Slashdot Mirror
Software-Defined Data Centers Might Cost Companies More Than They Save
storagedude writes "As more and more companies move to virtualized, or software-defined, data centers, cost savings might not be one of the benefits. Sure, utilization rates might go up as resources are pooled, but if the end result is that IT resources become easier for end users to access and provision, they might end up using more resources, not less. That's the view of Peder Ulander of Citrix, who cites the Jevons Paradox, a 150-year-old economic theory that arose from an observation about the relationship between coal efficiency and consumption. Making a resource easier to use leads to greater consumption, not less, says Ulander. As users can do more for themselves and don't have to wait for IT, they do more, so more gets used. The real gain, then, might be that more gets accomplished as IT becomes less of a bottleneck. It won't mean cost savings, but it could mean higher revenues."
GTFO.
'We are trying to prove ourselves wrong as quickly as possible, because only in that way can we find progress.' RPF
This.
Of course helping the business get its task done is the only reason IT exists at all. If the increased usage
results in a more profitable operation, then its a good thing. It is back to the first step of enterprise architecture, which is
what jobs does the business want IT to help it accomplish? If the business can accomplish more by using more resources,s
so much the better. It is not ITs job to ration IT resources, (except perhaps by charging back costs incurred). If the cost recovery
model recovers for the additional resources used, then It is not squeezed either. If the model is not recovering costs it does
need to be adjusted. IT is often cited for being to slow to meet business demands, so this helps. However it may well
reduce the size of the solution construction part of IT as those resources move to the business units. Here is where
a set of corporate standards would be useful. Of course we have seen over the last 20 years how a thousand flowersb
bloom in a branch of IT then the market prunes it back to 1 or 2. Take Ethernet versus Token Ring, or TCP/IP versus
Decnet, or the old SNA architecture. Or another example, MIPS,Sparc, Itanium, Motorola Chips, or the Vax hardware
architecture.
Faster internet? Download more data. Bigger hard drive? Store more porn. Better video compression algorithm? Larger files per hour.
The situation can be modelled by the theory of price elasticity from microeconomics. If the demand for cloud computing is highly elastic, then cutting the price of the resource in half will be met by more than twice the original quantity demanded, after adjusting for the long term growth in demand that would've occurred even without the price adjustment, i.e. the long term movement of the supply and demand curves. This high price elasticity case is what the Citrix guy is speculating could happen. I don't know that it makes sense, though. If the price of gasoline fell in half (let's say from $3.60 to $1.80/gallon) and stayed there for 12 consecutive months, long enough to convince people that it wasn't a fluke, people would undoubtedly drive more and use more gas, but I doubt they'd use twice as much as they are now.
Because when people read the label and see that the food is lower calorie or "more healthy"; they eat a larger amount of the food because they feel less guilty due to it being "more healthy"; and the additional consumption more than offsets the decrease in calorie count of the "healthier food"
So eating lower calorie foods makes you less healthy....
In the early 1950's, there were only a few computers (mainframes). The idea that we would now have only a few dozen computers in the world, which would each cost a fraction of a cent due to Moore's Law, sounds pretty dumb, doesn't it? Obviously the ability to do more is at least as important as declining cost for a fixed capability. Nothing new at all.
Virtualization makes it easier to stand up a new "server." True.
This simplicity will lead to using more "servers." Granted.
But those virtual servers require far less hardware than the old physical servers. Many of these virtual servers are used only a small percentage of the time. Depending on the load, 10, 20, or even more servers can run on one physical piece of hardware.
So even if we use, say, five times more "servers" with virtualization, we will be using fewer physical units--fewer "resources."
In short, the math is not so simple.
Since engineers have always worked on efficiency so pretty much everything you use these days is more efficient that the equivalent item from 30 years ago. However people in the US use more energy per capita than they did 30 years ago.(So for example instead of 4 people in a family watching the same 25" TV during prime time each one of them has their own and or they watch it more. End result is the amount of energy used to watch TV is greater even though the actual TV uses less power.)
Did you know 80 to 90% of the moderators on slashdot wouldn't recognize a troll even if one dragged them under a bridge.
If you had an energy source which was unlimited, people would piss it away. If aircraft had a 100% safe nuclear power source where planes could fly unlimited distances, the sky would be full of planes with people on joy rides. Likewise cars, likewise ships. Space shuttles burning hydrogen and oxygen rockets would have bi-weekly launches. We would have country sized air conditioners in the hottest of deserts, and people wouldn't bother shutting the door. If IT can provide a hundred million queries pre-built for novices, using huge amounts of resources in the process even when the vast majority of those queries will never be used or rarely used, but people want them 'just on spec', then they want them and resources will be wasted.
Refreshing that old quote: You implement Cloud, the new network structure, where they pay dollars an hour, have no health care, no retirement, no pollution controls, etc., and you're going to hear a giant sucking sound of US admin jobs being pulled out of this country. We have great telco agreements across the world. :)
Do US admins, technical staff, CS graduates, staff with double degrees really think US multinationals will let you work overtime with Seattle civilian aircraft engineers like wages for generations?
The "creativity and innovation" will be in security/privacy cleared front companies based in the USA to offer legal cover for huge out sourced backhaul.
Agility will be more in finding a "self-service model" faster based on ever cheaper staff in say a Vietnam or Laos.
Will the US end user see any price savings? The cost per core, per seat, per VM will be rental and with their 24/7 'local' hardware support for your software rental:
cost will chart in one direction: Up
Domestic spying is now "Benign Information Gathering"
Jevon's paradox is valid, but only under specific economic assumptions.
It's only true so long as there is more demand for the resource, and it's only a problem when the resource has a cost attached. Essentially, it's true in a "scarcity" economy, but not true under "post scarcity".
We've achieved "post scarcity" for several resources already; for example, phone calls and computer time.
Phone calls used to be expensive and billed by the minute, but nowadays it's virtually free. Similarly, computer time used to be metered and charged - in college, the CPU time for each program run was deducted from your account. Nowadays people can have as much un-metered computer time as they want.
CPU time and phone service aren't literally free, but the cost is so small as to be negligible.
Despite this, we do not see infinite consumption. People have a certain level of need for a resource, and when that need is met they stop consuming more. Coupled with a declining population, there is no reason to expect infinite consumption.
Your company may be using more resources than it needs... but so what? Computer resources are remarkably cheap - so cheap, in fact, that it may be more effective to ignore the problem. Optimize the biggest expenses first: if that turns out to be IT resources, then take a closer look. Otherwise, just ignore it.
(For another example of post-scarcity, consider the Chinese "dollar stores" that have cropped up. The cost of goods is so small that the time and expense of price tags makes a big difference. This is almost post-scarcity of tangible goods.)
This might be the stupidest article Slashdot has ever posted.
Before spouting off your moronic post to show everyone what a genius you are?
Read the sentence before your quote. He was citing the 150-year-old paradox.
If you have some sort of brain dysfunction that prevents you from remembering something for more than five seconds, my sincere apologies.
Talking about useless buzzwords.
So, we are back to the "you'll never need more than 640K" or "the world only needs 5 computers" or "a telephone per town will be good enough" logic?
Now, in the real world, we invent new things, and then make them as cheaply as possible, increasing use/ownership of the widget, and hopefully finding as many uses for the widget as possible. There are even films from the mid 20th century laying down this concept of capitalism and consumerism working at its most effective. But guess what? There are ALWAYS entrenched interest in the older and crappier ways of doing anything who will pay 'experts' to proclaim that the proposed new ways are stupid/dangerous/expensive or whatever.
Let me guess. Next up on Slashdot is a reference to Amdahl's law that 'proves' no general computer ever benefits from having more than 3 cores? Those that quote such 'laws' are ALWAYS too thick to comprehend the extremely limited set of circumstances to which the 'law' applies. Those that understand such special circumstances will never quote such 'laws'.
Datacenters are now shit.
Back in 2000, the rule, and it worked, was one app per server.
Now assholes are piling on the virtual servers, and piling as many apps as possible onto each server.
In addition, the agile process skips the architecture phase expecting the architecture to grow by consensus, So when engineers add functionality to a system, its guesswork in a short time frame based on what's already there, which may or may not be the be good place, or create yet another new app.
And never mind the junior developers who don't know what the fuck a CDM is, and insist on writing their own, incompatible shitty CDM, because they didn't learn that shit in school, creating more system overhead because of excessive data-translations.
Grasping with both hands for relevance, aren't we incumbents?
Love that everyone's response at Slashdot is to call bullshit.
I can see the FUD streaming out of the sales marketing teams now, "better stick with us or who knows what'll happen..."
You still need to hire someone to install and configure the virtualization apps. Not much different than running all your apps on the one piece of hardware. Virtualization: A solution in search of a problem, in a saturated market ...
AccountKiller
Work expands to fill the space given to it.
Give it no definable boundaries?
SpaceMonster: OOOOOOOH!!! *Wiggles fingers acquisitively*
Chas - The one, the only.
THANK GOD!!!
IT is really important and users need IT services even though they don't think they do, says IT services company
Datacenters at companies big enough to be using virtualization are measured by power use not by hardware / cpu power.
They have mandates that forbid them to use more power and that is the limiting factor.
Besides, virtualization as long as the servers are not static (ie always with the same resources), allows for using what is required with way more precision.
An accountant in my company is often bitching about the slow speed of internet banking, and the problem is there are a lot of bottlenecks between here and where the local bank holds their sensitive data on servers in India. It's the same with a lot of data entry of medical records.
I'm not suggesting it's a good idea but merely pointing out it's an idea that seagull management (makes noise, shits on everything then flies out) implemented in a lot of places at least five years ago.
Using more resources is exactly what happens... As hardware gets faster, software gets slower. While some of the slowness can be attributed to additional features and larger data sets, much of it is down to using higher level languages. Very few people bother writing efficient code anymore, on the basis they can always throw more hardware at it.
I have personal experience with a few games that were deemed too slow and rather than try to improve the code, they were simply shelved for a couple of years until the average hardware had caught up.
http://spamdecoy.net - free throwaway anonymous email - avoid spam!
The whole idea of SDDC and Cloud Computing is to basically end up with "IT as a Service". The rest are just marketing words. The goal is to have a service pretty much like electricity: you don't necessarily care where it comes from or how it's delivered to your premises. All you care is that it's there, it's reliable, it's consistent and you know exactly how much you are paying for.
The problem I've seen in the 10 years I've been in this particular industry, is that very few large companies are doing chargeback from IT to their internal customers or business units. IT has been historically seen as a shared cost for the company which adds tremendous pressure every year to cut more and more and try to leverage economies of scale whenever possible. Once you implement chargeback (even if it starts as a showback only) you can effectively pass that cost to the internal customer so you end up shaping their behavior depending on their own funds allocation, not IT's.
The next step is to have accurate forecasting so you know exactly how much infrastructure to have available, particularly if you implement service tiering. This doesn't mean that IT will have a free ride, and it will still be expected to be competitive with external cloud providers, but at least is something more manageable than the status quo.
"Might"
"Might" cost more than they save based on data gleaned from coal burning plants. I was going to call this an apples-to-oranges comparison, but those two things are actually fairly similar. This is more of an apples-to-hemidemisemiquaver comparison.
This signature is false.
Yes, I am picking this nit: OP, you mean "might end up using more resources, not fewer". The rule is, if you can count it (like marbles), use "fewer", if you can't count it (like water), use "less". This is basic English, not rocket science. C'mon, people!
Cheers, Tim -- Tim Janke Part mad scientist, part lion tamer: sr. software engineer, global team leader, project mana
Sticking stuff in teh cl0ud makes accounting clerks and order pickers transmogrify into software engineers?
Confucius say, "Find worm in apple - bad. Find half a worm - worse."
Look grasshopper. Virtualized and software defined are vastly different. One can enable the other, but stop confusing yourself and lots of other clueless
No it isn't. The exponential is a function that describes the relationship between two variables (e.g. time and number of rabbits). It doesn't describe the difference between different values of the same variable (number of rabbits on Monday and number of rabbits on Wednesday.
Confucius say, "Find worm in apple - bad. Find half a worm - worse."