Why Vista Release Date Really Slipped

anzev writes "A team manager for Windows for 5 years has decided to write a blog-essay about what caused Windows Vista project to miss the due date. Philip tells us in the blog, that Windows developers are writing an average of 5000 lines of code (which is *only* 1200 lines less than the national average of 6200 lines of code per year). He addresses issues like the Vista code being too complicated, the processes the developers have to follow too complex and a lot more. All in all it gives a nice insight into why Vista will be late, from a different perspective. Oh, and Slashdot gets mentioned too ;-)."

Comments

SLOC: Vista vs. Linux

[(1 -sqrt(5))*(2**-1)]

From TFA:

We shouldn't forget despite all this that Windows Vista remains the largest concerted software project in human history.

David Wheeler, for instance, calculated that Redhat 7.1 contained 30,152,114 physical source lines of code (SLOC), a 60% increase over 6.2 (and that was in 2001).

Linear extrapolation would take us to about eighty-two-million today, comfortably over Vista's projected fifty-million; but who's counting?

Re:SLOC: Vista vs. Linux

[Zigurd]

Now, if the Debian project managers were told to write specs for all n-thousand of these modules, and then told "deliver these modules so we can have the next 'eager beaver' release," then you'd be looking at a concerted effort.

You have answered youself: TFA asks if a project the size of Windows in controllable. In an environment where the tone is set by Steve Ballmer the answer sure looks like "No." Maybe "Hell no."

TFA also states Windows has 50 layers and circular dependencies. Linux has complex version and interface dependencies, too. But, evidently, the Windows dependencies are hairy enough to flummox 2000 smart people working in "concert" while Linux gets by with only a handful or people working in the same place at the same time and a much simpler process.

That is, Linux has better modularity and less process.

That points to the depth of the problem at Microsoft: They will have to change almost everything about how Windows is made in order to get a different result:

They have to stop telling developers to "do or die." Has that ever happened in the entire course of Linux development? Probably not. It's something that software project management can do without.

They have to get strong product management that knows which features are actually important, so you don't get that "do or die" message being sent to teams that are making things that don't add that much value.

They have to decouple development more: Why on Earth do you need to have 2000 people working in concert on any software project? That's a bug, not a feature.

They have to un-layer their management structure. 11 layers? That's ridiculous for a software company.

There is no one prescription for success. Apple succeeds by having very strong product management, so they know which features are actually important to the end user. Linux succeeds by having no product management at all, and having to adapt process to the practical constraints of being FOSS. Microsoft is stuck in the middle: Not enough product management strength to know which parts really deserve a "do or die" effort, and so much process, interdependency, and management layers that any of the 500 product managers Microsoft already employs that are smart enough to make these decisions can't possibly put them into effect.

Re:SLOC: Vista vs. Linux

[Dunkirk]

Excellent. That's exactly right. Just be sure to include the costs when comparing all of this. So... while you're right that you're comparing entire up-and-running, productive systems, the Linux one is still free (or, say, $60 for SuSE) versus the Windows one, which is now into the low thousands of dollars. Go ahead, keep adding things to comparison: firewalls, chat clients, web browsers, application servers, infrastructure services. All free. All included. And now you're talking about, literally, tens of thousands in licensing fees with Microsoft. And, in the case of any Linux distro I'm familiar with, all these packages are kept up to date with each other for security patches. And all the drivers (except nVidia, grrr) get updated with a new kernel. Truly one-stop-shopping.

About 7 years ago, I counted up that I had personally spent several thousand dollars on Microsoft products. I don't believe in pirating this stuff. Rather, when I thought about the kind of money I had been spending, and EXTRAPOLATED to what I'd spend in the future, I backtracked to the fact that I knew Linux pretty well, and I decided to make the switch. I've used Linux as my desktop as well as my servers since then. (And my first question to my local Mac-head was whether the new Mac's could triple boot OSX, Windows, and Linux.) I used Linux at work for years until my latest boss told me specifically that I couldn't run Linux. (I asked, if I was getting my job done, what it mattered. He didn't have much of an answer, but I let it drop.)

Since that time (Windows & Office 2000), the only money I given Microsoft has been the $15 co-pay for Windows XP that I got while attending the local college. At the time, I also got Office XP and MSDN Student Ed. for free. (Did I subsidize Microsoft with my tuition? Well, my employer did.) Yesterday, someone asked me to pirate a copy of Office for someone else, because that person didn't want to pay $400 to go buy a copy. I said no way. That's the whole point. Microsoft gets away with their practices because deals like the one with colleges, but that's a lot of money to a person who does volunteer work! That's the deal. It IS unfair. If you don't like it, then rightly be upset with Micrsoft's stranglehold on the market and do something about it. I told him to download OpenOffice and stick it on a CD for this other person. If 100% compatibility is worth hundreds of dollars to you, more power to you. But it's not to me, and I'm hoping that more people come to see it that way.

Re:SLOC: Vista vs. Linux

[Locutus]

TFA also states Windows has 50 layers and circular dependencies. Linux has complex version and interface dependencies, too. But, evidently, the Windows dependencies are hairy enough to flummox 2000 smart people working in "concert" while Linux gets by with only a handful or people working in the same place at the same time and a much simpler process.

That is, Linux has better modularity and less process.

So, 'integration' is what's holding Microsoft developers back while 'integration' is what the Microsoft marketing drones are pushing as a reason why MS Windows is better than GNU/Linux? Interesting. This should be an indication to those in the field that there's a major problem at Microsoft and the customer is taking the brunt of the impact of such a 'design' and the marketing boys need to do a bit more explaining how this 'integration' is a good thing.

I must say that the GNU/Linux camp must watch its step though. Too much reliance on KDE and/or Gnome can result in some of the same problems. Ok so the kernel will always be seperate, but what about all the utilities like sox, libjpeg, libxml, libwww, libtiff, etc? If these slowly start moving into the GUI, a slow moving(feature/upgrade/etc) desktop or worst an unstable and insecure desktop is still going to be bad for users. For many, a crashing GNU/Linux desktop is going to feel just like a crashing Windows OS/system. The movement toward compound document frameworks( openParts/etc ) could help here but there is definately a chance of the GNU/Linux desktopPC moving toward the same problems MS developers are seeing with Windows. Atleast to some extent.

But right now, the advantage is on the GNU/Linux and OSS side.

I wonder what Balmer or Gates is going to say when they next tell the market how the 'integration' of Windows is why they'll beat GNU/Linux in the long run and someone mentions this blog in response?

LoB

Yeah, whatever.

[Qbertino]

"This isn't some critical release patch. This isn't some driver that's long overdue. Microsoft never hand signed a sheet of paper telling me that I would have my copy of "Longhorn" by the end of 2005 or even 2006. It's a new operating system."

"Oooh, de poor little Vista Developers are sooo overworked. Lets give them a break."

No. Wrong. No break. And no extra auto-credit for being MS. I couldn't care less about Vista being delayed or not. But I will take every chance to turn the situation against all legends that cause people to think Computer == Windows. Usability == Doubleclick. Etc.

Reading that essay - from a Vista Guy with a position - gives of one clear message: Vista actually is a bloated weedy mess beyond any measure. And, guess what, making something new or not, the code that makes the unixes so usefull has been programmed allready and is in heavy field use for quite some time now. Somewhere between 10 and 20 years. After 30 years of unix, hardware finally is fast enough to run it on PDAs and cheap Notebooks. What x86 is to architecture - ancient, crazy, nutcase, but good enough for everything, even a Mac, Unix is to OSes - ancient, crazy, nutcase, but good enough for everything, even good enough for a Mac.
No, no break. Game over I say.
If MS has the guts to burn 10 Billion - 20 Billion on getting a new OS paradigm on to every plattform on the planet and do a good job at the same time they'll maybe make it. But even this late, jumping the OSS bandwagon and burning the cash it takes to take over the whole OSS service, distribution and customization sheebang would be cheaper and have better prospects.

Re:Yeah, whatever.

[jc42]

If MS has the guts to burn 10 Billion - 20 Billion on getting a new OS paradigm on to every plattform on the planet and do a good job at the same time they'll maybe make it.

It's perhaps worth mentioning that this was essentially IBM's approach back in the early days of "desktop" computers around 1980. There was this flock of little upstarts challenging IBM's growing stranglehold on the computer business by building small, cheap computers. IBM actually had a desktop computer, and it got very good reviews from its users. (What was it's 4-digit number? I've forgotten, but it was pretty good. ;-) Their problem was that, due to their development rules, they couldn't sell it for less than $50,000 and recover their development costs. And they couldn't sell more than a handful at that price.

So they farmed the job out to one of those startups, run by Bill Gates and a few of his buddies, handed them a few hundred million for marketing, and didn't impose the IBM development rules on them. The result was crap compared to any of the CP/M desktops, but with a marketing budget greater than the operating budgets of all the upstarts combined, the result was what IBM wanted.

Microsoft has understood the lesson of this from the start, and put their money into marketing rather than quality product. Until now, maybe. If the reports of their growing development rules and costs are true, they may be going the way of IBM. They'll produce a good system for the first time in their history, but to avoid going bankrupt from the cost, they'll have to get a very good price for it, and only the wealthiest (and stupidest) will pay that price. If this is true, we're seeing a repeat of the IBM/Microsoft story from a quarter century ago.

Of course, IBM didn't die. In fact, they completed their conquest of the "mainframe" market, which was willing to buy IBM no matter what the cost. They completely own that field, and development has pretty much come to a halt. Due to MS's market clout, we could see the same outcome. They will own the "desktop", and further development in that market will grind to a halt. They'll still sell to the "MS at any cost" market. But it won't matter to most of us, because we'll more and more consider "desktop" computers relics of a previous age. We'll stop worrying about making new systems "compatible with the desktop" (i.e., clones of MS's systems), just as 25 years ago we stopped worrying about whether our little computers were IBM compatible (and we didn't bother with PL/I or JCL ;-).

So what should we call the new thing we're building while ignoring IBM and Microsoft? "Web 2.0" seems to be out (and wasn't very good anyway), but the new thing will certainly be Net-based. Any good suggestions for a name that will supercede "desktop"? Maybe we need a catchy two-syllable name for the software going into the OLPC project. Push for making it a truly distributed, comm-based system without any central control, so the comm companies and local governments can't take it over as they're doing with the Web. We can base it on zillions of small components, so a company that refuses to follow standards can't make any inroads. The "new paradigm" will be as outside Microsoft's world view as PC/DOS was outside IBM's world view.

C'mon, we need a catchy new name ...

Re:Give Vista Developers A Break

[Moby Cock]

If Vista is as complicated as its specs say it is, I hope Microsoft takes another two years to get this done because I don't want to have to put up with Vista SP1, Vista SP2, Vista SP3, etc. down the line.

This is my beef with Vista. It is late and when it ships, I expect it to be buggy with many follow-on Service Packs. The reason it is late and buggy is the absurd devotion to backwards compatability. I don't understand it. I could accept software compatability, but the hardware aspect is mystifying. Microsoft could spend five more years trying to get Vista to be all things to all people, but its stupid. OS X ships every 18-24 months. Granted this is not a full up new OS, but these releases are much more significant than service packs. I am of the mind that this is possible because they make the hard decisions about what hardware and legacy support should be cut. (Although it is easier for Apple, since they build the hardware too. Perhaps the saga of Vista shows that the Apple model is inherently more technically sound.)

Source for averages?

[Anonymous Brave Guy]

I thought the number of finished lines of code per developer-day (that means debugged, documented, etc.) was only 20 for an average developer? A top developer will get closer to 10x that (mainly because when they write a lot of code in a day, they don't introduce lots of silly bugs that take a lot of time to correct later). Some developers actually have negative productivity overall (which makes sense when you consider the time spent by their colleagues to fix their bugs afterwards).

I can't remember where I saw those stats: probably something like Code Complete or The Mythical Man Month, I imagine, or possibly the IBM study into developer productivity at different ages (the one that says anyone under 25 is only good for documentation, and anyone 25-30 should only work on one project at once). Does anyone recognise the number?

I can't see any references in the blog post. Where do the figures of 6,200 (and the earlier 9,000) LOC/year come from?

it's kind of obvious what's wrong

[thechronic]

The guy is saying there are 50 layer dependencies, and tons of circular dependencies. It's software engineering 101, their model is wrong...they're not properly abstracting out each layer. I'm not a big fan of Linux, but every module can be decoupled in it, and modules work together even though they're written by completely different projects due to standards...that's how you design a proper system.

Re:it's kind of obvious what's wrong

[tgv]

Hey, Vista is not Software Engineering 101. Everyone who participated in a major software project probably ended up accepting circular dependencies. It's a compromise between the size and usability of a module on one hand and independency on the other. And if you don't want to duplicate code (talking about SE-101), things can get quite complex in such a big system.

Not that I'm looking forward of installing it on my Dual Core iMac, though...

Ballmer used to complain about this!

[Cybrex]

In the documentary miniseries "Triumph of the Nerds", there's an interview with Steve Ballmer where he describes the various factors that led to the fallout between Microsoft and IBM. One of the big things that he harps on is how the IBM programmers were too focused on KLOCs, while the M$ guys were striving for streamlined, efficient code.

Now we've got one of the head guys on the Vista project going on about KLOCs. Is anyone surprised? Me neither.

Re:Ballmer used to complain about this!

[Teach]

Now we've got one of the head guys on the Vista project going on about KLOCs.

Several others have brought up this concept. I'm not a defender of Microsoft *at all* (switched to Linux on the desktop in 2002), but 'philipsu' is not talking about KLOCs in this way.

Some companies use KLOCs as a measure of productivity, and incorrectly so. More lines of code does not mean 'better' code or better productivity.

TFA is sadly now removed by the author (does anyone have a copy in their browser cache?), but there is no evidence in it that Microsoft uses lines of code as any metric whatsoever. The author does bring up the low code output by Microsoft programmers to demonstrate empirically that their programmers (which are assumed to be talented and capable) are being hampered by a difficult task and broken system, both of which prevent the programmers from achieving code output anywhere near the industry average.

You may appreciate that I have tried to emphasize words frequently in my comment here to better match the style of the original article. :)

You have to appreciate the transparancy of MS

[Momoru]

It amazes me how transparent Microsoft lets itself be. The fact that someone can even post a blog like this about the company using internal company resources. As much as we rant about Microsoft being evil, you'd never see anything like this from Google, the only blogger about the internal day to day Google I know of was fired (Mark Jen).

Re:Summary == Wrong

[korbin_dallas]

Cowboy!

In a 'REAL' company, you have to do:
1. Requirements
2. Requirements Review (meeting)
3. Complete Requirements Review paperwork,and store that

1. Design
2. Design Review (meeting)
3. Complete Design Review paperwork, and store that
4. Complete a Design Checklist,and store that

1. Then you get to Code. (insert 1000 lines of code here)
2. Have a Code Review (meeting)
what this really means is that you need to tie up 2 other developers to check your 1000s of lines)
3. Revise all changes and complete Code.
4. Complete a Code Checklist and store that

1. Unit Test

1. Integration Test

1. Update and maintain the Software Design Document (ongoing).

Welcome to CMM!

For us the metrics are about 30 lines per day, but for Rule of Thumb we use 18.

The project I am on started as R&D,then it was fun, we just coded like you did. We did ALOT of good stuff. Then when the product sells and makes ALOT of $$$$, we started to get all these MANAGERS, the ones who thought we were all STUPID, piling onto the project SUKING up all our money. Then we get this CMM crap to do too. Some of it helps, At least if my Requirements change, I start at #1 and managment, if they have any questions, see what the impact of thier whims are.

Now if somoeone would just buy Serena and put PVCS out of its fscking misery. Oh Happy, Happy, Joy, Joy.

Re:Give Vista Developers A Break

[Spiked_Three]

LOL - I got a huge kick one time at a Microsoft SE meeting where some dumb SE gave Gates grief over Microsoft's implementation of DCE RPC. Gates ripped him a new butthole stating intricate details of DCE specs and it's failing points and how code was written to get around the problems.

There does not exist in all of slashdotdome more of a nerd than Bill Gates (nor anyone as rich).

Simple Lessons

[radtea]

The article describes the basic things that are wrong with virtually every late project:

• Managers who can't handle the truth
  
• Developers or lower level managers who lie to fulfill the psychological needs of managers who can't handle the truth
  
• A marketing-driven schedule culture that declares "drop dead dates", misses them, and nobody dies
  
• Input rather than output focused management (action control vs results control)
  
• Managers who interfere inappropriately with technical decision-making
  
• Excessive project mangement/process burden--the level of process needs to be just right for a project to run effectively, and needs to retuned on a per-project basis by people to whom success is more important than ego
  

He is describing a sick management culture, one peopled by individuals who are not part of a reality-based community and not aware of their own deficits. Projects run by people like this will always be late and frequently fail completely, because reality doesn't care about management egos.

This is pretty typical of modern management culture. It just shows up more clearly in this case because of the length and size of the project.

Re:Give Vista Developers A Break

[Overly Critical Guy]

Microsoft never hand signed a sheet of paper telling me that I would have my copy of "Longhorn" by the end of 2005 or even 2006.

Not only did they do so in what is called a press release, but they held public demos of Longhorn and proclaimed its target release date. And the original announced date was 2003.

In 2006, there were public statements made that Vista would be out by the end of the year, no matter what. That was another lie.

Re:The application of "concerted"

[iluvcapra]

Arguably, Debian developers are acting in much more harmony and conjuction than Vista devlopers, because the low overall communication between the individual developers makes the code and the interfaces between the modules all that cleaner, and encourages good code documentation.

TFA: Windows code is too complicated. It's not the components themselves, it's their interdependencies. An architectural diagram of Windows would suggest there are more than 50 dependency layers (never mind that there also exist circular dependencies). After working in Windows for five years, you understand only, say, two of them.

If you have 50 dependency layers, wheras Linux or Mac OS X have something like 10 (in the most pessimal case, for things like desktop Widgets), you've probably been spending most of your time writing new ways of doing old things, over and over and over, and you do this cuz

• Your original API was no good (I find this difficult to believe in a few cases)
• You are ignorant of the original API (bad documentation)
• Adding new features to add value (some of this in Vista, not as much as were promised, tho)
• Introducing dependencies helps sell products (hmm, I think you might have something there)

Also, the author puts a bit of attention into "number of lines written". This metric is worthless - if you pay people to code, you will end up with LOTS of code. Debian developers are working for free, are therefore lazy, and therefore write as little as possible to solve as many problems as possible.

Re:5000 lines of code a year?

[UnknownSoldier]

I used to hate schedules, until I learned to respect both sides of the _real_ issues:

1. Schedules tell me that you care more about the product then the people building it. That's one way to kill the company -- let people know that they are only being used, and they will give back the same medicine by refusing to work there. "Hey Joe, tell your friends that company X doesn't give a shit about us dev's having a personal life." It's fatal for a company to lose loyalty of its employees, because you can't motivate people, you can only stop from de-motivating them (E.A. knows that it can burn thru people; there are always more talent to "recruit" -- you think people that left are going to tell everyone what a great company they are to work for?)

2. But if we toss out the schedule, nothing will ever get shipped, because a program without a deadline is never "good enough" to ship -- there are always more interesting features to add. It will always be in a state of R & D. This is one reason why the majority of OSS sucks. It's someone pet project that is no longer being maintained. It's a hack job with no real thought towards maintaining the next version. Fortunately we have counter-examples like the Linux Kernal, Mame, Firefox, etc, showing us the right way to do things.

Unfortunately both extremes / views are flawed. The weakness is the other's strength, and vice versa. A schedule is a necessary evil. It's _supposed_ to be a way to balance theory + application -- to balance the goal, of making money by shipping a product, and by giving the creators (devs) enough time to build something cool that others will find useful.

Cheers