using a farm of FreeBSD boxes to serve content stored on monstrous Solaris database servers, and (gasp) franking out images of the database to static html pages several times a day.
Dejanews looks pretty naive too -- they must have naively dicretized their problem, like those naive scientific computation types have been doing for 30 years.
Damn, when WILL the world grow up to be as cool as you? I was starting to respect your opinions, and then you serve up this fragrant little turd (while some baboon moderates it up, presumably to encourage flaming).
Hint: market capitalization depends on the ideas and the implementation. Replication is not a new idea, either. Basically, your central thrust -- that transactions and state require single-box solutions -- is belied by 30 years of data processing in large businesses. It is not always *easy* to implement distributed solutions, but it is usually *cheaper* and at least *possible*.
This is actually not so hard. I thought about it for a while and several things occurred to me.
1) The benchmark should be run 3-way: Solaris x86 Linux 2.2.8 NT 4.0sp5
2) The high-end server must use a gigabit NIC
3) The low-end server must not be a PIII (Microsoft will almost certainly hack in optimizations to take advantage of KNI)
4) A test of database-backed web performance is mandatory (use MySQL... heheheh...)
5) The tests should be run in a neutral setting (maybe once each at VA and MS labs)
With these demands met, it should be possible to get a fair test, and even if Linux does get clobbered on the high-end box, Solaris should not. Linux will destroy NT on the lower-end box. So two sets of numbers are produced:
"Solaris outperforms NT4 in the enterprise" and "Linux destroys NT4 for entry-level servers"
which will be spun by Microsoft as
"NT4 outperforms Linux in the enterprise " and "NT4 offers better price/performance than Solaris"
but no one will care and the matter can be put to rest. Microsoft won't be conclusively demolished (and hence will be unlikely to try legal means to suppress the results) but neither will Linux, and (if these demands are made, loudly and publicly) MS will have to rise to the challenge.
It is a very important addition to the discussion as it fills in parts of McNealy's statements which were unclear. If I had any moderator points left (spent 'em all on the Thompson article) I'd do it. This is an insightful reply -- Sun has a great deal more invested in hardware than software, and Java is simply a means to their end. It isn't a moneymaker (unless you sell books) but both Sun and IBM are pushing furiously for Java because they stand to leverage their position in Big Iron for the future of computing. IBM in particular has written more Java code than Sun has -- don't get too excited, IBM has written more NT code than Microsoft, it just points out where the money lies.
is a whole hell of a lot more useful than Quake skills, to be quite frank. Hell, you can pimp your kinks out a lot easier than Quake skills.
I liked those Romans, they had their priorities straight. OTOH most people seem to think of programming as something twisted and perverted (witness VB "programmers") so maybe we can just lump real programming in there with hot wax and ball gags;-P.
The PC won over the mainframe? Perhaps in the general-utility-computing arena, but for truly obscene loads and outrageous availability requirements, mainframes still rule. Ask E-Schwab, for example, or REI.
Mainframes will never die. The legacy system of tomorrow will be mainframe transaction-processing systems fronted by SP/2 analysis clusters, with something like Linux or NT gating the whole mess to the web. I can almost guarantee it.
Then don't act like one. Your post made you sound like a spoiled child, which probably was not what you intended.
Much respect for the mail filtering system (though my old job entailed similar duties for 7000 people), but you really must deliver your point with more finesse if you want it to stick.
For example,
>> you can all F**K off
is not a persuasive rebuttal to anyone's argument.
I have compiled, built, and run a fair number of applications on (Debian, RedHat, and Open) Linux, Irix, SunOS, AIX, FreeBSD, and SCO... usually it is a bit of trouble but if the code is inherently portable, more a matter of some #ifdef's and symlinks. (Well, for SCO it's usually worse because they strive to provide broken development libraries... SCO is evil and must be destroyed.)
If you have trouble getting mod_jserv built and running, take a look at Ari Halberstadt's great guide to Using Apache-Jserv. It can save you time and frustration.
You could also hire him. (heheheh)
Finally, for Apache 1.3.4 and 1.3.6, save yourself some trouble by symlinking {APACHE_INSTALL}/bin to {APACHE_INSTALL}/sbin.
Take a look at Using Apache-Jserv and persevere. It took me a little while to set things up in Linux, but now we have embedded Java and Perl interpreters and everyone's delighted.
Also, I don't have to depend on an evil monolithic corporation for my server software this way.
If you're a pro, and want to write code that could comfortably live in an algorithms book (eg. "executable pseudocode"), Python is The Way. But I never got PyApache to compile and I don't like using anything but Apache so I gave up on Zope (couldn't get it to work via pcgi with Apache).
Then there's Perl. It's ugly and it WORKS. That's all there is to it. Do what you will, just don't let a newbie Perl hacker anywhere near a production mod_perl system or you will regret it. Perl is tightly integrated with the Apache API via mod_perl, so you can do damn near anything in Perl that you can in C. Like AOLServer and Tcl or Roxen and Pike, Perl and Apache are an item. For building up libraries of reusable functions and objects you create Perl modules, simple as that.
Java servlets are like Scheme programs if you do them right. You get a bunch of beautifully polished classes (because it just takes too damn long to write crappy code in Java) that are ideal for reuse, if you prototyped and designed well. Sessions are great, and the servlet API makes them trivial to use. Servlets are much, much better (IMHO) for production systems that have been slapped together as prototypes in something like Perl or Python. (an interesting extension to this idea is using JPython to write servlets... haven't heard of any recent successes but you could try) And the Apache-Jserv project is a great community thing -- very dynamic, lots of great new ideas. But Java is a (manually) compiled, strongly typed language and a servlet usually takes more time to write than the equivalent one-shot Perl script.
My boss likes writing servlets because he can't deal with Perl, but doesn't mind Java at all. I like prototyping in Perl because it's fast, but god damn it feels good to come up with an elegant class and stick it in our local project's.jar. People who hack on supercomputers and BSD guts invariably seem to love Python, but I would have trouble hooking up Python to Oracle as cleanly as we have done with Java and Perl (total abstraction from the database via JDBC and DBI). So go with what makes you happy, it'll keep you from burning out. And use ApacheDBI if you go with Perl.;-)
There are a lot of clever hacks which Hunter shows how to do, and he's *not* a dull blade. You might want to leaf through it before shooting off -- things like image manipulation and security are in this book, and I found those to be quite helpful. Oh, and sessions. Sessions are tres cool.
If you're a wizard then duh, it won't help you. In that case head on over to the cocoon website and think about contributing to the project. Show everyone what a stud you are and work with some really neat material at the same time.
There are a lot of clever hacks which Hunter shows how to do, and he's *not* a dull blade. You might want to leaf through it before shooting off -- things like image manipulation and security are in this book, and I found those to be quite helpful. Oh, and sessions. Sessions are tres cool.
If you're a wizard then duh, it won't help you. In that case head on over to the cocoon website and think about contributing to the project. Show everyone what a stud you are and work with some really neat material at the same time.
It's ugly and it's slow (well duh, debugging symbols and no optimization) but this is the first build in approximately forever that has worked for me. I was going to use the New and Improved build process to make my own but, well, it just *worked*. GtkStep makes it dump core, though.
Pretty cool... client side XSL would make this thing rule beyond all. Too bad I have no idea how to do that on the browser side.
If you're so valuable it should be trivial to get a job elsewhere for a reasonable salary. Decent programmers actually *are* in demand, you just have to force the issue.
Linux lacks a journaling file system, which means that if fsck can't fix a disk after a power failure or system crash, you're pretty well hosed and must restore from the most recent backups. That's not good for availability. You want a log of reads and writes to disk so that a consistent image can be rebuilt after a crash. Database vendors realized this decades ago, and high-end Unices followed suit with their filesystems.
Linux does SMP approximately as well as Windows NT, as of the 2.2 kernels. That's nothing to write home about, compared to Irix or Solaris.
Linux distributions do not yet conform to a base standard filesystem or level of library functionality (LSB standards). That's bad for interoperability. One hopes it will be fixed.
All of these can be seen as weaknesses. Or, if you are so inclined, they can be seen as goals. About the only thing which the above are NOT is evidence of Linux being mature. Note that the BSDs are no different (AFAIK) in this respect, save for the standards (only one distro of each).
Whether DH Brown was being misled or just didn't dig deep enough, they should not have said that NT is a contender. But the fact is that Linux isn't either, if you want high-end scalability *today*.
Slashdot Mirror
using a farm of FreeBSD boxes to serve content stored on monstrous Solaris database servers, and (gasp) franking out images of the database to static html pages several times a day.
Dejanews looks pretty naive too -- they must have naively dicretized their problem, like those naive scientific computation types have been doing for 30 years.
Damn, when WILL the world grow up to be as cool as you? I was starting to respect your opinions, and then you serve up this fragrant little turd (while some baboon moderates it up, presumably to encourage flaming).
Hint: market capitalization depends on the ideas and the implementation. Replication is not a new idea, either. Basically, your central thrust -- that transactions and state require single-box solutions -- is belied by 30 years of data processing in large businesses. It is not always *easy* to implement distributed solutions, but it is usually *cheaper* and at least *possible*.
Moreover, HA pretty much requires it.
This is actually not so hard. I thought about it for a while and several things occurred to me.
1) The benchmark should be run 3-way:
Solaris x86
Linux 2.2.8
NT 4.0sp5
2) The high-end server must use a gigabit NIC
3) The low-end server must not be a PIII
(Microsoft will almost certainly hack in
optimizations to take advantage of KNI)
4) A test of database-backed web performance is
mandatory (use MySQL... heheheh...)
5) The tests should be run in a neutral setting
(maybe once each at VA and MS labs)
With these demands met, it should be possible to get a fair test, and even if Linux does get clobbered on the high-end box, Solaris should not. Linux will destroy NT on the lower-end box. So two sets of numbers are produced:
"Solaris outperforms NT4 in the enterprise"
and
"Linux destroys NT4 for entry-level servers"
which will be spun by Microsoft as
"NT4 outperforms Linux in the enterprise "
and
"NT4 offers better price/performance than Solaris"
but no one will care and the matter can be put to rest. Microsoft won't be conclusively demolished (and hence will be unlikely to try legal means to suppress the results) but neither will Linux, and (if these demands are made, loudly and publicly) MS will have to rise to the challenge.
It is a very important addition to the discussion as it fills in parts of McNealy's statements which were unclear. If I had any moderator points left (spent 'em all on the Thompson article) I'd do it. This is an insightful reply -- Sun has a great deal more invested in hardware than software, and Java is simply a means to their end. It isn't a moneymaker (unless you sell books) but both Sun and IBM are pushing furiously for Java because they stand to leverage their position in Big Iron for the future of computing. IBM in particular has written more Java code than Sun has -- don't get too excited, IBM has written more NT code than Microsoft, it just points out where the money lies.
is a whole hell of a lot more useful than Quake skills, to be quite frank. Hell, you can pimp your kinks out a lot easier than Quake skills.
;-P.
I liked those Romans, they had their priorities straight. OTOH most people seem to think of programming as something twisted and perverted (witness VB "programmers") so maybe we can just lump real programming in there with hot wax and ball gags
Damn it feels good to be a gangsta...
Upgrade of course. Looks like you haven't switched to 2.2 yet... besides, who cares how long a box can stay up with a load average of 0.00?
Hell, NT could do that.
The PC won over the mainframe? Perhaps in the general-utility-computing arena, but for truly obscene loads and outrageous availability requirements, mainframes still rule. Ask E-Schwab, for example, or REI.
Mainframes will never die. The legacy system of tomorrow will be mainframe transaction-processing systems fronted by SP/2 analysis clusters, with something like Linux or NT gating the whole mess to the web. I can almost guarantee it.
Of course, Linux is being ported to the S/390...
>> I am NOT a halfwit...
Then don't act like one. Your post made you sound like a spoiled child, which probably was not what you intended.
Much respect for the mail filtering system (though my old job entailed similar duties for 7000 people), but you really must deliver your point with more finesse if you want it to stick.
For example,
>> you can all F**K off
is not a persuasive rebuttal to anyone's argument.
Supplying the schools of the sovereign nation of Chicago is not as impressive a publicity stunt as influencing an entire country.
Besides, anywhere in North America is likely to be stocked with bureaucrats who are in Microsoft's pocket (*cough* SLADE GORTON *cough*).
Hey, I've got an idea. How about they supply the schools of Redmond with free Linux WordPerfects?
;-)
$7 for the CDROM and shipping
;-)
$12 for the manual
$0 for RedHat. Maybe I'll send a donation.
to solving damn near any problem with Unix/Linux.
I have compiled, built, and run a fair number of applications on (Debian, RedHat, and Open) Linux, Irix, SunOS, AIX, FreeBSD, and SCO... usually it is a bit of trouble but if the code is inherently portable, more a matter of some #ifdef's and symlinks. (Well, for SCO it's usually worse because they strive to provide broken development libraries... SCO is evil and must be destroyed.)
YMMV.
If you have trouble getting mod_jserv built and
running, take a look at Ari Halberstadt's great
guide to Using Apache-Jserv. It can save you time and frustration.
You could also hire him. (heheheh)
Finally, for Apache 1.3.4 and 1.3.6, save yourself some trouble by symlinking {APACHE_INSTALL}/bin to {APACHE_INSTALL}/sbin.
Take a look at Using Apache-Jserv and persevere. It took me a little while to set things up in Linux, but now we have embedded Java and Perl interpreters and everyone's delighted.
Also, I don't have to depend on an evil monolithic corporation for my server software this way.
title says it all. GnuJSP for Apache-Jserv is at 0.99, the only thing left is for Sun to finalize the specification.
Oh, and did I mention Cocoon? Stefano's "XSP" idea (XML Server Pages) is probably the direction where JSP is headed anyhow.
...almost every good idea I've heard from engineers and programmers...
(Regarding the Microsoft problem, that is.)
No way that's gonna happen. Why is RMS wasting his time on hypothetical situations? ;-)
What RMS does is distill almost every good idea I've heard from engineers and programmers into a single page. This is a great article; read it.
If you're a pro, and want to write code that could comfortably live in an algorithms book (eg. "executable pseudocode"), Python is The Way. But I never got PyApache to compile and I don't like using anything but Apache so I gave up on Zope (couldn't get it to work via pcgi with Apache).
.jar. People who hack on supercomputers and BSD guts invariably seem to love Python, but I would have trouble hooking up Python to Oracle as cleanly as we have done with Java and Perl (total abstraction from the database via JDBC and DBI). So go with what makes you happy, it'll keep you from burning out. And use ApacheDBI if you go with Perl. ;-)
Then there's Perl. It's ugly and it WORKS. That's all there is to it. Do what you will, just don't let a newbie Perl hacker anywhere near a production mod_perl system or you will regret it.
Perl is tightly integrated with the Apache API via mod_perl, so you can do damn near anything in Perl that you can in C. Like AOLServer and Tcl or Roxen and Pike, Perl and Apache are an item. For building up libraries of reusable functions and objects you create Perl modules, simple as that.
Java servlets are like Scheme programs if you do them right. You get a bunch of beautifully polished classes (because it just takes too damn long to write crappy code in Java) that are ideal for reuse, if you prototyped and designed well. Sessions are great, and the servlet API makes them trivial to use. Servlets are much, much better (IMHO) for production systems that have been slapped together as prototypes in something like Perl or Python. (an interesting extension to this idea is using JPython to write servlets... haven't heard of any recent successes but you could try) And the Apache-Jserv project is a great community thing -- very dynamic, lots of great new ideas. But Java is a (manually) compiled, strongly typed language and a servlet usually takes more time to write than the equivalent one-shot Perl script.
My boss likes writing servlets because he can't deal with Perl, but doesn't mind Java at all. I like prototyping in Perl because it's fast, but god damn it feels good to come up with an elegant class and stick it in our local project's
If you're a wizard then duh, it won't help you. In that case head on over to the cocoon website and think about contributing to the project. Show everyone what a stud you are and work with some really neat material at the same time.
If you're a wizard then duh, it won't help you. In that case head on over to the cocoon website and think about contributing to the project. Show everyone what a stud you are and work with some really neat material at the same time.
What's the difference? Besides the availability of source for SSH, that is.
Rather than polished software like Windows 98...
It's ugly and it's slow (well duh, debugging symbols and no optimization) but this is the first build in approximately forever that has worked for me. I was going to use the New and Improved build process to make my own but, well, it just *worked*. GtkStep makes it dump core, though.
Pretty cool... client side XSL would make this thing rule beyond all. Too bad I have no idea how to do that on the browser side.
If you're so valuable it should be trivial to get a job elsewhere for a reasonable salary. Decent programmers actually *are* in demand, you just have to force the issue.
That's Disraeli, hoss. Not Churchill.
errr, 7 layer networking model.
Physical, uhhhh.... application layer... uhhh...
Yeah, so anyways, it's sort of a pretend thing that someone at IBM or BBN came up with a long time ago. Anyways, here's a reference:
http://www.europa.com/~dogman/osi/
Linux lacks a journaling file system, which means that if fsck can't fix a disk after a power failure or system crash, you're pretty well hosed and must restore from the most recent backups. That's not good for availability. You want a log of reads and writes to disk so that a consistent image can be rebuilt after a crash. Database vendors realized this decades ago, and high-end Unices followed suit with their filesystems.
Linux does SMP approximately as well as Windows NT, as of the 2.2 kernels. That's nothing to write home about, compared to Irix or Solaris.
Linux distributions do not yet conform to a base standard filesystem or level of library functionality (LSB standards). That's bad for interoperability. One hopes it will be fixed.
All of these can be seen as weaknesses. Or, if you are so inclined, they can be seen as goals.
About the only thing which the above are NOT is evidence of Linux being mature. Note that the BSDs are no different (AFAIK) in this respect, save for the standards (only one distro of each).
Whether DH Brown was being misled or just didn't dig deep enough, they should not have said that NT is a contender. But the fact is that Linux isn't either, if you want high-end scalability *today*.