Tuesday, September 20, 2005

More Wikipedia Weakness

It's often tempting to give Wikipedia a pass. After all, many of its most egregious errors get fixed over time--at least if the topic isn't controversial and the article has had enough time to "settle down" from breaking news or changing facts. But, as I've commented before here, here, and here, when Wikipedia is bad, it can be pretty bad.

Case in point, the other day I happened to run across an article on Data General AViiON servers, a topic with which I have a more than passing acquaintance. I was the product manager for the first AViiONs and handled marketing for many products of successive generations. Now this is the type of article on which one would probably be inclined to trust Wikipedia to get things more or less right. After all, it's an uncontroversial technical topic of the dead (but not too distant) past.

Don't trust those instincts. Apparently the article is also sufficiently off the beaten path that it hasn't had a chance to benefit from the Wikipedia "community" because it's rife with howling factual errors. It has SCO writing DG/UX (Data General's flavor of Unix) for example, contains basic misconceptions about Non-Uniform Memory Access (NUMA) architectures (an important piece of the AViiON line), and its storyline about DG's historical competition with DEC is pretty far off-base. But my intent here isn't to belabor the details--many will probably be fixed eventually, perhaps by me. And, I've seen trade press articles almost as bad. But this serves as yet another cautionary tale. Even where Wikipedia "should" be good, it often is not. Exercise appropriate caution.


Anonymous said...
This comment has been removed by a blog administrator.
Anonymous said...

Your article sucks!!!
Where's your brain???

Anonymous said...

I have rarely (three or four times), used this Wikipedia. I therefore did not know of it's quality, although an article by a 'EB' editor made me believe it was of very poor quality. The only DG history I know is from the "Soul" book, and even THAT limited knowledge shows me that this article on AViiON computers, and DG in general is not worth reading. I doubt you should try to correct it. Publish a book, yourself, on the latter days of DG (post 'Soul') and I'll BUY it !!


Gordon Haff said...

Hi JB,

I'm perhaps a bit less convinced than Andrew that Wikipedia is of consistently poor quality. However, my explorations of late convince me that thee's a lot of bad there and it's generally of not much better quality than the average Google search. Which may be a resonable standard for a random free source of information but not so good for THE Internet encyclopedia.

I find it a useful research tool, but I use it with skepticism.

Anonymous said...

Seeing as I actually wrote the article in question, I feel somewhat slighted. At a minimum I feel I should at least post where the data comes from.

With the exception of the naming issue, which I did not add, the sources of the data are primarily conversations with the authors of two web sites, most notably http://badabada.org/about.html. It's not like I just pulled this stuff out of my ass, I spent the better part of a week's worth of spare time trying to collect the information in question.

But of course you don't exactly outline which "howling" errors you found, except in rather nebulous terms -- for instance you seem to have a problem with my comment that DG's policy was to essentially chase Digital. This doesn't seem to be too "howling" to me. The SCO reference does seem to be outright wrong, but I can't recall where I originally saw that one.

But of course "Ghuff" did exactly what is expected to do when faced with these sorts of issues -- fix them. The article is clearly much better now (with the exception of the musings in the naming section, which should simply be removed), and assuming Gordon's memory is better preserved that the people I contacted, much more accurate.

Now let's compare: pull a copy of Britannica some time and look up Seymore Cray. It contains a number of outright incorrect facts (ie, he did NOT work on the Univac I, and Steve Chen designed the X-MP) and additional rather misleading statements, like Cray started with multiprocessors (not only not true, but anyone familiar with his history will recall that he hated the very concept). And it does this all in only three paras.

So by any measure, this is yet another example of the stunning success of the wiki. Right? Am I missing something here?


Gordon Haff said...


First of all, please note that, perhaps unlike Andrew, I don't think I've made a judgement that Wikipedia is "bad" or even (necessarily) fatally flawed. Rather, I've pointed out inaccuracy--which clearly does exist. And the fact that I and others can correct errors doesn't change the fact that those errors existed at a point in time. (And, yes, there are lots of errors in other reference sources as well. However, my experience is that there are more in Wikipedia than in Brittanica. However, in any case, my post wasn't comparative but illustrative of Wikipedia specifically.)

As for the DG AViiON article, since you ask. I'm sorry that you feel somewhat slighted but there truly were quite a few fundamental errors.

AViiON was not a series of NUMA multiprocessors. (Only the high-end was NUMA and only starting several years after AViiON launched.)

I have no objection to the comment that DG's strategy was essentially to chase Digital. I couldn't agree more. My objection was to the statement that DG decided to stop attempting to follow the dying DEC market; it wasn't dying at the time. DG was simply unable to follow DEC.

The SCO reference, as you say.

There was no "NUMA software" to port to NT. There were NUMA optimizations in DG/UX and some other software but nothing was ever ported to NT. (NUMA's just a form of SMP where memory access times vary depending on relative physical location.) Indeed, NT never ran on multi-block AViiON NUMA servers.

Sales slowed because of Unix consolidation and competition from commodity NT boxes, not for the reasons stated.

I agree that the name musings should just have been taken out. What you wrote has been widely reported and seems logical. It just isn't true as far as I know.

I appreciate that you did put effort into writing it, but a lot of it was frankly incorrect in some fairly basic ways. I'm not making a value judgement, but making a factual observation.

Anonymous said...

Trust me, I don't feel slighted by your pointing out errors! Far from it, I absolutely love it when that happens. I've had e-mail from the original author of the BASIC Star Trek game (yeah, THAT one), the author of ZGRASS, and any number of early Mac products that most people forget even existed. In one case I found a blog entry from someone on the Lotus Improv team that was convinced I was a member of the programming staff due to the historical details I included (for the record, I wasn't).

I really do try to do what the pooh-poohers claim doesn't happen -- I research my articles in depth, often buying reference books (ie, The NEWS Book, A Few Good Men from UNIVAC, etc.), using Toronto's online library for a lot of it (it has most computer magazine articles back into the 1980s) and tracking down the original team members on the 'net.

But, as you can see, that doesn't always mean the article ends up "correct". There is a definite risk that no matter how many sources you collect, when you weave the article together into a whole you end up telling the wrong story. In fact it's even scarrier when you get all the facts RIGHT and the narative is still wrong.

No, my feeling of being "slighted" is that your post seemed to be implying I just basically made it up. No, you _didn't_ say that, but it certainly _felt_ that way. Perhaps it was the word "howling".

But while we're here, I do have an additional question. You note "There was no "NUMA software" to port to NT." But I seem to recall that DG did indeed offer a complete line of such solutions, IIRC starting with the AV 86xx's? Were these not NUMA machines? What about the AV 20xxx's?

Gordon Haff said...

Hi Maury,

I'm not sure of the model numbers although AV/8000 series sounds right. In any case, they weren't NUMA systems, but rather uniform memory access 8-ways. The original plan was for a system based on a chipset from Corollary (later became Intel Profusion when the company was purchased by Intel). It used a multiported memory controller to basically bridge together the front side busses of 2 4-ways and all the memory. IIRC, the Corollary design was delayed and DG filled the gap with a similar 8-way design from a now-defunct company called Axil.

The AV/2000 (and successor AV/25000) were NUMA designs that connected together a version of Intel motherboards that had a special connector used by DG's (and Sequent's) SCI interconnect. These two servers were basically DG/UX only. (NT ran on a single block; it actually "worked" on larger systems but the scaling was terrible because NT, at the time, had no physical locality optimizations--and therefore DG never sold it.) The 88K-based AV/10000 was also a NUMA design but not one based on "commodity" building blocks like the AV/20K & AV/25K.

I know it's hard to get this sort of research right and memories become faultier over the years.:-) I had even forgotten that DG's first NUMA system was actually the AV/10000 before I dug into this again.

Anonymous said...

Don't whine, change the article! It is really as simple as that.