Dual-core Processors Challenge Licensing Models

ffub writes "Changes in hardware (such as dual-core processors and virtualisation) are making software licensing increasingly difficult for software firms. Companies still prefer the per-seat one-off license, while subscription models are favoured with software firms. But neither model reflects well the way software is used these days. The Economist looks at the situation and briefly touches on how Open Source could benefit from the muddle."

Comments

Maybe

[BHearsum]

Maybe this will get rid of licensing models that are 'per cpu'. I've never understood the logic in charging per CPU, anyone care to explain? One computer, one license. Or even better, no licenses.

Re:Maybe

[inode_buddha]

I would bet that the "per CPU" license model dates back to a time when CPU's were much more expensive; it could reasonably be assumed that there would be many users using one CPU. In other words, the business model is a couple of decades behind the technology.

Re:Maybe

[Anonymous Coward]

To me, it's like charging the driver of a larger car more to renew his plates, than the owner of a compact car. It doesn't make any sense.

In many jurisdictions, to encourage the population to drive smaller, more fuel efficient vehicles, they do exactly that. The US had a gas guzzler tax (don't know if they still have it) that was a one-time tax if your car didn't get 20 mpg or so.

Re:Maybe

[captaineo]

Rendering software is usually licensed per-CPU. It's a decent model since the number of CPUs in a studio roughly indicates how much it can afford to pay for software :). Though it seems likely that "per CPU" will soon become "per box" or "per OS instance" to avoid splitting hairs over the expanding jungle of multiprocessing technologies.

Re:Maybe

[byteherder]

To me, it's like charging the driver of a larger car more to renew his plates, than the owner of a compact car. It doesn't make any sense.

Most states charge based on the value of the car. This makes no sense other than trying to stick it to the rich. If you have a expensive compact car, you could pay more than someone with a inexpensive but larger car.

Charging based on weight makes more sense. The heavier the vehicle the more damage it does to the roadway. Thus larger cars should pay more, they cause more maintance to have to be done to the roads.

Re:Maybe

[Tim C]

I'd imagine that it's partly due to how much harder it is to write good multi-threaded code that scales well with increasing numbers of CPUs.

Yes, you're running the same code whether you have 1 CPU or 8, but if you do have more than one then you're actually benefiting from the additional effort (design, development, testing, etc). I imagine that the rationale is that it was harder and more expensive to write, why not charge more for it?

On top of that, the vast majority of multi-CPU users are business users, which tend to have more money and be prepared to spend it; you charge what the market will bear.

Re:Maybe

[nurb432]

I think it was more to get around the fact you could use one 'licensed seat' to access your application, but have 100's of people route thru that single seat. So they 'lose' revenue.

This way they can stick you for 'expected load'.

Remember too that once upon a time you were charged for use of that cpu TIME, not just a flat charge for access to it.. ( actually some of the big iron licenses is still based on a per cycle fee.. )

Re:Maybe

[Daniel Dvorkin]

No, it's the other way around; if you have many users on one CPU, charging per CPU makes no sense (unless you charge a lot.) The idea, of course, is for the software company to maximize its revenue, so by charging per CPU for big multiprocessor systems built on cheap commodity processors (which, of course, describes the majority of server setups these days) they can make more money. The justification (other than "we want more money") is that roughly, they expect the number of CPU's to scale with the number of users.

Re:Maybe

[S.O.B.]

Servers come in configurations of single up to 8 or even 16-way processors. Is it fair that a company with a server that has a single processor serving 100 users pays the same as a company that has a server with an 8-way processor serving 1000 users.

Per CPU licensing was a simple metric that allowed software companies to scale their pricing so that it was fair to both the entry level and high end customers.

As the article points out multi core processors are the processor companies' way of increasing performance without having to increase the clock speed and therefore keep temperatures down. Since software companies didn't care about the performance of a given processor, just how many you had, they shouldn't arbitrarily change the licensing model.

At the company I work for I know that because of the per CPU model we intentionally bought servers with fewer faster processors. Even though in most cases those servers were more expensive than machines with more processors the amount we saved on licensing costs more than made up for the additional hardware costs.

I suspect that in the end they'll end up with more of a performance based model similar to the MIP based licensing model on mainframes.

Re:Maybe

[TheCaptain]

Or even better, no licenses.

That is a really oversimplified and dangerous line of thought, IMHO. Even Linux and BSD have licenses...

Re:Maybe

[Spoing]

Maybe this will get rid of licensing models that are 'per cpu'. I've never understood the logic in charging per CPU, anyone care to explain? One computer, one license. Or even better, no licenses.

Where do you draw the boarders between one 'computer' and another?

If the licence were based on a per-metal-box basis, some clever folks would buy systems that are really clusters but are contained in a single box. Good for them, though it causes problems if you are the seller and supporter of the sofware.

Re:Maybe

[IntlHarvester]

Per CPU licensing was a simple metric that allowed software companies to scale their pricing so that it was fair to both the entry level and high end customers.

Which works as long as the hardware companies scale their prices with # of CPUs. Historically, going from 2 CPUs to 4 often quadrupled the price of a server, and going to 8 quadrupled it again.

The issue is that Intel and AMD are currently breaking this model. There isn't a substantial price difference between today's dual core system and yesterday's single core. But yet some software costs have doubled.

It's only a year or so until even laptops have dual-core chips, and $3000 Xeon/Opteron servers have chips with 4 or more cores. The whole assumption that 4 CPUs = Big Expensive System is going to have to change.

Database Licensing and the Web

[inmate]

This won't be the first time that licensing has faced such a crises.

In the early days of the web, I worked on a web-based project which connected to a MS SQL-Server database. The licensing issue was very confusing since the information in the database would be made available to anyone who came to site (and we expected a few hundred regular users), but technically everything would be accessed by through only one account (the webserver!).

I called the local MS office and they confirmed that we only need one licence for this model.
Based on this information, we rewrote a major internal application to be entirely browser based - and then dropped all our seat licences bar one.

Needless to say, MS had a absolute fit!

About a year later we received an incredibly confusing document outlining license-requirements for internet and intranet applications.

Re:Database Licensing and the Web

[blowdart]

There's actually a specific internet connection license for that sort of setup, however it's interesting to note that Microsoft have said, for licensing purposes, dual core CPUs count as a single cpu.

Compare to Oracle; if you buy a licence for a dual core machine, the second core is only counted as .75 of a CPU, as is each succeeding core. However Oracle rounds all numbers up, so .75 = one for licensing, and 1.75 = two, roughly the same cost as if you bought two licences. And so on. It's only a saving if you have 3 dual core cpus or more.

Re:Database Licensing and the Web

[mgv]

There's actually a specific internet connection license for that sort of setup, however it's interesting to note that Microsoft have said, for licensing purposes, dual core CPUs count as a single cpu.

Compare to Oracle; if you buy a licence for a dual core machine, the second core is only counted as .75 of a CPU, as is each succeeding core. However Oracle rounds all numbers up, so .75 = one for licensing, and 1.75 = two, roughly the same cost as if you bought two licences. And so on. It's only a saving if you have 3 dual core cpus or more.

Of course, microsoft used to allow you to have 4 cpu's for windows NT (this was back in the days when dual core stuff hadn't started).

Mostly, this is just about extorting as much money out of a paying customer as they can. If they charged a license per gigahertz of cpu speed, there would be an uproar when your software costs doubled when you upgraded your 1 GHz cpu to a 2GHz cpu.

When you look at it like this, you can see what a contrived concept that charging per core is.

Even if you argue that it takes more to write multithreaded code, that shouldn't make any difference between 2-4 cpu's. And in many cases the program utilisation might never even require that second core.

My 2c

Michael

Article mentions virtual servers

[putko]

I had not thought about the problem of virtual servers.

E.g. suppose I have a big-ass mainframe that emulates a few PCs, just to run Excel now and then (for legacy reasons). Once a month, we reconfigure the mainframe just for a batch job, so that some of its resources are used to simulate 10 PCs.

How do you price that? A mainframe license? 10 separate PC licenses? What about the fact that I'm only doing it now and then, and not using it regularly (8-10 hours a day)?

I just wish the article had used the term "price discrimination" -- that really explains it all.

Q: How much does it cost?
A: "How much ya got?"

We've heard this before...

[Zweideutig]

I know we have heard about this quite awhile ago on Slashdot, when Oracle wanted to consider a dual core CPU two processors I think companies like Oracle will be forced to think of dual core CPUs as simply one CPU that handles multiple threads well, especially with dual core CPUs not only coming from the Intel side, but also from IBM If I remember correctly Oracle found it difficult to determine the difference between dualcore and two CPUs. In the end, everyone will buy dual core, for the same reason everyone buys LCD monitors (it is seen as better, even if maybe it isn't.) Software companies will be forced to bend, hardware companies won't have to, because consumers are not going to put up with paying twice as much for what appears (on the outside) as one CPU. Should I be charged twice the parking fee because my 2001 Excursion has twice as many cylinders as the car beside it? I don't think so.

Re:We've heard this before...

[Snocone]

Am I missing something?

Yeah, a fair bit actually. If you're doing professional press work, digital photography, or video, you need the best true-to-life colour fidelity achievable on your monitor, and that means (very expensive) CRT, not LCD.

Also, I don't think any LCDs can match the pixel response time of CRTs, so the hardcore FPS gamer might notice a difference enough to prefer a CRT. My idea of a good game is more along Nethack lines, so I wouldn't personally know.

Re:We've heard this before...

[mrchaotica]

1. Yes, you should be charged twice as much to park your SUV, because it takes up twice as much space as a normal car!
2. Yes, your fat-ass SUV does need some serious downsizing, because although its size is proportionate to its engine, the engine is too damn big too.
3. I can't tell if you're being sarcastic or not about the Dunkin' Donuts and Burger King thing, but given that you don't appear to be joking about the Excursion I suspect you might not be. If that's the case, you should also feel sad about what a fat-ass you are.

Have a nice day.

CPU Licensing??

[lizdog]

Are we getting to a point where the term CPU loses its relevance? In gaming, is the power of monitor card selected as important than the speed of the CPU? Does the disk array attached to the database have more impact on speed than CPU? Should these also be factors in license models?

At least MS Got it right...

[Anonymous Coward]

Microsoft per cpu licenses are considered per socket, not per core. Makes getting a single dual-core cpu preferable to dual single-core cpus.

Robber Barons

[forq]

The software industry has gotten away with robbery for too long. Year over year they astound us with their skyrocketing costs, and as computing complexity goes up, they find more and more excuses to not deliver the support you're paying for. "We cannot support you because of X." X being any reason they can find. Upgrades, new hardware they don't have in their support matrix, virtualization. Whatever the reason, the very first order of business for those support folks when you call for help is to find a reason to not support you. And now they want more money. To pay the outsourced first level support folks that know all about how to determine if you're unsupportable, and nothing about how to support the products.

Ridiculous.

per seat licence

[vally_the_poo]

> Companies still prefer the per-seat one-off license

I understand now why they don't have any seat at my new work: everybody just sit on the floor, in a hippie style...

duh !

How about per cycle?

[wandazulu]

I'm surprised no one is talking about this, as it seems it was all the rage back-in-the-day (and I believe still going today): charge by the cycle for the app.

Case in point: I worked with IBM's MQSeries product as a link between a mainframe and a webserver. The MQSeries license for NT was something like a flat $6000. On the mainframe, however, it was some ungodly amount for the tapes, then they charged a per-cycle fee *and* a monthly maintenance contract.

As part of load testing, I wrote a program that would spit the complete works of shakespeare back and forth, over and over, to the mainframe and back using multiple threads. Two weeks of testing cost the company an extra $12,000 because of the cycles expended.

I noticed too that starting with SQL Server 7.0 that the explain plan feature can also show the number of cycles spent on a particular step. I would think Microsoft, with that info, could, if they wanted, go to a similar model with SQL Server if they so chose (and wanted to effectively kill the product).

And now that I think about it, my Unix account back in the early 90s had a cost associated with it too...I was allotted something like $1000 worth of what I assume was cpu time, and sure enough, enough attempts to get Nethack to compile and I was back in the office begging for more "money".

Ah, the good old days. I think.

Re:How about per cycle?

[Tablizer]

As part of load testing, I wrote a program that would spit the complete works of shakespeare back and forth, over and over, to the mainframe and back using multiple threads. Two weeks of testing cost the company an extra $12,000 because of the cycles expended.

To bilk or not to bilk

Per CPU licensing makes no sense anyway..

[wfberg]

Per CPU licensing makes no sense anyway. It gives no indication how heavily an application is used, or how important it is to a business. For databases, it would make more sense to have a license for X thousand transactions, or Y amount of data. After all, databases are used for doing transactions and storing data. (Don't let Oracle get wind of this idea though, I've got an Oracle database that's more than 1GB in size but compresses down to 30MB! This pricing model will be the ideal excuse for them to take up even more disk space..)

The reason licenses are tied to hardware or to seats is probably because it's easy to justify these as a "cost of doing business" to suits. While projects usually have the greatest difficulty getting an OK for money to go towards programmers, expensive hardware is purchased willy-nilly, on the basis of "well, now we've got this application, we need to run it, or else the money we spent on programming it is wasted!". So tying your database license to CPUs makes more of an afterthought. (Just like performance, scaleability and actual volumes are an afterthought).

The same goes for seats; you just HAVE to license one copy of Microsoft Office or an OS or a database for every employee, otherwise you're paying (some) employees for basically standing around! Then, to recover costs, you make sure they have very little access to things like notepads, pens, or copying machines, since those dimes add up, don't you know?

Call me a cynical bastard if you will..

licensing = overhead

[bromoseltzer]

In my experience in academic computing support, one of the biggest headaches is license management on Windows and Macs. We tend to have lots of different software packages installed in ad-hoc seats or small networks. Each one may want a dongle or a dedicated server environment. Each one has different contractual terms about student vs faculty vs research use. Etc.

All this, as I see it, is a pure waste of scarce resources. It is somewhat alleviated by sitewide licensing of a few products, but even these are not easy to administer. The whole scene is like the U.S. medical or tax system -- value is being delivered, but the administrative overhead is huge. All the costs of compliance are passed on to the end users and institutions.

What a difference with Linux and OSS! Easy licensing is a big plus and it's not well enough appreciated.

per-thread

[rhythmx]

Why don't the software companies license by something that they can control? A "number of threads" model would be more fair. Or at least, the license can't assume that all the hardware is there for it to use and profit from.

If I had an 8 processor server running an existing application that I also wanted a low-end DB server on, I could just buy a single thread license instead of an 8 cpu one. Later, if the DB server couldn't handle the load, I could simply upgrade it to a 2 or 3 thread server.

It's better than power unit licences for a start

[xixax]

Until quite recently, our database software was on power unit licences. A formula number of CPUs x MHz x architecture is used to work out how much it will cost you to run the database. Why? Well they want people who are running huge databases to pay more, and size of server(s) is a pretty good measure, Amazon isn't going to run on a single CPU. That is, they charge as much as they think the customer can afford.

While an interesting question, how does this question manage to rate as a "insightful"?

Xix.

One license per-disc and shut the hell up ?

[billcopc]

I love capitalism. No really, I love watching people test just how hard they can screw each other in the ass for money without getting shot. Here's how I see it:

Company ABC invests X money into developing product. They estimate sales of Y quantity. Divide X by Y to get a per-item cost, mark it up for profit and a support allowance, then sell it.

The fact that I might run their software on multiple CPU's, or that it might be accessed by Terminal Server, doesn't change a single thing for the developer. They don't need to work harder, they don't lose sleep at night, their kids won't end up on Springer. It doesn't matter whether I use it to index my MP3 files, or run a Fortune-500 business with it. They did their work, and they get paid for that work. What happens afterward is not their problem, and more importantly none of their goddamned business.

When people learn to take just compensation for their efforts, and give up the "fight" for riches, we'll wonder how we ever survived through capitalism. There is a set amount of monetary value in the world, the more you have, the less someone else has, and the more that person is likely to do nasty things to make up for the loss. So why don't you just be happy to eat every day and give me a goddamned break with your license gouging.

Wha-Whuuu?

[TheoMurpse]

Linux, the open-source operating system for Pentium-style processors

Did anyone else notice that line and do a doubletake? I parse that sentence as implying that Linux is only for Pentium-style processors.

Reminds me of a funny story...

[sanermind]

In college, for a chemistry class, the textbook included some a CD-ROM with some java software that we were supposed to use at home for our homework (or in a provided computer lab for the less fortunate). Anyway, I couldn't legally run it at home, because the shrink-wrap EULA prohibited running it on more than one CPU ...As my home system was a SMP athlon system (an affordable one too, using the XP-to-MP trick), I could not legally run it at home! What was even funnier was that when I mentioned this to the professor, he seemed terribly suprised that I actually read the EULAs. I suppose that most people don't, which you would think would strongly challenge their enforcability, but alas.