I can see why RH would want a facility in the SF Bay Area. The bay area has the highest population of software and hardware engineers in the world. Even on my short street I know several engineers or people working in related fields. The drawback is that there is a shortage of engineers here so they're competing with other companies (i.e. Cisco) for good engineers.
As for disasters, I think far more people in ths country are killed by tornadoes, hurricanes, thunderstormes, etc. than earthquakes. All modern buildings in CA have strict building codes designed to withstand an earthquake. At least it is possible to design buildings to withstand earthquakes, it's a bit harder to design a building to withstand a tornado or hurricane.
Besides, the weather out here is quite pleasant in the summer and the winter is fairly mild.
As a side note, what will happen when the New Madrid fault goes or the fault that runs along the eastern edge of the Rocky Mountains? At least here all the buildings are designed to withstand an earthquake. If those other faults go all bets are off.
The script kiddies are going to love this. I'm on a cable modem and run a Perl script called booby (see http://members.home.com/lazyx/booby which emulates BO. It's interesting to see how many script kiddies try hacking in without knowing everything they do is emulated and being logged. Most of the kiddies I see don't really know what they're doing, but I've seen some pretty malicious people out there.
The potential of this program is fairly large. If someone made an installer that would search out other systems on the LAN and install it on them as well this could be a nightmare (shudder) for Micro$oft shops. One more reason to not use M$ products.
Of course *NIX can be vulnerable as well to this type of trojan horse. The user security of *NIX may be better, but security is only as good as the user using it. The main difference, I believe, is that *NIX users are a lot more knowlegable about their systems and are much less likely to download and install software of questionable origins.
Unfortunately this won't work. In my case, if the upstream bandwidth exceeds 128Kbps the cable modem just starts dropping packets. What the modem should do is only drop packets which contain data and not ACK packets.
Also, I have been on @Home for 2 1/2 years. In the last year the service has become terrible. When I first joined, I received email about billing, was notified ahead of time about downtime, and was treated with respect. Tech support was excellent and the hold times were low. Reliability was high. Also, the AUP I signed said absolutely nothing about running servers and they were accepted.
Now it's the opposite. Mail and news is frequently down. 20% packet loss when trying to leave the @Home network is a frequent problem. Performance sucks. I no longer receive any email about billing (only spam from @Home). Hold times for tech support are frequently around 2 hours. All this and my rates have gone up.
I would be willing to pay extra to be able to run a server again, but @Home will not support @Work in residential areas, and if they did it would cost around $400/month! My cable modem is capable of 10Mbps full duplex. @Home should offer tiered service where one can pay more for higher quality of service.
My employer also has @Work. The packet loss is also very high at peak times.
The packet loss is not over the local links, but as soon as packets leave the @Home network. @Home does not have adequate peering with the rest of the Internet.
It could be enhanced. For example, it would be really cool to place a sink within a sink! That would vastly increase scalability and make it hierarchical more like a traditional file system.
For example, you expand the first sink until you see your personal document folder. You select that sink. From there you expand it until you see a folder for notes from your brother and you select it, and so on.
What's really cool about this is that it adds a sort of three dimensional method of accessing files rather than the simple tree and file list.
As a TCI@Home user I am quite pleased with the ruling. @Home has totally mismanaged things in my area. Packet losses of over 20% are not unusual. The mail server frequently crashes, and their tech support is totally unacceptable. It takes over 3 weeks to get a response from a technical question sent via Email, and hold times on the phone are measured in hours.
What really makes me mad is the fact that @Home spends billions on companies like Excite, yet their internal infrastructure is totally inadequate. @Home is trying to become another AOL. I don't want AOL, I want a high speed Internet connection.
Not only that, but my rates have gone up while my bandwidth has been capped.
I have been an @Home user for 2 1/2 years, and during the last year the service has gone to hell. When I first joined, there were no hold times for tech support. The tech support people were sharp and knew what they were talking about. The performance was excellent. The uptime was excellent. They sent out email telling us ahead of time about downtime. They also emailed us our monthly billing info.
Today the only email I get from @Home is spam. I don't get a monthly bill from them (except on my credit card). Downtime is never preannounced, and tech support always responds by saying "Clear the cache in your browser" as if that will fix the 20% packet loss.
Oh, the packet loss is not caused by my connection to the cable company. The packet losses are due to the fact that @Home has inadequate bandwidth to the rest of the internet. The packets are always lost as soon as traffic leaves the 172.16.x.x routers.
Another nice feature of Token Ring is the large frame sizes that are supported. I don't know about 100 or 155Mbps Token Ring, but the old 16Mbps Token Ring supports packets up to around 18K. Larger packets greatly improve bulk data transfer. Of course, ATM can do the same thing, with AAL5 supporting PDUs up to 64K with RFC1577 or 18K with LANE.
On a properly configured server, NT should be stable enough. When I worked at Adaptec and when our drivers were fully debugged, the servers were fairly stable when benchmarking and stress testing. The testing required for servers WRT NT is fairly extensive compared to client computers. I know, for example, that Dell's server stress test is very extensive. Very few network card vendors passed the test.
On ordinary consumer hardware, yeah, NT isn't that stable. Also, remember that these benchmarks don't match the real-world. In the real world a lot of stuff occurs that doesn't occur during benchmarks or stress tests which can cause NT to BSOD. Also, for the benchmarking, NT will likely only have to stay up for a few hours.
NT might very well beat Linux. Linux still has some shortcomings compared to NT for the test Mindcraft is about to run.
1. NT supports zero copy transmits out of the disk cache or the web server cache. This means that the data is directly DMAed by the NIC card out of the disk cache or user-space memory buffer out to the wire without having to be copied to kernel space. Linux (as of 2.2.7) does not support this (this requires changes to the Linux virtual memory code and to the networking code to support fragmented packets and locked memory).
2. NT (with Intel cards) supports Fast EtherChannel if connected to a Cisco switch. Linux does not. Fast EtherChannel, for those who don't know, allows multiple ports (up to 4) to be ganged together to behave as one virtual fat pipe (800Mbps with 4 100Mbps adapters running full-duplex, the usual configuration).
3. Intel adapters on NT can offload the TCP checksum calculation to hardware. Linux can not. I don't know if SP4 added that support to NT4 or not. It is definitely in NT5 (I worked on some code to take advantage of it).
4. If NT uses NetBEUI it will beat Samba. NetBEUI is much faster than NetBIOS over IP.
When I worked at Adaptec and we were competing against Intel for server tuned NICs I saw what NT could do. With 6 network cards set up with load balancing we saw over 500Mbps of throughput, and the server was less powerful than Mindcraft's. Intel wasn't too far behind us either. This was with NT4+SP3 on a quad 400MHz server.
Now, personally, if I had a choice I would almost always choose Linux over NT, but I'm just trying to be realistic. One of these days I need to sit down and hack some Linux networking code. Fast EtherChannel support would be a very nice addition to Linux.
I periodically go through the same thing with TCI/@Home. It seams that about once a year they change their billing system so it bills me twice. Then it takes 4 months for them to straighten it out.
TCI/@Home is totally screwed up. Here in Fremont, CA. it got to the point where we (the users) have organized and appealed to the city council. They also are quite annoyed with @Home's terrible customer service (typical hold times are measured in hours, not minutes) and terrible performance (20% packet loss to @Home's NAP connections).
Things have improved recently, but for a while my 28.8 was a *MUCH* faster than my 10Mbps cable modem.
As to the city council I went to the meeting. It was really fun watching @Home squirm. TCI tried to claim that they fixed the customer service problem. The city council then had the city manager call TCI/@Home customer service right there during the meeting. They couldn't get through. TCI had fun trying to backtrack and wiggle out of that one. TCI/@Home offered to donate $40,000 to the city for any program the city chose if the city agreed not to fine them (the fine was $18,000 for lack of service). The city council was ready to accept until the customer support fiasco and went ahead and fined them.
It was a very entertaining city council meeting and I hope to go to the next one. The city gave @Home 120 days to fix their problems or they'd get a new fine. The city council is also going to pass legislation requiring minimum standards. This will be the first city in the country to do so. All this because of TCI/@Home.
If Redhat is going IPO and if Redhat is smart, they'll introduce some form of poison pill. For those who don't know, a poison pill is basically something which goes into effect if there's a hostile takeover. A poison pill could be just about anything, such as all stock options become fully vested immediately. Poison pills can work quite well. Even without one, M$ wouldn't dare to buy Redhat with the DOJ looming over them. The FTC would likely block it in an instant.
Is this consistent with common carrier status?
on
ISP Sues Spammer
·
· Score: 1
I guess you don't remember Agis and spammers then. Agis gave the spammers a free reign on its network. The spam flowed like the mississippi and everyone got pissed off at Agis and firewalled them off. Agis now has about the best anti-spam record in the industry.
I'm sorry, but how would you feel if you were an ISP and someone ran a huge spam run against your users and your mail server ground to a halt under the load, with millions of undeliverable messages and what not.
I have yet to have a Micro$oft mouse last more than 6 months on me. After 6 months (even with careful cleaning) they always start mistracking. My collegues keep wondering what that swearing and banging is when I start banging the broken mouse against the desk in frustration. I have yet to have an optical mouse die on me either. I use an old Mouse Systems 3-button optical mouse and have never had a problem (except that the pads on the bottom are wearing out).
It very well might be that NT did beat Linux. Linux still has a long ways to go tuning wise. Let's face it, most Linux programmers don't have access to a quad Pentium III server with an intelligent RAID controller and multiple NIC cards.
Linux still has a ways to go. As someone who has worked on a number of device drivers (some of which were NDIS) I must say that NT could very potentially be faster than Linux as a web server.
For example, NT supports zero-copy transmits for TCP traffic. Linux (as of 2.2.6) does not. NT 5 supports offloading the TCP checksum calculation to hardware, which many of the newer NIC cards support (i.e. Intel and 3COM), Linux doesn't support this. NT 5 even supports hardware TCP segmentation assistance, again, Linux does not.
Copying data is painful and should usually be avoided. Linux as it currently stands must copy all of the data sent out over the network due to it's current architecture of using one buffer per packet. Before someone tells me that it can, first go read the source code (tcp.c and eepro100.c). Linux also does not support TCP checksum acceleration or TCP segmentation offloading.
I must also say that BSD can also perform zero copy transmits.
Also, servers are whole different animals than desktop computers. Servers often have much faster PCI busses (64-bit and/or 66MHz), much faster disk I/O capability via intelligent RAID adapters, and use certified networking hardware that is held to a much higher performance and reliability standard than desktop PCs.
Also, I know for a fact that the Intel NT driver has been very carefully tuned. At my last job we went head to head against Intel in the server NIC business. Whenever we would surpass them in performance, they would turn around and surpass us. In the end we beat Intel, but only because our descriptor format was simpler (modeled loosely after the DEC Tulip architecture with tons of interrupt reducing features). We spent many months tuning our drivers on quad XEON servers to handle multiple adapters and load balancing (Fast EtherChannel and our own proprietary method). We were doing over 500Mbps on a quad 400MHz XEON server with NT 4+SP3.
The last time I communicated with Alan Cox it sounded like a lot of work was needed to add the support for proper memory locking (especially on SMP boxes) to provide the zero-copy scatter-gather transmit. Before you say "what about receive?", a typical server has a transmit/receive ratio of 10/1, so it is very critical.
Now I'm not saying that NT is faster than Linux, only that it potentially could be. Micro$oft has spent a lot of time and effort tuning the networking in NT 5 so Linux needs to watch out.
I'm also not saying that Mindcraft is right either. After seeing their reports on Solaris and Netware vs. NT I am very skeptical about them as a company.
-Aaron -- Why is everything alway's Segmentation's fault?
It may be that Linux faired poorly in their test and that Mindcraft was unable to get decent tuning information. It is also possible that even if Linux were properly tuned it would still lag. Part of this comes from my experience with writing network device drivers (and I must admit I have worked on some NDIS drivers).
Instead of just crying foul, why not run some independent tests on some server hardware. Server hardware is a whole different class than desktop systems. The new systems have faster PCI busses (64 bit and/or 66MHz), faster disk I/O via intelligent RAID adapters, fast network adapters, and so forth. Hardware certification for servers is also much stricter. For example, as of the end of last year 3COM was not certified for Dell's server line because 3COM couldn't flip the bill for performance or reliability. Intel easily trounced 3COM in performance and reliability when we benchmarked.
At my last job we went head-to-head against Intel in the server NIC market. Every time we found new tricks in our drivers to up the performance, Intel would come along and do the same thing. Over our months spent tuning our drivers we more than tripled our performance and Intel made some significant progress as well. Now, the original driver architecture was very lean and clean (well, as lean and clean as possible in NDIS). We did numerous tricks to reduce slave PCI accesses and reduce (or even eliminate) interrupts to drop CPU utilization.
We also implemented load balancing and failover across multiple adapters via Fast EtherChannel and our own proprietary algorithm (which preceeded Fast EtherChannel).
In the end, we finally surpassed Intel, but I blame this only on the fact that the hardware interface was cleaner than Intels, basically modeled after the popular DEC Tulip architecture.
Now, for our tuning we spent many months of careful analysis with very heavy loads. We had 100 NT client computers pounding the server running many different tests, from many different sources. In the end we were able to achieve over 500Mbps throughput on a quad 400 MHz XEON server.
Now, I'm not knocking Linux. I think what needs to happen is for some serious performance analysis and tuning to take place. I am sure that there are numerous areas in Linux that can be tuned and enhanced to take advantage of these servers.
For example, NT 5 has support to offload TCP checksum calculations and even TCP segmentation to the hardware. Many of the new NIC cards support hardware checksum calculations, including 3COM and Intel. NT 5 also supports non-copy transmits, where data fragments are DMAed directly out of user-space buffers, thus eliminating the copy from user space to kernel space. Another feature NT has is the ability to DMA directly out of the disk cache.
Now, looking at the Linux 2.2.6 kernel (tcp.c and eepro100.c) it looks like it cannot support zero-copy transmits. It must copy, since skb's can only hold one fragment per packet. Copying data has a very noticable impact on performance in my experience, and should usually be avoided. I'm sure one of these days Alan Cox or someone will change this. The last time I communicated with Alan it sounded like it would require some major effort to eliminate this bottleneck since it would require some complex memory locking code.
Now, just comparing the architectures of NT and Linux, in this case the transmit path in NT could very conceivably be faster. NT 5 may very conceivably be much faster than Linux since it not only eliminates the copying stage, but can offload the checksum calculation to hardware.
Now I'm not going to say anything about reliability, since Linux almost always wins hands-down, at least according to the posts here, although I have rarely had any problems when I used NT (except for driver bugs).
Also, I'm not saying that NT is faster, either, since I have not performed any side-by-side benchmarking. I also must say that I'm very leery of Mindcraft especially over their reports of NT beating out Solaris as a web server and NT being a lot faster than Novell.
One final thing I might add is that if Linux puts a lot of effort into competing on the servers it very well might have a negative impact on the desktop. Often one must sacrifice simplicity and code size for performance. Supporting copyless transmits, for example, will require more complicated network drivers and more complex code dealing with buffer management, not to mention more complexity needed for the paging code (to handle memory locking) and the SMP code.
-Aaron -- Why is everything always Segmentation's fault?
I totally agree. Likewise the embedded world can be almost as bad (in some cases worse depending on the environment).
I had a lot of fun working on an OS/2 ATM device driver that was around 100KLOC lines of code. It was a dream compared to the 360KLOC NT equivelent driver written in C.
Then came NT 5... Can you say spending a week just trying to get the driver to install after Microsoft released a new beta that broke the install script (which MS wrote to begin with) and WinDBG crashing EVERY FIVE MINUTES LIKE CLOCKWORK?!?!? Then there was the constant crashing MS kept complaining about which was traced down to the fact that MS wasn't updating the registry (which we couldn't duplicate since they wouldn't send us their latest build). I spent 3x as much time just fighting stupid MS issues instead of working on the code.
Dealing with user space is much easier.
After dealing with WinDBG you don't know how privileged you are with GDB.
This is what I'm planning. There's some new technologies such as reflective LCD's and the micro-mirror stuff from TI that should make these projectors much cheaper. Right now reflective LCDs look the most promising. BTW, these are totally unrelated to active-matrix or your typical LCD technology.
I saw Descent on a 10' projection screen. All I can say is it's a totally different experience.
Think about it, you could sit back in that easy-chair and code away without having a CRT in your face. That will reduce a lot of eye-strain.
I expect that within a couple of years the price of these new generation projectors will fall to under $2K and possibly to around $1K.
Last project was a 15KLOC Java app (around 90 classes).
HDTV will not make desire for small sets go away.
on
Low Cost HDTV Cards
·
· Score: 1
The problem with a 19" or 13" HDTV is that it would be a total waste of money. For one thing, the CRT would not be able to take advantage of it (an expensive 19" CRT might, but who sits that close to their TVs). For the smaller sets, DTV is more than adequate.
Yeah, a 32" HDTV set looks awesom compared to a 32" NTSC TV, but why bother with all the extra expense involved. In a year you'll be able to get a 33" HDTV set for a bit under 3K. The problem is that it is like scaling a 32" TV to a 13" TV. It just looks too small. It's the detail. With all that extra detail it's like looking at a movie from twice as far back as the back row of a theatre. You keep wanting to get closer. If I had an HDTV card I'd have my eyeballs about 6" from the monitor (and be dying of eye-strain) watching a movie. I know, I've been watching line-quadrupled TV & DVD on a 19" SGI monitor and I looked at a 33" HDTV set a start-up company is coming out with. Yeah, the picture looks incredible and very sharp, but I'd much rather have a good 32" TV any day. It's such a different experience to see HDTV on a big screen. HDTV on a 10' screen is breathtaking.
As for an HDTV card for the PC, how many monitors can handle 1920x1080? For most of us this exceeds the dot pitch by a fair amount. Even if it matched the dot pitch, I'd have to have my eyes 3" away to see the detail.
I guess if you just want to watch TV on your PC all day the card is fine. But for watching HDTV all the time on a PC monitor I just shudder.
The recommended minimum size display for HDTV is 60". I saw HDTV on a 33" display and it isn't worth it. Heah, you have a great picture, but you're missing the whole experience.
Also, as for the line doubler, there arn't very many line doublers out there that look good. Most of them introduce all kinds of artifacts from de-interlacing.
In the next couple of years, the price of HDTV projection displays should come down significantly. There's some new solid-state technology coming out, both TI's micro-mirror stuff (which some Japanese companies are copying) and reflective LCDs (very different than other LCD technology). Right now the reflective LCD technology sounds the most promising. As for the plasma wall screens, the color on them is too washed out and they put out too much heat (if you look carefully you'll find a bunch of fans in those displays).
I have a high-quality line-quadrupled AV system hooked up to a 19" CRT right now, running 1440x960 until I get my 33" monitor fixed. It is no comparison to a 33" TV for entertainment purposes. It's no fun watching a movie when you need to sit 12" away to see the detail.
It will be a couple of years until HDTV is ready for prime time. When it comes, nobody will want their old TV sets or converter boxes.
I pity those people walking out of Price Club with their brand-new 60" NTSC projection TVs.
Also, HDTV should be able to interact nicely with computers, since more than audio and video can be included in the signal. For example, you could click on an object and bring up more information or treat it like a URL.
While HDTV looks extremely clean, it's a total waste unless you have a 60" display, which is the recommended minimum size. HDTV projection displays will come down significantly in the next couple of years due to some new non-CRT technology coming down the pipe which are solid-state. The new technologies can easily reproduce the HDTV resolutions and beyond.
The other big advantage of HDTV (which a PC monitor will not take advantage of) is the 16:9 aspect ratio, which means that all HDTV broadcast and recorded movies will be wide-screen.
HDTV on a 19" CRT is not much comparison to even regular TV on a good 32" set from an entertainment perspective.
How can I claim this? I have a very good line-quadrupled DVD setup currently running 1440x960 through a 19" monitor (until I get my other monitor fixed). If I'm only a couple feet away it looks great, but that's no way to enjoy a movie. Of course, HDTV is a big improvement over the line-doubled/quadrupled NTSC. Also, for regular TV if the line doubling is not done properly there are all kinds of annoying artifacts that show up due to de-interlacing. Most de-interlacers I have seen have this problem to varying degrees, the exceptions being Snell and Wilcox ($30,000) and a Phillips unit (DVX8000 - $5K) and some units based on a chip from Genesis.
As for combining HDTV and a PC, they're a perfect match in many ways. The HDTV format allows for more than just plain audio and video to be transmitted. I.E. you could click on an object on the screen and get more information on it.
As for recording HDTV, the latest specs I've seen are 3 1/2 hours of HDTV on a VHS tape or 7 hours of DTV, and since it's digital there should be no quality degredation.
Now, instead of running HDTV on a PC monitor I'd love to be able to hook my PC up to an HDTV monitor and play Quake on a 60" screen at 1920x1080!
I know there's a big markup. On the high-end equipment a markup of 50% isn't that unusual, often higher. Part of what you are paying for is for someone to really design something right without marketing constantly trying to figure out how much cost they can cut.
I stand corrected on the price of DACs. The new Burr-Brown DACs have come down significantly in price and are now around $3 in 1000 quantity. The last time I looked over there they were on the order of $40.
As for building high-end equipment, which I have been involved with, it isn't cheap. A lot more care needs to be taken in terms of selecting matched components and using quality parts (i.e. no tantalum caps or carbon resistors in the signal path). None of the consumer-grade equipment I've seen does a decent job of using quality parts. Open up most Sony or other cheap equipment and you'll find it's full of 10% carbon resistors, tantalum caps, cheap electrolytics, and jumpers all over the place on a phenolic PC board with marginal solder connections and a just barely adequate power supply using LM78xx class regulators for the analog section (which are terrible in terms of noise).
Most consumers wouldn't notice much difference with a high-end DAC with their $200 pair of Bose or Sony speakers. Someone with a pair of NHT's, Magnapan, Martin Logan, or Wilson's will definitely hear a difference because the speakers will be able to much more accurately reproduce the sound without muddying it with their own color.
The high end equipment does not make these compromises. The BOM is higher and the quantities sold are much lower (since the average consumer doesn't have a system that could notice it with their $200 pair of Bose speakers). Due to the quanty sold, the manufacturer must charge more to stay in business. If you're Sony or Panasonic then you plan on hundreds of thousands if not millions of units selling. If you're Theta, Pass Labs, Audio Research, etc. then you figure you'll sell less than 10,000 units. If you were to sell 10,000 units at $100 profit each and you have a number of employees you'd go out of business real fast.
This isn't to say that Sony or Magnavox couldn't make a high-end player. They could, but they're constantly looking at how they can cut costs which defeats the purpose.
I can very well see why a good DVD audio player would cost $1000. Have you tried pricing what it takes to build a good high-quality DAC that can handle 20 bits, let alone 24 bits? More than that, you need more expensive op-amps, better quality resistors, capacitors, filters, and so on. A high-quality 24-bit DAC isn't cheap either. The noise floor becomes much more critical and hence a much more expensive design is in order.
For good quality audio equipment you don't get away with using cheap electrolytic capacitors or carbon resistors any more or the really cheap op-amps and voltage regulators. Now you need to move to 1% or better metal film resistors (or better) and polypropolyne/polystyrene caps and either discrete or very high quality op-amps. The power supply needs to be beefier and have better noise immunity. More than likely, a separate power supply is needed for the digital vs analog portions of the circuit.
Oh yeah, now that you're 24-bit you want to use balanced XLR connectors. For that nice clean balanced output you now need two DACs per channel running inverted for a total of 4 DACs.
A starting 24-bit DAC probably starts at $20 each. You're at $80 already for only one piece. Now another $60-80 for op-amps, $30 for resistors and caps. Now throw in the power supply for $80 for a good toroid transformer and double regulated supplies, independent for each channel and for the analog and digital portions. You're up to $250 already and nothing has been covered for the transport or other circuitry. Add video support and AC3 5.1 channel audio and you could easily exceed $1000 for parts alone.
Of course, most of those here couldn't tell the difference between a CD and a cassette, given how fond you are of MP3 format. MP3 has some serious shortcomings quality-wise, and some could be fixed without too much difficulty (i.e. using a logrithmic scale for dividing frequencies rather than a linear scale).
For those of us who really like to try and reproduce music (rather than noise) we do notice the shortcomings of 44100/16. Hell, even consumer equipment typically tries to enhance it (i.e. 8x oversampling) since trying to make a 20khz brick wall filter is about impossible without screwing up the phase and other artifacts.
The filtering involved in the recording side also effects the sound since all input sound must be filtered to reject any signal above or at 22050 to eliminate aliasing.
DVD audio is a long-welcomed addition by hi-fi entheusist since CD's are so screwed up. The recording industry took the minimal acceptable standards and went with those much the same as they pushed cassettes, another bare-minimum standard, or VHS, which was not properly engineered from the beginning (unlike betamax, which used a much better head geometry).
Of course, the DVD audio spec supports several sampling rates besides 96khz and several bit depths from 16-bits to 24 bits. This means those 10-CD collections on those late-night commercials could be replaced by a single DVD.
Also, if they support AC3 and DTS audio DVDs all the better. Full-surround DTS audio would be a nice medium to play with.
CDs are ancient technology, and anyone who wants to replace them with MP3 should go back to their 3" computer speakers and sound blasters.
You don't need to go to an expensive college to get a good degree. I've met many sharp people who graduated from San Jose State. I myself graduated from U.C. Santa Cruz in Computer Engineering. The UC system isn't all that unreasonable either when compared to many of these private Ivy League colleges. The thing is, you need to find a college that has a good cirriculum and offers a wide range of choices. Also, at least at UCSC, I was required to take numerous courses outside my CE major (i.e. music, writing, Latin American Studies, and so forth).
You don't have to graduate from Harvard or MIT to get a good education. Of course that doesn't mean that you'll get a good education down at the local Junior College. The JC's typically arn't in the same calibar as the 4-year colleges.
One other thing college does is to require you to work on things you probably wouldn't choose to do on your own. Yeah, you might think that that class on microprocessor design is boring, but at least you were exposed to it so you would know what it was about. I found some of my most enjoyable classes were subjects I would never have studied on my own.
And finally, by getting that piece of paper, you have demonstrated that you have some skill and that you can provide the effort needed to succeed in a new job.
In my experience, most companies want one of two things before hiring, either prior work experience or a degree. It's almost impossible to get your foot in the door beyond a sys-admin (if even that) without that piece of paper.
I was lucky in that fresh out of high school I was writing assembly code, but that was only through a fluke set of circumstances. Most of my friends had no such breaks.
As for those of you who say "I can earn $150,000/year without college", yeah, you can, but not straight out of high school. It's going to be a much more difficult climb to reach the levels of those with a degree. With that degree you start much further up the ladder, and your job prospects are much better at least for the first few years. After your first few years, experience becomes the dominant factor.
College did several things for me. I was a geek through and through prior to entering college, but college opened up whole new avenues for me that there is no way I could have explored otherwise. Things like building a microprocessor would have been out of the question, or playing around with GL on an SGI (prior to Open GL).
Many of the things I learned in college were invaluable, besides just that piece of paper. Other things were not very valuable. There were a number of classes that were basically a waste of time, but that was just preparing me for the real world.
Perhapse it depends on the college. I just interviewed a new college grad for an entry-level embedded programming job, yet the grad couldn't perform simple things. I asked about the difference between a linked list and a binary tree and how they relate to Big-O notation when searching. No answer. I asked the grad to write a C function to convert an integer to an ASCII string. Again, the grad was at a total loss.
For those who say they got nothing out of college, either you didn't want anything out of college or you were some super genious before entering. Either that or you went to some brain-damaged college.
Prior to college I had done a fair amount of programming and exploration. I knew 80x86 assembly cold and all the main data types used. In college I was able to greatly build on my experiences. Also, that piece of paper has been useful since it allows me to get a lower insurance rate.
Slashdot Mirror
I can see why RH would want a facility in the SF Bay Area. The bay area has the highest population of software and hardware engineers in the world. Even on my short street I know several engineers or people working in related fields. The drawback is that there is a shortage of engineers here so they're competing with other companies (i.e. Cisco) for good engineers.
As for disasters, I think far more people in ths country are killed by tornadoes, hurricanes, thunderstormes, etc. than earthquakes. All modern buildings in CA have strict building codes designed to withstand an earthquake. At least it is possible to design buildings to withstand earthquakes, it's a bit harder to design a building to withstand a tornado or hurricane.
Besides, the weather out here is quite pleasant in the summer and the winter is fairly mild.
As a side note, what will happen when the New Madrid fault goes or the fault that runs along the eastern edge of the Rocky Mountains? At least here all the buildings are designed to withstand an earthquake. If those other faults go all bets are off.
The script kiddies are going to love this. I'm on a cable modem and run a Perl script called booby (see http://members.home.com/lazyx/booby which emulates BO. It's interesting to see how many script kiddies try hacking in without knowing everything they do is emulated and being logged. Most of the kiddies I see don't really know what they're doing, but I've seen some pretty malicious people out there.
The potential of this program is fairly large. If someone made an installer that would search out other systems on the LAN and install it on them as well this could be a nightmare (shudder) for Micro$oft shops. One more reason to not use M$ products.
Of course *NIX can be vulnerable as well to this type of trojan horse. The user security of *NIX may be better, but security is only as good as the user using it. The main difference, I believe, is that *NIX users are a lot more knowlegable about their systems and are much less likely to download and install software of questionable origins.
Unfortunately this won't work. In my case, if the upstream bandwidth exceeds 128Kbps the cable modem just starts dropping packets. What the modem should do is only drop packets which contain data and not ACK packets.
Also, I have been on @Home for 2 1/2 years. In the last year the service has become terrible. When I first joined, I received email about billing, was notified ahead of time about downtime, and was treated with respect. Tech support was excellent and the hold times were low. Reliability was high. Also, the AUP I signed said absolutely nothing about running servers and they were accepted.
Now it's the opposite. Mail and news is frequently down. 20% packet loss when trying to leave the @Home network is a frequent problem. Performance sucks. I no longer receive any email about billing (only spam from @Home). Hold times for tech support are frequently around 2 hours. All this and my rates have gone up.
I would be willing to pay extra to be able to run a server again, but @Home will not support @Work in residential areas, and if they did it would cost around $400/month! My cable modem is capable of 10Mbps full duplex. @Home should offer tiered service where one can pay more for higher quality of service.
My employer also has @Work. The packet loss is also very high at peak times.
The packet loss is not over the local links, but as soon as packets leave the @Home network. @Home does not have adequate peering with the rest of the Internet.
It could be enhanced. For example, it would be really cool to place a sink within a sink! That would vastly increase scalability and make it hierarchical more like a traditional file system.
For example, you expand the first sink until you see your personal document folder. You select that sink. From there you expand it until you see a folder for notes from your brother and you select it, and so on.
What's really cool about this is that it adds a sort of three dimensional method of accessing files rather than the simple tree and file list.
-Aaron
As a TCI@Home user I am quite pleased with the ruling. @Home has totally mismanaged things in my area. Packet losses of over 20% are not unusual. The mail server frequently crashes, and their tech support is totally unacceptable.
It takes over 3 weeks to get a response from a technical question sent via Email, and hold times on the phone are measured in hours.
What really makes me mad is the fact that @Home spends billions on companies like Excite, yet their internal infrastructure is totally inadequate. @Home is trying to become another AOL. I don't want AOL, I want a high speed Internet connection.
Not only that, but my rates have gone up while my bandwidth has been capped.
I have been an @Home user for 2 1/2 years, and during the last year the service has gone to hell. When I first joined, there were no hold times for tech support. The tech support people were sharp and knew what they were talking about. The performance was excellent. The uptime was excellent. They sent out email telling us ahead of time about downtime. They also emailed us our monthly billing info.
Today the only email I get from @Home is spam. I don't get a monthly bill from them (except on my credit card). Downtime is never preannounced, and tech support always responds by saying "Clear the cache in your browser" as if that will fix the 20% packet loss.
Oh, the packet loss is not caused by my connection to the cable company. The packet losses are due to the fact that @Home has inadequate bandwidth to the rest of the internet. The packets are always lost as soon as traffic leaves the 172.16.x.x routers.
Another nice feature of Token Ring is the large frame sizes that are supported. I don't know about 100 or 155Mbps Token Ring, but the old 16Mbps Token Ring supports packets up to around 18K. Larger packets greatly improve bulk data transfer. Of course, ATM can do the same thing, with AAL5 supporting PDUs up to 64K with RFC1577 or 18K with LANE.
On a properly configured server, NT should be stable enough. When I worked at Adaptec and when our drivers were fully debugged, the servers were fairly stable when benchmarking and stress testing. The testing required for servers WRT NT is fairly extensive compared to client computers. I know, for example, that Dell's server stress test is very extensive. Very few network card vendors passed the test.
On ordinary consumer hardware, yeah, NT isn't that stable. Also, remember that these benchmarks don't match the real-world. In the real world a lot of stuff occurs that doesn't occur during benchmarks or stress tests which can cause NT to BSOD. Also, for the benchmarking, NT will likely only have to stay up for a few hours.
NT might very well beat Linux. Linux still has some shortcomings compared to NT for the test Mindcraft is about to run.
1. NT supports zero copy transmits out of the disk cache or the web server cache. This means that the data is directly DMAed by the NIC card out of the disk cache or user-space memory buffer out to the wire without having to be copied to kernel space. Linux (as of 2.2.7) does not support this (this requires changes to the Linux virtual memory code and to the networking code to support fragmented packets and locked memory).
2. NT (with Intel cards) supports Fast EtherChannel if connected to a Cisco switch. Linux does not. Fast EtherChannel, for those who don't know, allows multiple ports (up to 4) to be ganged together to behave as one virtual fat pipe (800Mbps with 4 100Mbps adapters running full-duplex, the usual configuration).
3. Intel adapters on NT can offload the TCP checksum calculation to hardware. Linux can not. I don't know if SP4 added that support to NT4 or not. It is definitely in NT5 (I worked on some code to take advantage of it).
4. If NT uses NetBEUI it will beat Samba. NetBEUI is much faster than NetBIOS over IP.
When I worked at Adaptec and we were competing against Intel for server tuned NICs I saw what NT could do. With 6 network cards set up with load balancing we saw over 500Mbps of throughput, and the server was less powerful than Mindcraft's. Intel wasn't too far behind us either. This was with NT4+SP3 on a quad 400MHz server.
Now, personally, if I had a choice I would almost always choose Linux over NT, but I'm just trying to be realistic. One of these days I need to sit down and hack some Linux networking code. Fast EtherChannel support would be a very nice addition to Linux.
-Aaron
I periodically go through the same thing with TCI/@Home. It seams that about once a year they change their billing system so it bills me twice. Then it takes 4 months for them to straighten it out.
TCI/@Home is totally screwed up. Here in Fremont, CA. it got to the point where we (the users) have organized and appealed to the city council. They also are quite annoyed with @Home's terrible customer service (typical hold times are measured in hours, not minutes) and terrible performance (20% packet loss to @Home's NAP connections).
Things have improved recently, but for a while my 28.8 was a *MUCH* faster than my 10Mbps cable modem.
As to the city council I went to the meeting. It was really fun watching @Home squirm. TCI tried to claim that they fixed the customer service problem. The city council then had the city manager call TCI/@Home customer service right there during the meeting. They couldn't get through. TCI had fun trying to backtrack and wiggle out of that one. TCI/@Home offered to donate $40,000 to the city for any program the city chose if the city agreed not to fine them (the fine was $18,000 for lack of service). The city council was ready to accept until the customer support fiasco and went ahead and fined them.
It was a very entertaining city council meeting and I hope to go to the next one. The city gave @Home 120 days to fix their problems or they'd get a new fine. The city council is also going to pass legislation requiring minimum standards. This will be the first city in the country to do so. All this because of TCI/@Home.
If Redhat is going IPO and if Redhat is smart, they'll introduce some form of poison pill. For those who don't know, a poison pill is basically something which goes into effect if there's a hostile takeover. A poison pill could be just about anything, such as all stock options become fully vested immediately. Poison pills can work quite well. Even without one, M$ wouldn't dare to buy Redhat with the DOJ looming over them. The FTC would likely block it in an instant.
I guess you don't remember Agis and spammers then. Agis gave the spammers a free reign on its network. The spam flowed like the mississippi and everyone got pissed off at Agis and firewalled them off. Agis now has about the best anti-spam record in the industry.
I'm sorry, but how would you feel if you were an ISP and someone ran a huge spam run against your users and your mail server ground to a halt under the load, with millions of undeliverable messages and what not.
I have yet to have a Micro$oft mouse last more than 6 months on me. After 6 months (even with careful cleaning) they always start mistracking. My collegues keep wondering what that swearing and banging is when I start banging the broken mouse against the desk in frustration. I have yet to have an optical mouse die on me either. I use an old Mouse Systems 3-button optical mouse and have never had a problem (except that the pads on the bottom are wearing out).
-Aaron
Sorry about the double posting. Netscape 4.51 barfed for the first posting and I didn't think it made it.
It very well might be that NT did beat Linux. Linux still has a long ways to go tuning wise. Let's face it, most Linux programmers don't have access to a quad Pentium III server with an intelligent RAID controller and multiple NIC cards.
Linux still has a ways to go. As someone who has worked on a number of device drivers (some of which were NDIS) I must say that NT could very potentially be faster than Linux as a web server.
For example, NT supports zero-copy transmits for TCP traffic. Linux (as of 2.2.6) does not. NT 5 supports offloading the TCP checksum calculation to hardware, which many of the newer NIC cards support (i.e. Intel and 3COM), Linux doesn't support this. NT 5 even supports hardware TCP segmentation assistance, again, Linux does not.
Copying data is painful and should usually be avoided. Linux as it currently stands must copy all of the data sent out over the network due to it's current architecture of using one buffer per packet. Before someone tells me that it can, first go read the source code (tcp.c and eepro100.c). Linux also does not support TCP checksum acceleration or TCP segmentation offloading.
I must also say that BSD can also perform zero copy transmits.
Also, servers are whole different animals than desktop computers. Servers often have much faster PCI busses (64-bit and/or 66MHz), much faster disk I/O capability via intelligent RAID adapters, and use certified networking hardware that is held to a much higher performance and reliability standard than desktop PCs.
Also, I know for a fact that the Intel NT driver has been very carefully tuned. At my last job we went head to head against Intel in the server NIC business. Whenever we would surpass them in performance, they would turn around and surpass us. In the end we beat Intel, but only because our descriptor format was simpler (modeled loosely after the DEC Tulip architecture with tons of interrupt reducing features). We spent many months tuning our drivers on quad XEON servers to handle multiple adapters and load balancing (Fast EtherChannel and our own proprietary method). We were doing over 500Mbps on a quad 400MHz XEON server with NT 4+SP3.
The last time I communicated with Alan Cox it sounded like a lot of work was needed to add the support for proper memory locking (especially on SMP boxes) to provide the zero-copy scatter-gather transmit. Before you say "what about receive?", a typical server has a transmit/receive ratio of 10/1, so it is very critical.
Now I'm not saying that NT is faster than Linux, only that it potentially could be. Micro$oft has spent a lot of time and effort tuning the networking in NT 5 so Linux needs to watch out.
I'm also not saying that Mindcraft is right either. After seeing their reports on Solaris and Netware vs. NT I am very skeptical about them as a company.
-Aaron
--
Why is everything alway's Segmentation's fault?
It may be that Linux faired poorly in their test and that Mindcraft was unable to get decent tuning information. It is also possible that even if Linux were properly tuned it would still lag. Part of this comes from my experience with writing network device drivers (and I must admit I have worked on some NDIS drivers).
Instead of just crying foul, why not run some independent tests on some server hardware. Server hardware is a whole different class than desktop systems. The new systems have faster PCI busses (64 bit and/or 66MHz), faster disk I/O via intelligent RAID adapters, fast network adapters, and so forth. Hardware certification for servers is also much stricter. For example, as of the end of last year 3COM was not certified for Dell's server line because 3COM couldn't flip the bill for performance or reliability. Intel easily trounced 3COM in performance and reliability when we benchmarked.
At my last job we went head-to-head against Intel in the server NIC market. Every time we found new tricks in our drivers to up the performance, Intel would come along and do the same thing. Over our months spent tuning our drivers we more than tripled our performance and Intel made some significant progress as well. Now, the original driver architecture was very lean and clean (well, as lean and clean as possible in NDIS). We did numerous tricks to reduce slave PCI accesses and reduce (or even eliminate) interrupts to drop CPU utilization.
We also implemented load balancing and failover across multiple adapters via Fast EtherChannel and our own proprietary algorithm (which preceeded Fast EtherChannel).
In the end, we finally surpassed Intel, but I blame this only on the fact that the hardware interface was cleaner than Intels, basically modeled after the popular DEC Tulip architecture.
Now, for our tuning we spent many months of careful analysis with very heavy loads. We had 100 NT client computers pounding the server running many different tests, from many different sources. In the end we were able to achieve over 500Mbps throughput on a quad 400 MHz XEON server.
Now, I'm not knocking Linux. I think what needs to happen is for some serious performance analysis and tuning to take place. I am sure that there are numerous areas in Linux that can be tuned and enhanced to take advantage of these servers.
For example, NT 5 has support to offload TCP checksum calculations and even TCP segmentation to the hardware. Many of the new NIC cards support hardware checksum calculations, including 3COM and Intel. NT 5 also supports non-copy transmits, where data fragments are DMAed directly out of user-space buffers, thus eliminating the copy from user space to kernel space. Another feature NT has is the ability to DMA directly out of the disk cache.
Now, looking at the Linux 2.2.6 kernel (tcp.c and eepro100.c) it looks like it cannot support zero-copy transmits. It must copy, since skb's can only hold one fragment per packet. Copying data has a very noticable impact on performance in my experience, and should usually be avoided. I'm sure one of these days Alan Cox or someone will change this. The last time I communicated with Alan it sounded like it would require some major effort to eliminate this bottleneck since it would require some complex memory locking code.
Now, just comparing the architectures of NT and Linux, in this case the transmit path in NT could very conceivably be faster. NT 5 may very conceivably be much faster than Linux since it not only eliminates the copying stage, but can offload the checksum calculation to hardware.
Now I'm not going to say anything about reliability, since Linux almost always wins hands-down, at least according to the posts here, although I have rarely had any problems when I used NT (except for driver bugs).
Also, I'm not saying that NT is faster, either, since I have not performed any side-by-side benchmarking. I also must say that I'm very leery of Mindcraft especially over their reports of NT beating out Solaris as a web server and NT being a lot faster than Novell.
One final thing I might add is that if Linux puts a lot of effort into competing on the servers it very well might have a negative impact on the desktop. Often one must sacrifice simplicity and code size for performance. Supporting copyless transmits, for example, will require more complicated network drivers and more complex code dealing with buffer management, not to mention more complexity needed for the paging code (to handle memory locking) and the SMP code.
-Aaron
--
Why is everything always Segmentation's fault?
I totally agree. Likewise the embedded world can be almost as bad (in some cases worse depending on the environment).
I had a lot of fun working on an OS/2 ATM device driver that was around 100KLOC lines of code. It was a dream compared to the 360KLOC NT equivelent driver written in C.
Then came NT 5... Can you say spending a week just trying to get the driver to install after Microsoft released a new beta that broke the install script (which MS wrote to begin with) and WinDBG crashing EVERY FIVE MINUTES LIKE CLOCKWORK?!?!? Then there was the constant crashing MS kept complaining about which was traced down to the fact that MS wasn't updating the registry (which we couldn't duplicate since they wouldn't send us their latest build). I spent 3x as much time just fighting stupid MS issues instead of working on the code.
Dealing with user space is much easier.
After dealing with WinDBG you don't know how privileged you are with GDB.
This is what I'm planning. There's some new technologies such as reflective LCD's and the micro-mirror stuff from TI that should make these projectors much cheaper. Right now reflective LCDs look the most promising. BTW, these are totally unrelated to active-matrix or your typical LCD technology.
I saw Descent on a 10' projection screen. All I can say is it's a totally different experience.
Think about it, you could sit back in that easy-chair and code away without having a CRT in your face. That will reduce a lot of eye-strain.
I expect that within a couple of years the price of these new generation projectors will fall to under $2K and possibly to around $1K.
C, C++, Java, assembly, and CSH at work.
Last project was a 15KLOC Java app (around 90 classes).
The problem with a 19" or 13" HDTV is that it would be a total waste of money. For one thing, the CRT would not be able to take advantage of it
(an expensive 19" CRT might, but who sits that close to their TVs). For the smaller sets, DTV is more than adequate.
Yeah, a 32" HDTV set looks awesom compared to a 32" NTSC TV, but why bother with all the extra expense involved. In a year you'll be able to get a 33" HDTV set for a bit under 3K. The problem is that it is like scaling a 32" TV to a 13" TV. It just looks too small. It's the detail. With all that extra detail it's like looking at a movie from twice as far back as the back row of a theatre. You keep wanting to get closer. If I had an HDTV card I'd have my eyeballs about 6" from the monitor (and be dying of eye-strain) watching a movie. I know, I've been watching line-quadrupled TV & DVD on a 19" SGI monitor and I looked at a 33" HDTV set a start-up company is coming out with. Yeah, the picture looks incredible and very sharp, but I'd much rather have a good 32" TV any day. It's such a different experience to see HDTV on a big screen. HDTV on a 10' screen is breathtaking.
As for an HDTV card for the PC, how many monitors can handle 1920x1080? For most of us this exceeds the dot pitch by a fair amount. Even if it matched the dot pitch, I'd have to have my eyes 3" away to see the detail.
I guess if you just want to watch TV on your PC all day the card is fine. But for watching HDTV all the time on a PC monitor I just shudder.
The recommended minimum size display for HDTV is 60". I saw HDTV on a 33" display and it isn't worth it. Heah, you have a great picture, but you're missing the whole experience.
Also, as for the line doubler, there arn't very many line doublers out there that look good. Most of them introduce all kinds of artifacts from de-interlacing.
In the next couple of years, the price of HDTV projection displays should come down significantly. There's some new solid-state technology coming out, both TI's micro-mirror stuff (which some Japanese companies are copying) and reflective LCDs (very different than other LCD technology). Right now the reflective LCD technology sounds the most promising. As for the plasma wall screens, the color on them is too washed out and they put out too much heat (if you look carefully you'll find a bunch of fans in those displays).
I have a high-quality line-quadrupled AV system hooked up to a 19" CRT right now, running 1440x960 until I get my 33" monitor fixed. It is no comparison to a 33" TV for entertainment purposes. It's no fun watching a movie when you need to sit 12" away to see the detail.
It will be a couple of years until HDTV is ready for prime time. When it comes, nobody will want their old TV sets or converter boxes.
I pity those people walking out of Price Club with their brand-new 60" NTSC projection TVs.
Also, HDTV should be able to interact nicely with computers, since more than audio and video can be included in the signal. For example, you could click on an object and bring up more information or treat it like a URL.
-Aaron
While HDTV looks extremely clean, it's a total waste unless you have a 60" display, which is the recommended minimum size. HDTV projection displays will come down significantly in the next couple of years due to some new non-CRT technology coming down the pipe which are solid-state. The new technologies can easily reproduce the HDTV resolutions and beyond.
The other big advantage of HDTV (which a PC monitor will not take advantage of) is the 16:9 aspect ratio, which means that all HDTV broadcast and recorded movies will be wide-screen.
HDTV on a 19" CRT is not much comparison to even regular TV on a good 32" set from an entertainment perspective.
How can I claim this? I have a very good line-quadrupled DVD setup currently running 1440x960 through a 19" monitor (until I get my other monitor fixed). If I'm only a couple feet away it looks great, but that's no way to enjoy a movie. Of course, HDTV is a big improvement over the line-doubled/quadrupled NTSC. Also, for regular TV if the line doubling is not done properly there are all kinds of annoying artifacts that show up due to de-interlacing. Most de-interlacers I have seen have this problem to varying degrees, the exceptions being Snell and Wilcox ($30,000) and a Phillips unit (DVX8000 - $5K) and some units based on a chip from Genesis.
As for combining HDTV and a PC, they're a perfect match in many ways. The HDTV format allows for more than just plain audio and video to be transmitted. I.E. you could click on an object on the screen and get more information on it.
As for recording HDTV, the latest specs I've seen are 3 1/2 hours of HDTV on a VHS tape or 7 hours of DTV, and since it's digital there should be no quality degredation.
Now, instead of running HDTV on a PC monitor I'd love to be able to hook my PC up to an HDTV monitor and play Quake on a 60" screen at 1920x1080!
-Aaron
I know there's a big markup. On the high-end equipment a markup of 50% isn't that unusual, often higher. Part of what you are paying for is for someone to really design something right without marketing constantly trying to figure out how much cost they can cut.
I stand corrected on the price of DACs. The new Burr-Brown DACs have come down significantly in price and are now around $3 in 1000 quantity. The last time I looked over there they were on the order of $40.
As for building high-end equipment, which I have been involved with, it isn't cheap. A lot more care needs to be taken in terms of selecting matched components and using quality parts (i.e. no tantalum caps or carbon resistors in the signal path). None of the consumer-grade equipment I've seen does a decent job of using quality parts. Open up most Sony or other cheap equipment and you'll find it's full of 10% carbon resistors, tantalum caps, cheap electrolytics, and jumpers all over the place on a phenolic PC board with marginal solder connections and a just barely adequate power supply using LM78xx class regulators for the analog section (which are terrible in terms of noise).
Most consumers wouldn't notice much difference with a high-end DAC with their $200 pair of Bose or Sony speakers. Someone with a pair of NHT's, Magnapan, Martin Logan, or Wilson's will definitely hear a difference because the speakers will be able to much more accurately reproduce the sound without muddying it with their own color.
The high end equipment does not make these compromises. The BOM is higher and the quantities sold are much lower (since the average consumer doesn't have a system that could notice it with their $200 pair of Bose speakers). Due to the quanty sold, the manufacturer must charge more to stay in business. If you're Sony or Panasonic then you plan on hundreds of thousands if not millions of units selling. If you're Theta, Pass Labs, Audio Research, etc. then you figure you'll sell less than 10,000 units. If you were to sell 10,000 units at $100 profit each and you have a number of employees you'd go out of business real fast.
This isn't to say that Sony or Magnavox couldn't make a high-end player. They could, but they're constantly looking at how they can cut costs which defeats the purpose.
I can very well see why a good DVD audio player would cost $1000. Have you tried pricing what it takes to build a good high-quality DAC that can handle 20 bits, let alone 24 bits? More than that, you need more expensive op-amps, better quality resistors, capacitors, filters, and so on. A high-quality 24-bit DAC isn't cheap either. The noise floor becomes much more critical and hence a much more expensive design is in order.
For good quality audio equipment you don't get away with using cheap electrolytic capacitors or carbon resistors any more or the really cheap op-amps and voltage regulators. Now you need to move to 1% or better metal film resistors (or better) and polypropolyne/polystyrene caps and either discrete or very high quality op-amps. The power supply needs to be beefier and have better noise immunity. More than likely, a separate power supply is needed for the digital vs analog portions of the circuit.
Oh yeah, now that you're 24-bit you want to use balanced XLR connectors. For that nice clean balanced output you now need two DACs per channel running inverted for a total of 4 DACs.
A starting 24-bit DAC probably starts at $20 each. You're at $80 already for only one piece. Now another $60-80 for op-amps, $30 for resistors and caps. Now throw in the power supply for $80 for a good toroid transformer and double regulated supplies, independent for each channel and for the analog and digital portions. You're up to $250 already and nothing has been covered for the transport or other circuitry. Add video support and AC3 5.1 channel audio and you could easily exceed $1000 for parts alone.
Of course, most of those here couldn't tell the difference between a CD and a cassette, given how fond you are of MP3 format. MP3 has some serious shortcomings quality-wise, and some could be fixed without too much difficulty (i.e. using a logrithmic scale for dividing frequencies rather than a linear scale).
For those of us who really like to try and reproduce music (rather than noise) we do notice the shortcomings of 44100/16. Hell, even consumer equipment typically tries to enhance it (i.e. 8x oversampling) since trying to make a 20khz brick wall filter is about impossible without screwing up the phase and other artifacts.
The filtering involved in the recording side also effects the sound since all input sound must be filtered to reject any signal above or at 22050 to eliminate aliasing.
DVD audio is a long-welcomed addition by hi-fi entheusist since CD's are so screwed up. The recording industry took the minimal acceptable standards and went with those much the same as they pushed cassettes, another bare-minimum standard, or VHS, which was not properly engineered from the beginning (unlike betamax, which used a much better head geometry).
Of course, the DVD audio spec supports several sampling rates besides 96khz and several bit depths from 16-bits to 24 bits. This means those 10-CD collections on those late-night commercials could be replaced by a single DVD.
Also, if they support AC3 and DTS audio DVDs all the better. Full-surround DTS audio would be a nice medium to play with.
CDs are ancient technology, and anyone who wants to replace them with MP3 should go back to their 3" computer speakers and sound blasters.
You don't need to go to an expensive college to get a good degree. I've met many sharp people who graduated from San Jose State. I myself graduated from U.C. Santa Cruz in Computer Engineering. The UC system isn't all that unreasonable either when compared to many of these private Ivy League colleges. The thing is, you need to find a college that has a good cirriculum and offers a wide range of choices. Also, at least at UCSC, I was required to take numerous courses outside my CE major (i.e. music, writing, Latin American Studies, and so forth).
You don't have to graduate from Harvard or MIT to get a good education. Of course that doesn't mean that you'll get a good education down at the local Junior College. The JC's typically arn't in the same calibar as the 4-year colleges.
One other thing college does is to require you to work on things you probably wouldn't choose to do on your own. Yeah, you might think that that class on microprocessor design is boring, but at least you were exposed to it so you would know what it was about. I found some of my most enjoyable classes were subjects I would never have studied on my own.
And finally, by getting that piece of paper, you have demonstrated that you have some skill and that you can provide the effort needed to succeed in a new job.
In my experience, most companies want one of two things before hiring, either prior work experience or a degree. It's almost impossible to get your foot in the door beyond a sys-admin (if even that) without that piece of paper.
I was lucky in that fresh out of high school I was writing assembly code, but that was only through a fluke set of circumstances. Most of my friends had no such breaks.
As for those of you who say "I can earn $150,000/year without college", yeah, you can, but not straight out of high school. It's going to be a much more difficult climb to reach the levels of those with a degree. With that degree you start much further up the ladder, and your job prospects are much better at least for the first few years. After your first few years, experience becomes the dominant factor.
College did several things for me. I was a geek through and through prior to entering college, but college opened up whole new avenues for me that there is no way I could have explored otherwise. Things like building a microprocessor would have been out of the question, or playing around with GL on an SGI (prior to Open GL).
Many of the things I learned in college were invaluable, besides just that piece of paper. Other things were not very valuable. There were a number of classes that were basically a waste of time, but that was just preparing me for the real world.
Perhapse it depends on the college. I just interviewed a new college grad for an entry-level embedded programming job, yet the grad couldn't perform simple things. I asked about the difference between a linked list and a binary tree and how they relate to Big-O notation when searching. No answer. I asked the grad to write a C function to convert an integer to an ASCII string. Again, the grad was at a total loss.
For those who say they got nothing out of college, either you didn't want anything out of college or you were some super genious before entering. Either that or you went to some brain-damaged college.
Prior to college I had done a fair amount of programming and exploration. I knew 80x86 assembly cold and all the main data types used. In college I was able to greatly build on my experiences. Also, that piece of paper has been useful since it allows me to get a lower insurance rate.