Guy Kewney's (late) Weekend Diary

Monday

This was the week where a great friendship may have come to an end. I refer to my love affair with Laplink.

I absolutely couldn't live without Laplink to reach my office from home. It gives me incredibly fast file transfer over a modem (typically, I move 50 meg worth of files at an average speed of one minute per meg or faster) and it also give me remote control of my Windows 95 desktop. This week, however, I've just had to live without it.

Monday was when it started; the modem at home dropped the line half way through the transfer. Home staff (as I delicately describe my wife in this context) kindly rebooted the modem, and the rest of the transfer started apparently normally. Alas, when I got home, I found the line had dropped again.

Considering that Laplink users typically work from home, I feel it's kinda dumb for Traveling Software (sic) to offer phone support only from nine to five. I mean, you get home at the end of the day, and dial the office, and there's a problem. Who you gonna call?

I called San Jose. They talked me through the problem, and the more we did, the worse it got. Eventually, it got to the point where Laplink couldn't even open files on the local machine, never mind see what was on the remote one.

They made me delete the Laplink.INI file. That cured the immediate lock up. "Do you think that will cure the modem problem," I asked. "Well, what I think is, after all these months of hard work, the INI file just got a bit tired," said the laconic help desk wizard. "Wore out."

Not a good day otherwise. Peter Coffee, brilliant guy at PC Week in the US, has written a book on How to Program with Java; published by Macmillan's Que tech books subsidiary. It comes with a CD. On the CD is software: Symantec's Cafe, Kedit, Java Workshop from Sun... I gave it to a young friend who was keen to learn, for evaluation.

My young friend gave the book back after the weekend. "I'm useless. I can't make any of it work. It's too hard. I'm stupid." His mother tells me his self-esteem is not in need of this sort of thing, and if I have any more "help" to give like this, I can help someone else. So, puzzled, I run the CD.

The Sun Java Workshop is an evaluation-only version. It expires in August 96. Symantec Cafe turns out to be Cafe Lite. No documentation at all, just a readme file which I could upload here without significantly adding to the length of my diary, or significantly challenging anybody's learning skills. Also, doesn't seem to work.

It's so annoying! Stupid, stupid, stupid; checking with Macmillan I find that they just didn't run the software and so didn't know they'd built the CD with the wrong version. And the author, they say, had wanted to include documentation, but they decided it would take too long. And the book is actually well written, and probably a very good guide to Java programming.

Coffee is really his name, by the way. I happen to know him. I don't dare ring him up, because if he were me (if you see what I mean) he'd be hopping mad, and probably say something indiscreet about his publishers... whatever, I don't think his next book will come out from Que.

Tuesday

An angry phone call from "someone at Motorola" who won't give a name.

"That was a very irresponsible piece you wrote about us battling with IBM over the cost of new silicon foundries for next-generation processors," he explains. "You should perhaps not write about things you don't understand."

Well, well; I knew it was a hot story, but I didn't realise it was that hot! Time to dive into the silicon industry, I think.

It turns out that Digital Equipment, Intel, Motorola and IBM are all quietly wetting themselves about the size of the future giant desktop system. Fact is, you don't need a very powerful processor to run a Network Computer. If NCs do take off, the market for desktop PCs will stagnate.

Today's market just barely supports the also-ran processor makers. Intel, of course, makes 50 per cent profit on its processors. Everybody else has to charge less, and sell fewer machines. Well, they make money, but not as much as they'd like.

But when Andy Grove (chairman of Intel) says he has an "economic mandate" to invest in a $10 bn fab for the next generation of processors, he's not joking. What he means, of course, is that he hopes he has one, because the chips you need to make to keep Moore's Law alive in 2010 would be 10 GHz processors about three inches square, and you'd have to sell five times as many per year as Intel sells Pentiums today.

But, says my source, nobody has a fab capable of making the processors for 2003, never mind 2010. Dec's Hudson 3 plant is making amazing Alpha processors today, but it's nearing obsolescence. IBM's processor technology is a lot better than most, but it runs out of steam inside four years. Motorola (as I mentioned a fortnight ago) has other irons in the fire.

Actually, not irons, nor a fire; ColdFire is what is driving Motorola forward. It's a re-engineered family of 68000 processors, miles cheaper, and already equally powerful. So the ColdFire 5102 you may first have heard about in 1994 is as powerful as a 68040, and already antique; we now have numbers like the 5204 and 5206, and clock-doubled third-generation ColdFire processors are now emerging with superscalar machines due next year.

As little as six months ago, authorities were saying that ColdFire was "not much better than the ARM or PPC" architectures for embedded (things like car control systems) processing. Now, they are seeing Motorola's investment in technology start to pay off; it's racing ahead of rivals, and the better it does, the more Motorola wants to invest in ColdFire, and the less it wants to invest in PowerPC.

Tricky, because you see, Microsoft has released Windows CE for PPC, and

Motorola will have a CE machine out by summer.

Watch this space; it's going to get nasty before they all see sense, and set up a consortium silicon fab with licences for processors. I think it'll be an Intel fab, because of one reason; Intel processors are useless for embedded applications. But we'll see.

Oh, Laplink? -- still ongoing. This was the help-desk day. I discovered the modem still dropping the line. What was interesting was that the modem wasn't doing it. Laplink was doing it. I know this because I would see the message saying: "Line dropped" but the modem lights were still all lit. And if I asked Laplink to take over remote control of the office machine, it did. The file transfer module insisted there was no modem link, and the modem was "busy." Yes, indeed.

The Laplink help line goes through to Amsterdam. The help desk operative was French. He spoke some English, though it was difficult to be sure how much because he had a faulty microphone in his phone headset. If he put it in front of his mouth, it kept bumping his chin, and generating static. And if he moved it away from his mouth, I couldn't hear him.

"Installed any new software, printers, that sort of thing?" he asked.

No, I hadn't. "OK, that makes it tricky...try this:"

He made me stop using the Windows 95 TAPI drivers, and set the software up to run Laplink directly. Turns out Laplink doesn't know about obscure modems like a Hayes Accura 288, or a Courier DS V.34 (market leading brands). We did our best with "compatible" options. Obviously, I could only do this at one end; so I told him I'd call back, and set out for the office. "Where have you been?" asked my colleagues. I think I was relatively even-tempered in my reply, but apologies, chaps, if I was rude...

In the office, I set it all up as at home. No improvement. So I rang the

Frenchman. He was out; I spoke to an Irishman.

"So it says here, you've installed new software and a new printer, and it's all stopped working?"

Argh!

Wednesday

Thanks for the memory, Hyundai. Well, all I can say is, I'm damn glad I bought my "variance controlled ram" from Vanguard last week. this week, memory is unobtainable, because of politics.

As I understand it, it has to do with Hyundai versus Dresden.

Dresden is a new town built on a pile of rubble in what was East Germany; employment there is at a premium. There's a factory there owned by Siemens, here they make semiconductors -- primarily SIMMs for PCs. And in order to keep 2,500 odd folks in work, Germany has to invest in new silicon foundry equipment.

Now, opinions differ on what is a "fair price" for memory. This time last year, memory prices were definitely over the odds; there was a shortage resulting from the launch of Windows 95, if you recall, and prices just didn't drop the way they normally do in the silicon business. And then there was the famous "golf course" meeting between all the Korean chip makers, and they tried to preserve the delicious profit margins they'd all started making. Naturally, it didn't work; one of them broke ranks.

Since then, it's been a steady escalation, ending up with genuine dumping. Last week, you could get an eight meg SIMM for $21, or a sixteen meg SIMM for $48.

I think we should look inside Hyundai for the person who led the race; their pricing was described by one competitor as a "scorched earth" policy. "Last one standing owns the field," said one bitter-sounding rival. So, OK, that's what you'd expect rivals to say. What's interesting is that the Korean Government appears to agree.

It seems that Siemens went to the EC and asked for support; and when they didn't get much (German protection of German industry isn't popular, I think, in Brussels) approached Motorola and Texas Instruments. They put serious pressure on the Korean Government: "Do something about these prices, or else..."

So Korea's lords and masters told LG and Samsung and Hyundai to reduce memory output by 37 per cent TODAY! or else... and the next thing we knew, everybody who had stocks of DRAM had put a stop on sales. You can't get them. They aren't answering the phone. They knew whatever they have today will be worth 20 per cent more tomorrow.

I expect it all to end in March, when the EC will impose FMV again, Fair

Market Value. It will mean, once again, that memory is more expensive in the European markets. As I said, I'm glad I upgraded my Performa to 32 meg last week...

Thursday

A reader wants to know: "I've got a Pentium machine, 120 MHz. Can I upgrade to MMX?"

Yes, if you have a socket seven, is the official Intel response. Or is that Socket Five? Intel will get back to me. "Why don't you ask at this evening's MMX party?" says the marcoms department. A nice idea, says I, but I have a reader on the other line; can't I talk to a product manager? "They're all in America for training."

The party is at Planet Hollywood. Budweiser "beer" of the American, not Czech variety. And there are lots and lots of Intel product people there. "How can I tell whether a PC is upgradeable to MMX?"

They don't know.

"Maybe we'll put together a list of OEMs who claim to be upgradeable,"

suggests one of them. But no, they don't have any way of running a software check on a PC to see if it will support MMX upgrade. "Quite a good idea, though; we'll look into it."

Call to one of my best silicon sources, for insight on the Motorola versus IBM battle. "Sorry, too sensitive to comment," he says. "I've been specifically asked if I've spoken to you about this. You're in trouble, Guy."

Hmmm.

Friday

Traveling Software has now made me switch modems twice at both ends; the

system is steadily degrading. "We'll send you a fax with a new initialisation string," says the nice Irish support guy. [There's a short silence] "Oh. It seems our fax server is down. Here's the init string: control-Em, tilde, AT ampersand Eff,..."

It doesn't work.

"Tell you what, I'll call into your office machine, and we'll establish that we can control that."

He can't.

"I'll call back."

He doesn't. Actually, he may have done, but BT chooses this moment to throw its teddy out of the pram; all incoming direct-dial lines are "temprilly out of order," says a semi-literate robot.

Lunch with Dave Sikkora of Fourfront -- makers of Web Wacker and Web printer, and gurus of "content management." Content management; everybody's doing it, and nobody knows where it's going, have you noticed?

Prediction of the day: 1997 will be the year of client-server software on the Web. Oh, and next prediction; 90 per cent of it will not work properly.