Re:Why bother?
on
Palmtop NetBSD
·
· Score: 2, Interesting
I've got an iPaq that runs Familiar Linux, and it's a great mobile hacking and experimentation platform. True, I wouldn't want to write full apps with the tiny onscreen keyboard, but it's nice to be able to either try out an algorithm or quick script idea in Python on a full-featured Linux system that fits in a coat pocket. If I had the extra couple of hundred dollars lying around, I'd definitely make the switch to one of the WinCE devices with even a tiny keyboard, and probably be able to use for for most of the tasks I'd normally assign to either my PDA or laptop.
Personally, at least, I've found that at least 80% of my time on either kind of portable device map fine into basic console tools: mutt for email, links or w3m for quick web browsing, and remote server administration via ssh, and a bit of Ruby scripting here are there. Not too much there that requires a GUI, (even if I liked the WinCE UI, which I don't) or support for the huge library of mediocre shareware and Windows-specific integration and synchronization that seem to dominate the PalmOS market.
Absolutely. It's not different than, say, getting your managers to let the developers switch over to using Linux for their dev workstations, or MySQL or PostgreSQL instead of Access or MS SQL Server -- it's a cost/benefit comparison, which you have to *figure out* the correct answer to. Developing software in-house is almost always at least as expensive as buying it; however, if the off-the-shelf solution doesn't have features or options you really do need, then you're going to have to get the vendor ($$), a consultant ($$$), or someone in-house ($-$$, plus a lot of waiting) to do the customization. Plus, once you've customized an outside solution, it may or may not still be eligible for normal support and upgrades from the original vendor.
I've found that having a short (1-3 page) analysis, which lists the functional requirements for the system first, and then breaks down where the front-running commercial packages do and don't meet those needs, along with an estimate of the cost and time needed to adapt them to work (if it's even possible, of course; sometimes, you don't get access to source code, or even an extension API). Along with that information, you need at least a ballpark estimate of the development time and costs associated with doing it in-house, which should actually be fairly accurate. If possible, get a non-programmer who has been involved in development projects before to look over the estimates for your solution, and call BS on you where they can.
It's definitely a battle you can win, but first you need to have more than your hacker's disdain for other people's code to prove that the in-house way is best.
The major labels are already getting $0.65 from every track sold through the Apple music channel. I would expect that most of the independent labels would club baby seals for that kind of profit margin.
Bands don't get salaries from the big labels -- just as the parent post stated, the usual arrangement is for the label to advance the band a certain amount on signing, and then deduct royalties that would be owed them from that initial amount. In order to start collecting any money after that first payment, a band usually has to go gold within their first two albums released on the contract.
Even worse, most of the promotional and support services you talk about are actaully paid for by the band, again out of their future royalties. Many groups actually end up *owing* the label money after their first album and tour, which only further binds them to the label's "artistic direction" for them.
The only thing you get from a major label by signing with them is a temporary invitation to ride their distribution network to promote your work. Artists pay their own way, and lose all rights to their work in the process. Of course, that's often better than the alternative, since independent groups and labels have effectively no leverage with radio stations and major venue owners, esp. given the current trend of media consolidation
You just have to look at the likely outcome from this week's news about the new FCC relaxation of ownership rules for a chilling view of where we're going. I don't think that Apple is going to turn things around overnight, but any arrangement which gives the indies more opportunity to reach a paying audience is fine by me.
I agree on the crypto issues, but forgive the Nullsoft folks for one simple reason: anything that pushes public key encryption closer to the mainstream is golden in my book. I don't think that the current implementation will survive much more than a couple of weeks, but the idea of a PK-based secure messaging and file transfer mesh should live on and improve.
Transitioning AIM over to the WASTE protocol seems highly unlikely, since WASTE starts to break down with less than 100 users connected, due to a number of technical limitations and design decisions. In a (legal, since it was before the recent notification that the distribution wasn't legit) test yesterday, I connected to a mesh that varied from 60-120 users throughout the day. When the higher numbers of users were on, my broadband connection (~1mbps) was nearly saturated with broadcast packets, (chat, search, and "heartbeat" pings) which left very little bandwidth for any useful work.
It also doesn't handle presence very well: users are only detected leaving the mesh by a timeout on their reply packets. That means that you end up with a number of "ghost" users who may have either disconnected or just gone inactive, who the system can't really determine the status of.
I think you've hit it on the head there. AOL/Nullsoft may or may not have a problem with their employees working on open source projects, (i.e., Mozilla) but they have to be acutely aware of any potential licensing no-nos, esp. given the current SCO/Linux debacle.
Personally, I think it's an interesting project, but needs some serious work before it could be a viable alternative to existing chat and filesharing apps -- the design docs distributed show a number of issues with the wire protocol, including its overuse of broadcast messages, and the high (i.e., 40 bytes per packet!) overhead added for message checksumming, routing info, etc.
Re:Downfall of muds....
on
Saving MUDs?
·
· Score: 3, Insightful
The hosting issue is definitely one problem with MUDs, MOOs, etc. Starting and running a web-based community can be extremely cheap these days, (at least with the kind of userbase that most MU* servers deal with) given the prevalence of $10 web hosting packages with PHP, CGI, MySQL, etc. However, to run a MUD or similar, you need a server with full shell access, support for persistent processes, and at least some control over the firewall rules, to allow client connections to the odd port numbers that the servers tend to use.
Of course, there are technological solutions to this problem: use a Java applet that directly connects to an XML-RPC/SOAP CGI server, or to the database backend; build a DHTML-based UI for a traditional MUD-style environment; or even just pool resources with other MUD admins to get one of the cheap "virtual server" accounts somewhere like JohnCompanies, and share it amongst several running game server instances.
Personally, I'm very interested in the second option, and have been hacking together pieces of a MUD-style backend that could use a browser-based interface to do real-time chat, building and development, etc. Looking at systems like the Zope Management Interface, and other OSS content-management and thru-the-web development tools, I think it should be entirely possible to migrate the MUD experience to a web interface, which would simplify both hosting and use by a wider range of players.
...and take some of those gigs that barely cover gas and food, just to get your music out there for people to listen to. Most of the indie "success stories" (Big Head Todd & the Monsters, Ani DiFranco, etc.) based their reputations on their live performances, with album sales being just another means of helping to support that primary occupation.
You can't just throw your music at people who've never heard or seen the band before, and expect them to gladly fork over $10-20 on the chance that it could be good. Until you have some "known" artists, (i.e., they can attract a crowd on the basis of their name and rep for their shows) the label itself isn't going to be a good promotional vehicle. Once one or two of your artists have started to attract some attention, though, the label's name can be an attractor for new talent, and for listeners looking for more of that kind of music.
It can be done -- my father has been making a living as an independent musician for a number of years, and after establishing a sizeable local following for his live shows, has managed to self-publish and sell out several 1000-unit batches of his recent CDs. However, it took at least 3-4 years of low-paid live shows, interviews and solo accoustic sets on local radio stations, etc., before he was able to do so.
I actually thought about the IEEE1394 cluster idea for a while, after I bought a $20, 3-port card for my desktop machine. Shortly after getting it home and installed, I had a thought of building a star-topology cluster, with each node using its three ports to connect to three others, shortening the distance between them vs. a daisy-chained setup.
Unfortunately, the CPU utilization figures I've seen for the Linux kernel support for IP-over-IEEE1394 suggests that you couldn't really get too close to maximum bandwidth without pegging the CPU and PCI bus, which mostly defeats the purpose of using it instead of basic 100Mbit Ethernet.
I spend a fair amount of time working on a ThinkPad 560 (Pentium-120, 40MB RAM, 800 MB HD, 800x600x16bit LCD), and I have to say that most of the "standard" desktop environments (GNOME, KDE) choke a system like that.
Windowmaker is indeed a good choice for a basic window manager and launcher/task manager, but it doesn't provide a lot of the file management and drag-and-drop support that even kids expect from their computers these days.
I'd recommend ROX Filer (rox.sf.net), which does a good job of basic file management, image preview, etc.; XFCE might be an option, as well, esp. once they release their current development version.
I have an OpenBrick (http://openbrick.org), which works very nicely as a thin client box connecting to faster/noisier machines located elsewhere in the house; obviously, laptops work well for this, too, and since the bulk of the processing is offloaded to the server, even a fairly old machine works well, so long as it can run X and/or a VNC client.
With wireless cards thrown into the mix, you can just stuff the bigger machines into the attic, a closet, etc., and then set up your thin client workstations wherever you like, without fear of noise and space consumption making them a nuisance.
All of this breaks down as soon as you want to run a recent game of any sort, of course, but that's what the console systems in the living room are for.
...find a decent pawn shop in the area, and check out the stones and settings they have. If you've done some looking at more "reputable" jewelry stores, and learn how to look at a stone with a loupe to check for visible flaws, color, etc., you can get a very nice piece of "ice" for about 1/3 the price, and they'll usually throw in a simple setting in white or yellow gold for free.
My parents did it, (no engagement ring, just wedding bands) I've done it already with the engagement ring, and I think I'll probably be able to get her to go for it on the wedding bands, as well.
Going the custom-design route can be very cool, but unless you know a jewelry designer and/or metalsmith, it's probably just a good way to end up with some very nice mental images or sketches of the ring of your dreams.
I know it is, and always will be, fashionable to bash Microsoft around here, but I really wish folks would stop talking about their developers as though they were drooling morons. I've known people who worked for Microsoft, and met a number of extremely bright developers, researchers, and even (gasp!) non-technical staff from their various branches, and most of them were capable, bright people.
Microsoft's screening and interview processes for programming positions are famously tough, and they attract many of the best and brightest straight out of school. Like any techie working for a large corporation, of course, they can't spend all day posting to web discussion boards about all the amazing PHP code they've been hacking together, but that doesn't mean that they do bad work.
Just becuase Microsoft's overall presence in the technology and business worlds is agressive, manipulative, and derivative of others' innovations doesn't mean that the folks "in the trenches" aren't at least as bright (and likely much more capable) as the usual/.'er.
And yet, there's good old-fashioned PostgreSQL, the original Timex of open source database servers...it'll take a lickin', and keep on tickin'!
Seriously, though, the high-load performance of pgsql still impresses me; for any application where I don't need the replication or hot-backup features of Oracle, I'd hold up a sturdy Postgres-backed (and mod_{perl,python,ruby}-powered) website against the dreaded/. effect any day.
When clients ask me why they should use PostgreSQL instead of MySQL, I tell them it's the "baby Oracle" to match MySQL's "baby MS SQL Server".
Not to incite any sort of Linux-vs-BSD war or anything, but have you looked at NetBSD? The 1.6 distro runs on 386s and up, (probably 286s too, if it has a math coprocessor) installs from two floppies with drivers for just about everything, and runs just as well on PCs, Amigas, old 68k Macs, etc.
Without the development tools, headers, etc., you can easily fit it into a 200MB image size, and go down from there by manaully removing items from the default installation sets. Plus, there are binary packages for most of the common apps you would want to use, which you can install with a simple command like: setenv PKG_PATH "ftp://ftp.netbsd.org/pub/NetBSD/packages/1.6/i386/All"; pkg_add abiword.
I have been a big Linux user (and booster) for many years, but if you want a lean UNIX-like system that runs great on old boxen, without going back five years on the versions of all your software (which is a big security win all by itself), I highly recommend NetBSD.
The in-memory datastructures for the vector search aren't a big deal for the scale I'm working with -- I'm basically able to just use hashes for all the vectors at indexing time, which lets them grow with the collection in O(n) time. What I'm more concerned with is an efficient on-disk structure for repeated queries of the same index, so that I don't have to re-read the entire index into memory every time I run a simple lookup.
As far as spoofing is concerned, though, I'm afraid that LSI and other "semantic" search methods will only outpace spammed results for a while, since the same basic algorithms that let you determine the relevance of a document could quite conceivably be used to generate the "ideal" match for any search you can think of.
For the porter stemming algorithm, check out the following:
http://www.tartarus.org/~martin/PorterStemmer/ There are implementations for just about any language I could imagine someone using for serious web development.
Re:This is probably the problem.
on
RAMdisk RAID?
·
· Score: 1
Notice the original restriction in the posting, though: on a limited budget. SCSI RAID really isn't the first solution I'd offer for that problem.
One thing I didn't think of in my previous post, though, was the XServe RAID units that Apple released recently...$6k gets you a 720GB Fibre-Channel RAID array in a 3U enclosure. Not bad at all, really.
Not really cost-effective
on
RAMdisk RAID?
·
· Score: 2, Informative
Assuming that you'd have to buy at least some of the Gb networking hardware (switches, cables, etc.), you're really not going to be saving much. Assuming at least $100 per 'RAMdisk server', you'll be spending $1200+ for a ~20GB RAID array that will lose everything the minute the power blinks, not to mention drawing several kilowatts of AC.
On the other hand, if you just throw four 100GB ATA-100 drives in a standard tower case with a decent IDE RAID controller, you get five times as much storage for probably about half the money.
Also, remember that most low-to-mid-range PCs can't actually fully take advantage of a gigabit network link, since the PCI bus and CPU get saturated long before the network does.
One potential extension not raised in the article (though it was perhaps hinted at in the discussion of attributes attached to each document) is the use of explicit hypertext link information to affect the vector value of each document in a collection. One could even 'stem' the target URLs to the top-level domain, affecting the same kind of term-space reduction as the Porter algorithm does for English text.
Now that I think of it, to some extent, both the Google PageRank methodology and some of the emerging weblog and news-tracking tools utilize hyperlink information very effectively to build a sort of loose semantic grouping between pieces of web content. However, from what I understand of the PageRank algorithm, at least, the conceptual model is much closer to basic flow graphs than vector spaces, and it (arguably unfairly) more or less completely ignores 'island' sites or pages, no matter how closely they match a particular query.
I'm actually working on a simple search engine implementation for a couple of small websites, and may have to tinker a bit with some of this.
One thing I'm not sure about, though: are there any standard, efficient *persistent* data stores/structures especially suitable for this kind of vector data? I'm hoping to work in Ruby, which unfortunately lacks a complete mathematical data library like PDL, but can easily hook in either a simple C library or a native implementation of any appropriate structures.
Please, don't start out making a quite viable point (namely, that "shrink-wrap" or "pre-installed" Linux systems are far more user-friendly) and the screw it up by saying that desktop Linux is ready to go head-to-head with OS X or Windows XP. My two primary computers run Linux and NetBSD, and for my purposes (90% of my time is spent in vi, bash, and perl/python/ruby scripting) they're great.
However, there's no way that I would seriously try to give a Linux system to my girlfriend, or even my fairly tech-savvy younger brother, simply because I know that they'd spend a lot of time trying to figure out how to do things that were idiotically simple on their Macs and damn-near impossible under Linux.
It's getting better, especially in the area of basic UI consistency for desktop apps -- and as much as I can't stand running a RedHat system, they're leading the way with the BlueCurve unification of GNOME and KDE. However, it'll probably be at least a couple more years before the Linux desktop situation improves enough to make a switch a less-than-excruciatingly-painful experience for the large majority of PC/Mac users.
Of course, by that time, the OpenBeOS crew will either have given up, or pulled off a working system, in which case I hope to be able to forget about XWindows, custom-built kernels, and the thousand twisty little/etc files, all *not* alike.
While I agree that the above post seemed to be in a "toungue-in-cheek" kind of voice, it also made a valid point: for any average business user, even one who is comfortable with doing their own Windows installations/upgrades, the free *NIX world (and the BSDs in particular, w/o RedHat or similar to put out the slick, documented installers) is a challenge to get into.
Personally, I'm very happy doing day-to-day coding and net work on an old Thinkpad 560 running NetBSD, and will probably try FreeBSD 5.0 if the release installer supports wireless PCMCIA cards (NetBSD 1.6 did, much to my surprise!). However, the value of the BSDs is much like the value of Debian, IMHO: they offer lots of low-level control, so you *can* do crazy things (and install them in crazy places) that the maintainers couldn't have imagined.
The most interesting works of modern fiction I've read were seldom examples of the finest writing style -- that's what the "classics" are for. Their value is more in the concepts and characters they introduce than in their use of language.
I think that Doctorow's short stories do a good job of presenting a scenario based on a popular current meme, while the idea is still fresh.
Slashdot Mirror
I've got an iPaq that runs Familiar Linux, and it's a great mobile hacking and experimentation platform. True, I wouldn't want to write full apps with the tiny onscreen keyboard, but it's nice to be able to either try out an algorithm or quick script idea in Python on a full-featured Linux system that fits in a coat pocket. If I had the extra couple of hundred dollars lying around, I'd definitely make the switch to one of the WinCE devices with even a tiny keyboard, and probably be able to use for for most of the tasks I'd normally assign to either my PDA or laptop.
Personally, at least, I've found that at least 80% of my time on either kind of portable device map fine into basic console tools: mutt for email, links or w3m for quick web browsing, and remote server administration via ssh, and a bit of Ruby scripting here are there. Not too much there that requires a GUI, (even if I liked the WinCE UI, which I don't) or support for the huge library of mediocre shareware and Windows-specific integration and synchronization that seem to dominate the PalmOS market.
Absolutely. It's not different than, say, getting your managers to let the developers switch over to using Linux for their dev workstations, or MySQL or PostgreSQL instead of Access or MS SQL Server -- it's a cost/benefit comparison, which you have to *figure out* the correct answer to. Developing software in-house is almost always at least as expensive as buying it; however, if the off-the-shelf solution doesn't have features or options you really do need, then you're going to have to get the vendor ($$), a consultant ($$$), or someone in-house ($-$$, plus a lot of waiting) to do the customization. Plus, once you've customized an outside solution, it may or may not still be eligible for normal support and upgrades from the original vendor.
I've found that having a short (1-3 page) analysis, which lists the functional requirements for the system first, and then breaks down where the front-running commercial packages do and don't meet those needs, along with an estimate of the cost and time needed to adapt them to work (if it's even possible, of course; sometimes, you don't get access to source code, or even an extension API). Along with that information, you need at least a ballpark estimate of the development time and costs associated with doing it in-house, which should actually be fairly accurate. If possible, get a non-programmer who has been involved in development projects before to look over the estimates for your solution, and call BS on you where they can.
It's definitely a battle you can win, but first you need to have more than your hacker's disdain for other people's code to prove that the in-house way is best.
The major labels are already getting $0.65 from every track sold through the Apple music channel. I would expect that most of the independent labels would club baby seals for that kind of profit margin.
Bands don't get salaries from the big labels -- just as the parent post stated, the usual arrangement is for the label to advance the band a certain amount on signing, and then deduct royalties that would be owed them from that initial amount. In order to start collecting any money after that first payment, a band usually has to go gold within their first two albums released on the contract.
Even worse, most of the promotional and support services you talk about are actaully paid for by the band, again out of their future royalties. Many groups actually end up *owing* the label money after their first album and tour, which only further binds them to the label's "artistic direction" for them.
The only thing you get from a major label by signing with them is a temporary invitation to ride their distribution network to promote your work. Artists pay their own way, and lose all rights to their work in the process. Of course, that's often better than the alternative, since independent groups and labels have effectively no leverage with radio stations and major venue owners, esp. given the current trend of media consolidation
You just have to look at the likely outcome from this week's news about the new FCC relaxation of ownership rules for a chilling view of where we're going. I don't think that Apple is going to turn things around overnight, but any arrangement which gives the indies more opportunity to reach a paying audience is fine by me.
I agree on the crypto issues, but forgive the Nullsoft folks for one simple reason: anything that pushes public key encryption closer to the mainstream is golden in my book. I don't think that the current implementation will survive much more than a couple of weeks, but the idea of a PK-based secure messaging and file transfer mesh should live on and improve.
Transitioning AIM over to the WASTE protocol seems highly unlikely, since WASTE starts to break down with less than 100 users connected, due to a number of technical limitations and design decisions. In a (legal, since it was before the recent notification that the distribution wasn't legit) test yesterday, I connected to a mesh that varied from 60-120 users throughout the day. When the higher numbers of users were on, my broadband connection (~1mbps) was nearly saturated with broadcast packets, (chat, search, and "heartbeat" pings) which left very little bandwidth for any useful work.
It also doesn't handle presence very well: users are only detected leaving the mesh by a timeout on their reply packets. That means that you end up with a number of "ghost" users who may have either disconnected or just gone inactive, who the system can't really determine the status of.
I think you've hit it on the head there. AOL/Nullsoft may or may not have a problem with their employees working on open source projects, (i.e., Mozilla) but they have to be acutely aware of any potential licensing no-nos, esp. given the current SCO/Linux debacle.
Personally, I think it's an interesting project, but needs some serious work before it could be a viable alternative to existing chat and filesharing apps -- the design docs distributed show a number of issues with the wire protocol, including its overuse of broadcast messages, and the high (i.e., 40 bytes per packet!) overhead added for message checksumming, routing info, etc.
WASTE_PUBLIC_KEY 20 1536 baka_boyB 45B34CD34A 7860C9048E5 9D5623041E23 99D7AF393FC EC213736FDD0 ECC7BB132E1 0032ADE97819 93A9D64E1A3 CFBEBCED1440 86C8286CE65 C643C6541F34 629CCD3D9AA STE_PUBLIC_KEY_END
CFCCBFDE4C297EB05F705721BC606A1E4FAA408
D1665C4791CDEEB3B676DDEE8109E4BF0E2FA
29F3BD73B0AE3FC32B4713C5D1837F8003B2F
5450557B37730D9F8F9B853617E043891C314
2A4CAFF29CD69BEE9E05373C96ED8B37C0E87
2A5251A348A6027C6675159C029D2BEEAA274
6D952FCB8A62291FDE3CD9870003010001
W
The hosting issue is definitely one problem with MUDs, MOOs, etc. Starting and running a web-based community can be extremely cheap these days, (at least with the kind of userbase that most MU* servers deal with) given the prevalence of $10 web hosting packages with PHP, CGI, MySQL, etc. However, to run a MUD or similar, you need a server with full shell access, support for persistent processes, and at least some control over the firewall rules, to allow client connections to the odd port numbers that the servers tend to use.
Of course, there are technological solutions to this problem: use a Java applet that directly connects to an XML-RPC/SOAP CGI server, or to the database backend; build a DHTML-based UI for a traditional MUD-style environment; or even just pool resources with other MUD admins to get one of the cheap "virtual server" accounts somewhere like JohnCompanies, and share it amongst several running game server instances.
Personally, I'm very interested in the second option, and have been hacking together pieces of a MUD-style backend that could use a browser-based interface to do real-time chat, building and development, etc. Looking at systems like the Zope Management Interface, and other OSS content-management and thru-the-web development tools, I think it should be entirely possible to migrate the MUD experience to a web interface, which would simplify both hosting and use by a wider range of players.
...and take some of those gigs that barely cover gas and food, just to get your music out there for people to listen to. Most of the indie "success stories" (Big Head Todd & the Monsters, Ani DiFranco, etc.) based their reputations on their live performances, with album sales being just another means of helping to support that primary occupation.
You can't just throw your music at people who've never heard or seen the band before, and expect them to gladly fork over $10-20 on the chance that it could be good. Until you have some "known" artists, (i.e., they can attract a crowd on the basis of their name and rep for their shows) the label itself isn't going to be a good promotional vehicle. Once one or two of your artists have started to attract some attention, though, the label's name can be an attractor for new talent, and for listeners looking for more of that kind of music.
It can be done -- my father has been making a living as an independent musician for a number of years, and after establishing a sizeable local following for his live shows, has managed to self-publish and sell out several 1000-unit batches of his recent CDs. However, it took at least 3-4 years of low-paid live shows, interviews and solo accoustic sets on local radio stations, etc., before he was able to do so.
I actually thought about the IEEE1394 cluster idea for a while, after I bought a $20, 3-port card for my desktop machine. Shortly after getting it home and installed, I had a thought of building a star-topology cluster, with each node using its three ports to connect to three others, shortening the distance between them vs. a daisy-chained setup.
Unfortunately, the CPU utilization figures I've seen for the Linux kernel support for IP-over-IEEE1394 suggests that you couldn't really get too close to maximum bandwidth without pegging the CPU and PCI bus, which mostly defeats the purpose of using it instead of basic 100Mbit Ethernet.
I spend a fair amount of time working on a ThinkPad 560 (Pentium-120, 40MB RAM, 800 MB HD, 800x600x16bit LCD), and I have to say that most of the "standard" desktop environments (GNOME, KDE) choke a system like that.
Windowmaker is indeed a good choice for a basic window manager and launcher/task manager, but it doesn't provide a lot of the file management and drag-and-drop support that even kids expect from their computers these days.
I'd recommend ROX Filer (rox.sf.net), which does a good job of basic file management, image preview, etc.; XFCE might be an option, as well, esp. once they release their current development version.
I have an OpenBrick (http://openbrick.org), which works very nicely as a thin client box connecting to faster/noisier machines located elsewhere in the house; obviously, laptops work well for this, too, and since the bulk of the processing is offloaded to the server, even a fairly old machine works well, so long as it can run X and/or a VNC client.
With wireless cards thrown into the mix, you can just stuff the bigger machines into the attic, a closet, etc., and then set up your thin client workstations wherever you like, without fear of noise and space consumption making them a nuisance.
All of this breaks down as soon as you want to run a recent game of any sort, of course, but that's what the console systems in the living room are for.
...find a decent pawn shop in the area, and check out the stones and settings they have. If you've done some looking at more "reputable" jewelry stores, and learn how to look at a stone with a loupe to check for visible flaws, color, etc., you can get a very nice piece of "ice" for about 1/3 the price, and they'll usually throw in a simple setting in white or yellow gold for free.
My parents did it, (no engagement ring, just wedding bands) I've done it already with the engagement ring, and I think I'll probably be able to get her to go for it on the wedding bands, as well.
Going the custom-design route can be very cool, but unless you know a jewelry designer and/or metalsmith, it's probably just a good way to end up with some very nice mental images or sketches of the ring of your dreams.
I know it is, and always will be, fashionable to bash Microsoft around here, but I really wish folks would stop talking about their developers as though they were drooling morons. I've known people who worked for Microsoft, and met a number of extremely bright developers, researchers, and even (gasp!) non-technical staff from their various branches, and most of them were capable, bright people.
/.'er.
Microsoft's screening and interview processes for programming positions are famously tough, and they attract many of the best and brightest straight out of school. Like any techie working for a large corporation, of course, they can't spend all day posting to web discussion boards about all the amazing PHP code they've been hacking together, but that doesn't mean that they do bad work.
Just becuase Microsoft's overall presence in the technology and business worlds is agressive, manipulative, and derivative of others' innovations doesn't mean that the folks "in the trenches" aren't at least as bright (and likely much more capable) as the usual
And yet, there's good old-fashioned PostgreSQL, the original Timex of open source database servers...it'll take a lickin', and keep on tickin'!
/. effect any day.
Seriously, though, the high-load performance of pgsql still impresses me; for any application where I don't need the replication or hot-backup features of Oracle, I'd hold up a sturdy Postgres-backed (and mod_{perl,python,ruby}-powered) website against the dreaded
When clients ask me why they should use PostgreSQL instead of MySQL, I tell them it's the "baby Oracle" to match MySQL's "baby MS SQL Server".
Not to incite any sort of Linux-vs-BSD war or anything, but have you looked at NetBSD? The 1.6 distro runs on 386s and up, (probably 286s too, if it has a math coprocessor) installs from two floppies with drivers for just about everything, and runs just as well on PCs, Amigas, old 68k Macs, etc.
6 /All"; pkg_add abiword.
Without the development tools, headers, etc., you can easily fit it into a 200MB image size, and go down from there by manaully removing items from the default installation sets. Plus, there are binary packages for most of the common apps you would want to use, which you can install with a simple command like: setenv PKG_PATH "ftp://ftp.netbsd.org/pub/NetBSD/packages/1.6/i38
I have been a big Linux user (and booster) for many years, but if you want a lean UNIX-like system that runs great on old boxen, without going back five years on the versions of all your software (which is a big security win all by itself), I highly recommend NetBSD.
The in-memory datastructures for the vector search aren't a big deal for the scale I'm working with -- I'm basically able to just use hashes for all the vectors at indexing time, which lets them grow with the collection in O(n) time. What I'm more concerned with is an efficient on-disk structure for repeated queries of the same index, so that I don't have to re-read the entire index into memory every time I run a simple lookup.
As far as spoofing is concerned, though, I'm afraid that LSI and other "semantic" search methods will only outpace spammed results for a while, since the same basic algorithms that let you determine the relevance of a document could quite conceivably be used to generate the "ideal" match for any search you can think of.
For the porter stemming algorithm, check out the following:
http://www.tartarus.org/~martin/PorterStemmer/
There are implementations for just about any language I could imagine someone using for serious web development.
Notice the original restriction in the posting, though: on a limited budget. SCSI RAID really isn't the first solution I'd offer for that problem.
One thing I didn't think of in my previous post, though, was the XServe RAID units that Apple released recently...$6k gets you a 720GB Fibre-Channel RAID array in a 3U enclosure. Not bad at all, really.
Assuming that you'd have to buy at least some of the Gb networking hardware (switches, cables, etc.), you're really not going to be saving much. Assuming at least $100 per 'RAMdisk server', you'll be spending $1200+ for a ~20GB RAID array that will lose everything the minute the power blinks, not to mention drawing several kilowatts of AC.
On the other hand, if you just throw four 100GB ATA-100 drives in a standard tower case with a decent IDE RAID controller, you get five times as much storage for probably about half the money.
Also, remember that most low-to-mid-range PCs can't actually fully take advantage of a gigabit network link, since the PCI bus and CPU get saturated long before the network does.
One potential extension not raised in the article (though it was perhaps hinted at in the discussion of attributes attached to each document) is the use of explicit hypertext link information to affect the vector value of each document in a collection. One could even 'stem' the target URLs to the top-level domain, affecting the same kind of term-space reduction as the Porter algorithm does for English text.
Now that I think of it, to some extent, both the Google PageRank methodology and some of the emerging weblog and news-tracking tools utilize hyperlink information very effectively to build a sort of loose semantic grouping between pieces of web content. However, from what I understand of the PageRank algorithm, at least, the conceptual model is much closer to basic flow graphs than vector spaces, and it (arguably unfairly) more or less completely ignores 'island' sites or pages, no matter how closely they match a particular query.
I'm actually working on a simple search engine implementation for a couple of small websites, and may have to tinker a bit with some of this.
One thing I'm not sure about, though: are there any standard, efficient *persistent* data stores/structures especially suitable for this kind of vector data? I'm hoping to work in Ruby, which unfortunately lacks a complete mathematical data library like PDL, but can easily hook in either a simple C library or a native implementation of any appropriate structures.
Please, don't start out making a quite viable point (namely, that "shrink-wrap" or "pre-installed" Linux systems are far more user-friendly) and the screw it up by saying that desktop Linux is ready to go head-to-head with OS X or Windows XP. My two primary computers run Linux and NetBSD, and for my purposes (90% of my time is spent in vi, bash, and perl/python/ruby scripting) they're great.
/etc files, all *not* alike.
However, there's no way that I would seriously try to give a Linux system to my girlfriend, or even my fairly tech-savvy younger brother, simply because I know that they'd spend a lot of time trying to figure out how to do things that were idiotically simple on their Macs and damn-near impossible under Linux.
It's getting better, especially in the area of basic UI consistency for desktop apps -- and as much as I can't stand running a RedHat system, they're leading the way with the BlueCurve unification of GNOME and KDE. However, it'll probably be at least a couple more years before the Linux desktop situation improves enough to make a switch a less-than-excruciatingly-painful experience for the large majority of PC/Mac users.
Of course, by that time, the OpenBeOS crew will either have given up, or pulled off a working system, in which case I hope to be able to forget about XWindows, custom-built kernels, and the thousand twisty little
While I agree that the above post seemed to be in a "toungue-in-cheek" kind of voice, it also made a valid point: for any average business user, even one who is comfortable with doing their own Windows installations/upgrades, the free *NIX world (and the BSDs in particular, w/o RedHat or similar to put out the slick, documented installers) is a challenge to get into.
Personally, I'm very happy doing day-to-day coding and net work on an old Thinkpad 560 running NetBSD, and will probably try FreeBSD 5.0 if the release installer supports wireless PCMCIA cards (NetBSD 1.6 did, much to my surprise!). However, the value of the BSDs is much like the value of Debian, IMHO: they offer lots of low-level control, so you *can* do crazy things (and install them in crazy places) that the maintainers couldn't have imagined.
The most interesting works of modern fiction I've read were seldom examples of the finest writing style -- that's what the "classics" are for. Their value is more in the concepts and characters they introduce than in their use of language.
I think that Doctorow's short stories do a good job of presenting a scenario based on a popular current meme, while the idea is still fresh.
Don't wait for your favorite development tools and languages to get ported to PocketPC -- bring your PocketPC device to them!