Dr. Jerry Pournelle

Email Me

Why not subscribe now?

Chaos Manor Subscribe Now

Useful Link(s)...


Hosting by

Powered by Apache

Computing At Chaos Manor

Mailbag for September 11, 2006
Jerry Pournelle jerryp@jerrypournelle.com
Copyright 2006 Jerry E. Pournelle, Ph.D.

September 11, 2006

Mail was very light this week.

Jerry, here's the review of the router I told you about last Thursday evening at LASFS:

Last week, my router started getting flaky. It would stop transmitting packets, and not start up again until I power-cycled it. When it got to the point that I had to do this several times a day, my sister and I decided to get a new one. As we don't have much money, price was a consideration. We ended up with a TRENDnet router, model TW-BRF114U to be exact. The quick-start guide said that I could configure it with either Internet Explorer or Netscape, but I figured that if either of those would work, so would Firefox. Instead, the router would hang every time I tried to save the configuration. Not only did I need to power-cycle the router, I had to reboot my Win98 SE machine because it suddenly had a non-routable IP address, and trying to renew it failed.

OK, it's time to call tech support. The manual gave a number in LA county and said that tech support was available between 7 AM and 6 PM Pacific Standard Time, Monday through Friday. I can only guess that this company completely ignores Daylight Savings Time. Calling, I listened to the phone tree and selected Customer Service as the closest choice available. After a short wait, I was speaking to a human being. Alas, she was unable to help me because she's not tech support; tech support is available 24/7 at a different number, which she gave me. If I had needed support for the first time outside of office hours, there would have been no way for me to find out, as their phone tree doesn't inform you of how to get tech support.

I called the new number (an 866 number, unlike their main number) and only waited five minutes to speak to a tech. No complaint there! He knew what the problem is: their router's setup pages work ONLY with IE and Netscape. Yes, I know that they are the recommended browsers, but I had no way of knowing they're the only supported browsers. Considering that most routers today run Linux inside, I find it hard to understand why it doesn't support the most common browser used under Linux.

Now to configure the router. The setup pages are clear, and well-written except for one thing: there's only room for one DNS number. When you save the configuration, the router restarts, taking 25 seconds, then refreshing the page. This is where it hangs under Firefox. The tech I spoke to told me that there's "something in the Firefox Security settings that blocks what we're using." If they're using anything more complicated than a simple page refresh with a 25 second delay, they're using the wrong thing.

After I got it working, I looked around the rest of the config. Among other things, you can do Port Forwarding, URL blocking and other neat things, all under Advanced, where Aunt Minnie would never dare to look, even if she were aware she needed them. It can also send your current IP address to a Dynamic DNS site, if you have a cable/DSL modem that passes through the public IP, not exactly the most common thing.

There's also a place there, if you poke around, to add additional DNS numbers.

Suddenly, at this point, I stopped routing and couldn't get it back again. I called Tech Support again, and the tech got my number so that a senior tech could call back. As it was about 5:30, I asked for a call within 30 minutes. It came at 3:30 PM the next day, long after I'd solved the problem on my own by putting in the router's IP as a Default Gateway. I don't know why this was needed, as Linux doesn't need it, nor does my sister's Win2K box, and it was probably a temporary issue.

One thing I wasn't able to make their tech understand was that if I'd have decided to boot into Linux and use it to set the router up, I'd never have succeeded unless I could hook the computer up directly to the DSL without the router and find a copy of Netscape for Linux because the only browser I have under FC 5 is Firefox. If I were running a pure Linux LAN, I'd have had to borrow a laptop with Windows on it to set up/administer my router and that's not a Good Thing.

Now that I've gotten past the configuration problems, the router seems to be working OK. The programmers put a lot of thought into providing as much functionality as they could in a low-priced router, suitable for Aunt Minnie. Alas, whoever built the interface doesn't know how to write browser-independent HTML, and whoever wrote the manual didn't make the fact clear. When I add in the problems getting to Tech Support and the absurd delay in returning my call, I can't in conscience recommend this router. If, as and when they address these issues, I may change my mind.

Joe Zeff

I think I see a different moral to this story than you did, but it's interesting. Thanks.

Your page: Chaos Manor, Column 313, Part 3 - The User's Column, August 2006 has a discussion of GPU versus multi-core (and more powerful, in general) CPUs.

First, I agree with the statement that graphics processing is different from general purpose processing, that there is room for one thousand to one million improvement in processing power, and and further that this is gain is unlikely to come from increasing CPU speed by a small factor or increasing the number of "cores" to 64 or so.

However, memory bandwidth is will be (and might already be) a limiting factor for multiple general purposes CPUs in a system well before 64 cores are reached.

Therefore, even to get the thousand or so performance inprovement that might be reached for general purpose CPUs, there will have to be changes in the memory model, which is one underlying memory with a few ports and a relatively small (compared to main memory size) local cache for each core and synchronization of cached between cores. (Nothing new here, just the same things that have been discussed for parallel processing for more than 30 years.)


I replied before seeing: "Computing At Chaos Manor", Mailbag for August 28, 2006

Regarding email from Charlie,
   "Subject: multiple cores, and memory access"

The problem addressed is the one that I was talking about, but the implementation can't be done with multiple address and data lines.

I think that the parallel processing message passage archtectures of the 1980s to early 1990s would be useful in the range of up to a few thousand CPUs. Elxsi (link) had an architecture that seemed nice when I looked at it in the 1980s, but I don't remember the details enough to say for sure that it would work with chip level multiple CPUs.

To Be Continued...