "Weak" AI has been amazingly successful, often due to the efforts of those trying to invent HAL.
Of course the rest of the semester was devoted to "Weak AI". But it's quite useful stuff!
The enemies of Democracy are
I administer hundreds of virtual machines and virtualization has solved a few different problems while introducing others.
Virtualization is often sold as a means to completely utilize servers. Rather than having two or three applications on two or three servers, virtualization would allow condensing of those environments into one large server, saving power, data center floor space, plus allowing all the other benefits (virtual console, ease of backup, ease of recovery, etc..).
In one sense it did solve the under-utilization problem. Well, actually it worked around the problem. The actual problem was often that certain applications were buggy and did not play well with other applications. If the application crashed it could bring down the entire system. I'm not picking on Windows here, but in the past the Windows systems were notorious for this. Also, PCs were notoriously unreliable (but they were cheap, so we weighed the cost/reliability). To "solve" the problem, applications were segregated to separate servers. We used RAID, HA, clusters, etc., all to get around the problem of unreliability.
Fast forward a few years and PCs are a lot more reliable (and more powerful) but we still have this mentality that we need to segregate applications. So rather than fixing the OS we work around it by virtualizing. The problem is that virtualization can have significant overhead. On Power/AIX systems, the hypervisor and management required can eat up 10% or more of RAM and processing power. Terabytes of disk space across each virtual machine is eaten up in multiple copies of the OS, swap space, etc.. Even with dynamic CPU and memory allocation, systems have significant wasted resources. It's getting better, but still only partially addresses the problem of under-utilization.
So what's the solution? Maybe a big, highly reliable box with multiple applications running? Sound familiar?
because virtualization only works for large companies with many, many servers
You're full of crap. At my company, a coworker and I are the only one handling the virtualization for a single rackful of servers. He virtualizes Windows stuff because of stupid limitations in so much of the software. For example, we still use a lot of legacy FoxPro databases. Did you know that MS's own FoxPro client libraries are single-threaded and may only be loaded once per instance, so that a Windows box is only capable of executing one single query at a time? We got around that by deploying several virtualized instances and querying them round-robin. It's not perfect, but works as well as anything could given that FoxPro is involved in the formula. None of those instances need to have more than about 256MB of RAM or any CPU to speak of, but we need several of them. While that's an extreme example, it serves the point: sometimes with Windows you really want a specific application to be the only thing running on the machine, and virtualization gives that to us.
I do the same thing on the Unix side. Suppose we're rolling out a new Internet-facing service. I don't really want to install it on the same system as other critical services, but I don't want to ask my boss for a new 1U rackmount that will sit with a load average of 0.01 for the next 5 years. Since we use FreeBSD, I find a lightly-loaded server and fire up a new jail instance. Since each jail only requires the disk space to hold software that's not part of the base system, I can do things like deploying a Jabber server in its own virtualized environment in only 100MB.
I don't think our $2,000 Dell rackmounts count as "super-servers" by any definition. If we have a machine sitting their mostly idle, and can virtualize a new OS instance with damn near zero resource waste that solves a very real business or security need, then why on earth not other than because it doesn't appeal to the warped tastes of certain purists?
Dewey, what part of this looks like authorities should be involved?
Actually, the funny thing is, real snake oil actually does what it was originally supposed to do. "Snake oil" comes from traditional Chinese medicine (as a cure for joint pain), and was made from the fat of the Chinese water snake, Enhydris chinensis. It is extremely high in omega-3 fatty acids (particularly EPA), and is very similar to what is sold today as fish oil. Omega-3 fatty acids (in particular, EPA) are now known to reduce the progression and symptoms of rheumatoid arthritis.
Now, in the US, a variety of hucksters took fats from any old snake (if it even involved snake oil at all) and made all sorts of miraculous, unsubstantiated claims about what it would do. But concerning in its original role in Chinese medicine, snake oil likely did exactly what it was claimed to do.
Look at me, still talking while there's science to do.
In fact, this was an internal web based app for our office, which dealt with hotel reservations.
When setting up a new hotel on the system, the users (our staff), had to find and supply the telephone number as part of the standard contact details we needed for every hotel.
Do you know of any hotel that DOESN'T have a telephone, and if so, how would we call them to make a reservation ?
There are sometimes instances where some fields MUST be filled in, otherwise the whole record becomes worthless.