It gets a bit weird actually. Apparantly if you *buy* the license to a rh server then you can't install it on any other machine from the rh EULA agreement. However because it is GPL'd I believe you should legally be able to strip out all the trademarked pictures etc and make it "available" as an iso publically (I.E. redistribute it for free) Now everyone *but* you can install it on other machines w/out buying a per license fee. God knows what kind of mess it would be if someone modified the gpl'd stripped version and the original person "reinstalled" rhel from that.
After seeing that non-response to the academic pricing, I was annoyed enough to call redhat up to see what is going on. They have not finalized the pricing but it is going to be $25->$30 a year per machine w/ some minumum number of machines. They also will offer a "proxy server" for $2500 so you can roll your own updates.
(1) The proxy server idea is stupid since apt is ported over to redhat. If we know how to setup a ftp server we automagically get a "proxy" server. Unless they are not going to let us grab the updated rpms (from our $25 per node licensed "up2dates") and copy it to the server which if they dont as far as I can tell violates the gpl (unless they do an "endrun" by providing rpm's w/ "trademarked" pictures in it).
(2) $25 a machine/year is more expensive than windows for updates! Yes redhat is a smaller company BUT at microsoft is writing their own gui, writing their own kernel, writing their own patches, writing their own driver specs. Redhat is NOT doing these things on their own (but to be fair they are contributing but having per machine licensing is a trick worthy of SCO not a linux company).
(3) We are in a physics dept and run "oscared" images for a smallish beowulf cluster (50 dual nodes) and have two more beowulf clusters and one public access workstation cluster in our department alone. It makes *no* sense to pay $25 a node for this since we "automirror" when things go awry. I personally will move over to debian (or wait and see what fermilabs etc is going to do). This redhat fiasco is a fiasco. Using "trademarked" pictures to do an endrun around the gpl to get "per processor" licensing is an end run around the GPL and ought to be treated as such. (This is, as far as I can tell, what they are doing w/ "enterprise" redhat to prevent me from buying one "enterprise" redhat and apt-proxying my other nodes). Feel free to correct me if I am mistaken.
Whoa that is very cool. I didn't realize Atlas was a MS project actually. Whaley should honestly get a lot more credit that I think he does (at least from the current sourceforge description). Admittedly, i'm a physicist and we tend to be a bit biased about the importance about these things:) But Atlas really epitomized one of the coolest ideas I'd seen in recent times.
-avi
I think this is actually what most two pass compilers will do. The intel compilers have this option for instance. Basically you compile w/ profiling on and on the second pass it using the runtime profiling data to recompile source to a new object code. (btw the intel compilers are "free" for opensource/nonprofit use). W/ regards to using GA/SA as part of the optimization it would be (to be honest) a bit weird to have a nondeterministic compiler since in different runs you may end up w/ different optimizations.
There is a variation of this that is very clever that optimizes BLAS and LAPACKroutines "empirically" called ATLAS that has been around a while, originally (and perhaps still) a research project by err Clint Whaley from tennessee (BLAS/LAPACK are numerical routines that do a slew of thing people generally find useful linear algebra and vectors). These routines will often time sit "under" many mathematical packages like matlab/octave/maple/mathematica/scilab as well as make up the core of much custom scientific computing packages (or even libraries like the "Gnu Scientific Library")
Basically the jist is atlas "empirically" (read: use an optimizer for instance like GA, though empirical may actually mean brute force in this case) to optimize various parameters that will affect things like optimize the routine for the cache size of the processor etc. The cool thing about this, is they can get w/in 10% of hand machine coded BLAS/LAPACK libraries w/out the pain!
Out of curiousity what is in the "enteprise" version? The irony of redhat is that while it is reasonably straightforward to "upgrade" desktops to the latest and greatest redhat we generally keep "server" machines on a version or two behind. The intel compilers had an issue w/ redhat9 until some work arounds were found. OSCAR the node management utility for mpi/beowulf clusters seems to be unstable for rh9 such that the vendor gave us redhat 8 on our machines. As far as I am aware this is a standard redhat 8. If node management/clustering gridware/mpi linux clusters don't need "enterprise" who does and what could it possibly include that couldn't be rpm'd or "apt-getted" from elsewhere? I can't imagine that the non-enteprise contains a crippled kernel.
(I just googled a bit and an ssl'd apache is included, anything else?)
By saying they are no longer supporting standard rh standard does this translate to just no iso's or just alias fedora rh?
Are you positive that the wattage is the same when under no load? I forget the actual numbers but the amp readings jump sky high when we actually use our dual xeon machines (correspondingly the temperature of the chips jump from 22C to 60C). I got the impression that the current draw is highly dependent on what you are doing w/ the processor on most modern processors..
One problem w/ this sort of research is it comes off as very, very hypocritical to the rest of the world. Particularly the 3rd world. It is exactly for this reason that India/Pakistan have not signed the nonproliferation treaty. America wants to have it both ways, (1) Though shalt not deal w/ chemical/biological/nuclear weapons on "mass direction" (2) Except if you are america. The US can get away w/ this because it is an economic powerhouse and sanctions affect smaller countries on a large scale. But this blatant hypocracy gives credence to any powerhouse thug wannabe to start/develop any "mass destruction" programs because the "bad guys" (read americans) are already doing it and so if they don't they will get wiped out. You can't expect people to follow rules that you yourself do not follow. It doesn't work for children and it doesn't work for countries.
He's right at least in the scientific community. Most of us prof/gprof our code immensely and take advantage of everything the hardware can give us at the very least on the compiler level.
This is why you'll have developers extoll the virtues of single multiply adds that IBM's power architecture has if they are doing numerically intensive code (now incorporated into the G5) or perhaps the memory management of the Sun boxes if they are running code that necessitates having > 4 gb ram.
Things like "cache size" matters to developers in a way that it does not to sysadmins. On the other hand I get the impression that much of the "sparc is god" raraing comes from development/running of Java code or running/developing websites. In any case it makes much more sense to me that developers do profiling of their own code and thus know the performance issues of the architecture than a sysadmin.
Libertarian except when it comes to your money eh? So I should contribute to your 100% markup and six figure salary because i've got to support the "scene" and the local shops? I dont think. I will be realistic however, I make enough money that if I want to buy 1/2 records and/or i'm going used cd shopping I buy locally. However if I buy 5-10 cds I *will* buy mailorder and/or the local shop that actually is worth going to. Maybe you can get away w/ 100% mark up and $15-$16 dollar in bfe Il because you have no competition in the "indie" market and charge what you like (as you said your only competition is the net which gives you your insane markup). However out here "free market" capitalism works enough in the indie market to have well priced local stores ($12-$13) w/ well priced options. If they didn't I dont see any need to be apologetic about going mailorder not providing you w/ a six figure salary w/ 100% markup on your localized monopoly.
No no it has nothing to do w/ being "indie" its just marketing. At $15-16 a cd I'll rather go mailorder which sells major labels as well. There are plenty of record stores in most major cities that sell cds at $12-13 including a fair amount of major label music and they apparantly all make money off of it. As a consumer of more than your average amount of cds it doesn't make sense to pay $15-$16 a pop when there are plenty of stores selling for much cheaper (let alone mail order). Whatever, if major labels are retailing at $13 it doesn't make any sense for the indies to now cost more. Or perhaps, come to think of it, it does since the indies tend to have "cult" followings of people who will buy the records anyway..
Wow! I remember thinking it was madness it had shot up to $10 on this ludicrous war and $15-$16 was the peak of the insanity. Rational markets indeed:!
(From article)
"In other news, SCO's stock surged $4.97, or 32 percent, to close at $20.50 Wednesday, after Deutsche Bank analysts Brian Skiba and Matthew Kelly initiated coverage of the company with a "buy" rating and a $45 price target for the stock"
Argh the over priced indie store. Don't get me wrong I think its cool that you are doing so well but I can't imagine shopping at your store w/out being a bit annoyed (indie store w/ Sam goody prices). The prices in ny/philly/boston aren't that high. Even newberry comics (boston chain) has the $12 standard.
This is nothing like your bike example. On a bike you know that it is going to tip over when you stop. W/ the segway there is a possibility that if you press the acceleration the thing will die even w/out a warning light! They are obviously setting the "warning light" to go on way too late, hitting the acceleration shouldn't make the thing die ever w/out a warning.
I just got back from the Peru "software libre" conference. It was reasonably informative. The most common recurring theme that I gathered (though not explicitly put in these terms).
(2) Cost, cost, cost.. This is particularly a big deal for the government.
Someone from Oxfam (the NGO) who was sitting in the audience mentioned that software licenses are 2nd biggest export of money in third world countries,
(3) Practical examples of software development in the south america. One particular example was the development of an (my words) "polytechnic" education course system based on free software in brazil. (Also IBM South America was there but he basically got SCO questions:))
(4) The moral choice of Free Software by Stallman who else;)
There is a fix for this thats inherent in unix but for some reason in this age of "rpm hell" people have forgotten. Incompatible versions of libraries should carry different major version names while compatible versions may carry minor versions. If people stick to this (which unfortunately in practice doesn't really hold up i.e. all the __ctype_b errors w/ precompiled code w/ the glibc in the last rh upgrade), one could/SHOULD actually create for all libraries a back-compatible set of rpms which carries all previous major versions of that library.
for instance my rh9 has a glibc-2.3.2-27.9 we could make an rpm called compat-lib-glibc-2.3.2-27.9 which will carry libc.so.5 libc.so.4 libc.so.3 libc.so.2 libc.so.1
especially in this day an age of huge hard drives this seems like at least a partial fix of this horrible rpm hell. It may also get us to be a bit more careful about major/minor revisioning and make sure that minor versions are truly compatible while major versions are truly not.
On the other hard the "forced upgrade" cycle definitely promotes the idea of providing source w/ ones binaries at it becomes much easier for "distributions" to ensure compatibility.
Poindexter is going to resign! now (nytimes registration required blah blah)
-bloo
Re:Alvin Fernald, Danny Dunn, and Henry Reed
on
The Big Kerplop
·
· Score: 1
Wow danny dunn!! I had completely forgotten about that series, I had read this entire series when I was back in england (~9 yrs old) and had thought this was some brit series that america hadn't gotten. The other brit. series like this was something called the "famous five" which was a series of 5 kids on adventures. I remember when I came to america the other series that I thought was quite good (though very 50s genre sci fi) was the "tom swift" series that our local library had. I've never heard of this alfred series but i'll definitely look it up (i'm actually a college level physics prof now so it'll have to be a secret guilty pleasure:))
I use pine! I'm not sure what you mean by "default" but what is wrong w/ doing something like imap? There is nothing wrong w/ picking a default for people i.e. mozilla but let people choose/configure what they would like. (that is is it switchable?) Personally I only know a few people who use mozilla.
Just so you know, xeons are pretty cheap if you scale back to 2.66 dual xeons (these are xeon not xeon mp chips) you can get them for ~1600 a 1U node w/ a gig of memory.
Are you in europe? The De/vision one hit me as well. My understanding is w/ de/vision they switched (in germany) to BMG so they are having "major label woes". The one that hit us right before then was VNV nation singles (the genesis ones). I am actually now really wary of buying european cds anymore w/out first staring at the package for a long time to verify that this isn't an issue. Incidentally, the domestic american releases of these albums do not have this issue, that is the domestic de/visions were not copyprotected (two,remixed, devolution etc).
I happened to be on the alt.synthpop and got a response from de/visions manager actually a
while back on this issue:
"From lorenz"
as far as i know, i can't influence it, because there are BMG guidelines. all the BMG-cds are printed at the same place in germany. Don't know what will happened during the next 12 month, because the next DE/VISION CD is sceduled for 2003 and then they maybe will have new guidelines or copyright protections.
I'm a bit surprised by the blanket sun v*80 machines w/ sun chips run faster than the intel/xeon chips. The code I run is numerically intensive and memory intensive. I have a v880 based on the 933cu chip (4 proc/8 gigs of ram) as well as a "graphic workstation" w/ 2 gigs of ram and 2 of the same chip. I also have bought some cheapy dell "xterminals" that cost $650. Two of these machines are sdram based PIV motherboards w/ 512mb of ram and a 2.4 ghz chip. Using intel's compilers for *our code* the intel chips are wallclocking at twice as fast as our sun chips. (Non MPI single processor runs that are probably using currently about 200 mb of memory) I also have an old rdram based PIV machine also running at 2.4ghz and get another 50% speedup over the sdram based mb (slightly more expensive glorified xterminal). Some of the things we noticed that was a bit surprising is that intel's compilers did really incredible things w/ our code in contrast to gnu while the sun compiler gave us hardly any speedup. Quite honestly, the only case when the sun machine will be worth using is if we need to use (stablely) > 4 gigs of ram.
I've seen a few comments of the hotswapping redundancy issue. I think this is important but I think that sun oversells this. If you have a critical datastore or somesuch yes it is worth that sort of money, but there are a lot of things that are "computationally" intensive that doesn't need the immediate restore losslessness. i.e. the who cares if the machine dies, resubmit the job on a different node, google philosophy. The is afterall the underlying truth behind all the distributed projects and the "beowulf" cluster issue. Needless to say I see a lot of people saying the sun boxes at 1ghz are equivalent to 2.4 ghz xeons. This is how sun is trying to sell their blade servers. I can currently build xeon machines at ~1500 a dual and amd duals at ~1100 a dual while the sun blades are going at (I believe) $999 a processor/blade and those chips are way slower than my v880. I have say $80k to get the most cpu/$ I can afford. Given the fact I can reasonably distribute my code (i.e. don't need "large memory machines > 16gb") and I don't need redundancy and/or terabyte level fiber storage etc (i.e. job dies I resubmit it, blade philosophy) I honestly find it hard to believe that my money is better spent on buying single processor blades than buying dual processor xeons at not much more.
-bloo
I have a v880 w/ the 933mhz cu chips w/ 8gigs of ram and 4 processors and fiber drives as well as a 2 processor workstation based on the same chip. For our code which is pretty memory intensive and cpu intensive (numerical ode integration) running on *old* sdram based 2.4 ghz pIV using intel's compilers the intel code runs about twice as fast. The intel processor machines were cheap $650 dell boxes (at the time) that we bought to use as glorified "xterminals". I also have a PIV box that is based on rdram which was about a (at the time) a $1600 dell box and I get about another 50% speedup on that machine in comparison to the old dell machines. Its only a single point of data and should be taken as such but thats how the numbers played out for us. Incidentally, one of the biggest surprises were intel's compilers gave a drastic improvement on the intel chips over gnu, while we got practically no speedup going from gnu to suns own compilers on the v880.
Have you checked the power draw on the machine? Admittedly its a low end machine but I would guess that the power draw is significantly more than a generic ~$50 firewall/switch/router to the point where the ~$50 switch would pay for itself w/in a year or two..
Slashdot Mirror
It gets a bit weird actually. Apparantly if you *buy* the license to a rh server then you can't install it on any other machine from the rh EULA agreement. However because it is GPL'd I believe you should legally be able to strip out all the trademarked pictures etc and make it "available" as an iso publically (I.E. redistribute it for free) Now everyone *but* you can install it on other machines w/out buying a per license fee. God knows what kind of mess it would be if someone modified the gpl'd stripped version and the original person "reinstalled" rhel from that.
(1) The proxy server idea is stupid since apt is ported over to redhat. If we know how to setup a ftp server we automagically get a "proxy" server. Unless they are not going to let us grab the updated rpms (from
our $25 per node licensed "up2dates") and copy it to the server which if they dont as far as I can tell violates the gpl (unless they do an "endrun" by providing rpm's w/ "trademarked" pictures in it).
(2) $25 a machine/year is more expensive than windows for updates! Yes redhat is a smaller company BUT at microsoft is writing their own gui, writing their own kernel, writing their own patches,
writing their own driver specs. Redhat is NOT doing these things on their own (but to be fair they are contributing but having per machine licensing is a trick worthy of SCO not a linux company).
(3) We are in a physics dept and run "oscared" images for a smallish beowulf cluster (50 dual nodes) and have two more beowulf clusters and one public access workstation cluster in our department alone. It makes *no* sense to pay $25 a node for this since we "automirror" when things go awry. I personally will move over to debian (or wait and see what fermilabs etc is going to do). This redhat fiasco is a fiasco. Using "trademarked" pictures to do an endrun around the gpl to get "per processor" licensing is an end run around the GPL and ought to be treated as such. (This is, as far as I can tell, what they are doing w/ "enterprise" redhat to prevent me from buying one "enterprise" redhat and apt-proxying my other nodes). Feel free to correct me if I am mistaken.
best regards,
-bloo
Whoa that is very cool. I didn't realize Atlas was a MS project actually. Whaley should honestly get a lot more credit that I think he does (at least from the current sourceforge description). Admittedly, i'm a physicist and we tend to be a bit biased about the importance about these things :) But Atlas really epitomized one of the coolest ideas I'd seen in recent times.
-avi
I think this is actually what most two pass compilers will do. The intel compilers have this option for instance. Basically you compile w/ profiling on and on the second pass it using the runtime profiling data to recompile source to a new object code. (btw the intel compilers are "free" for opensource/nonprofit use). W/ regards to using GA/SA as part of the optimization it would be (to be honest) a bit weird to have a nondeterministic compiler since in different runs you may end up w/ different optimizations.
-bloo
as make up the core of much custom scientific computing packages (or even libraries like the "Gnu Scientific Library")
Basically the jist is atlas "empirically" (read: use an optimizer for instance like GA, though empirical may actually mean brute force in this case) to optimize various parameters that will affect things like optimize the routine for the cache size of the processor etc. The cool thing about this, is they can get w/in 10% of hand machine coded BLAS/LAPACK libraries w/out the pain!
(I just googled a bit and an ssl'd apache is included,
anything else?)
By saying they are no longer supporting standard rh standard does this translate to just no iso's or just alias fedora rh?
-bloo
Are you positive that the wattage is the same when under no load? I forget the actual numbers but the amp readings jump sky high when we actually use our dual xeon machines (correspondingly the temperature of the chips jump from 22C to 60C). I got the impression that
the current draw is highly dependent on what you are doing w/ the processor on most modern processors..
One problem w/ this sort of research is it comes off as very, very hypocritical to the rest of the world. Particularly the 3rd world. It is exactly for this reason that India/Pakistan have not signed the nonproliferation treaty. America wants to have it both ways, (1) Though shalt not deal w/ chemical/biological/nuclear weapons on "mass direction" (2) Except if you are america. The US can get away w/ this because it is an economic powerhouse and sanctions affect smaller countries on a large scale. But this blatant hypocracy gives credence to any powerhouse thug wannabe to start/develop any "mass destruction" programs because the "bad guys" (read americans) are already doing it and so if they don't they will get wiped out. You can't expect people to follow rules that you yourself do not follow. It doesn't work for children and it doesn't work for countries.
-bloo
This is why you'll have developers extoll the virtues of single multiply adds that IBM's power architecture has if they are doing numerically intensive code (now incorporated into the G5) or perhaps the memory management of the Sun boxes if they are running code that necessitates having > 4 gb ram.
Things like "cache size" matters to developers in a way that it does not to sysadmins. On the other hand I get the impression that much of the "sparc is god" raraing comes from development/running of Java code or running/developing websites. In any case it makes much more sense to me that developers do profiling of their own code and thus know the performance issues of the architecture than a sysadmin.
Libertarian except when it comes to your money eh? So I should contribute to your 100% markup and six figure salary because i've got to support the "scene" and the local shops? I dont think. I will be realistic however, I make enough money that if I want to buy 1/2 records and/or i'm going used cd shopping I buy locally. However if I buy 5-10 cds I *will* buy mailorder and/or the local shop that actually is worth going to. Maybe you can get away w/ 100% mark up and $15-$16 dollar in bfe Il because you have no competition in the "indie" market and charge what you like (as you said your only competition is the net which gives you your insane markup). However out here "free market" capitalism works enough in the indie market to have well priced local stores ($12-$13) w/ well priced options. If they didn't I dont see any need to be apologetic about going mailorder not providing you w/ a six figure salary w/ 100% markup on your localized monopoly.
-avi
No no it has nothing to do w/ being "indie" its just marketing. At $15-16 a cd I'll rather go mailorder which sells major labels as well. There are plenty of record stores in most major cities that sell cds at $12-13 including a fair amount of major label music and they apparantly all make money off of it. As a consumer of more than your average amount of cds it doesn't make sense to pay $15-$16 a pop when there are plenty of stores selling for much cheaper (let alone mail order). Whatever, if major labels are retailing at $13 it doesn't make any sense for the indies to now cost more. Or perhaps, come to think of it, it does since the indies tend to have "cult" followings of people who will buy the records anyway..
-bloo
"In other news, SCO's stock surged $4.97, or 32 percent, to close at $20.50 Wednesday, after Deutsche Bank analysts Brian Skiba and Matthew Kelly initiated coverage of the company with a "buy" rating and a $45 price target for the stock"
Argh the over priced indie store. Don't get me wrong I think its cool that you are doing so well but I can't imagine shopping at your store w/out being a bit annoyed (indie store w/ Sam goody prices). The prices in ny/philly/boston aren't that high. Even newberry comics (boston chain) has the $12 standard.
This is nothing like your bike example. On a bike you know that it is going to tip over when you stop. W/ the segway there is a possibility that if you press the acceleration the thing will die even w/out a warning light! They are obviously setting the "warning light" to go on way too late, hitting the acceleration shouldn't make the thing die ever w/out a warning.
I just got back from the Peru "software libre" conference. It was reasonably informative. The most common recurring theme that I gathered (though not explicitly put in these terms).
:))
;)
(1) Free software allows "nascent" software developers/local economies to jump start development.
(2) Cost, cost, cost.. This is particularly a big deal for the government.
Someone from Oxfam (the NGO) who was sitting in the audience mentioned that software licenses are 2nd biggest export of money in third world countries,
(3) Practical examples of software development in the south america. One particular example was the development of an (my words) "polytechnic" education course system based on free software in brazil. (Also IBM South America was there but he basically got SCO questions
(4) The moral choice of Free Software by Stallman who else
-bloosqr
There is a fix for this thats inherent in unix but for some reason in this age of "rpm hell" people have forgotten. Incompatible versions of libraries should carry different major version names while compatible versions may carry minor versions. If people stick to this (which unfortunately in practice doesn't really hold up i.e. all the __ctype_b errors w/ precompiled code w/ the glibc in the last rh upgrade), one could/SHOULD actually create for all libraries a back-compatible set of rpms which carries all previous major versions of that library.
for instance my rh9 has a glibc-2.3.2-27.9
we could make an rpm called compat-lib-glibc-2.3.2-27.9 which will carry
libc.so.5
libc.so.4
libc.so.3
libc.so.2
libc.so.1
especially in this day an age of huge hard drives this seems like at least a partial fix of this horrible rpm hell. It may also get us to be a bit more careful about major/minor revisioning and make sure that minor versions are truly compatible while major versions are truly not.
On the other hard the "forced upgrade" cycle definitely promotes the idea of providing source w/ ones binaries at it becomes much easier for "distributions" to ensure compatibility.
Poindexter is going to resign!
now (nytimes registration required blah blah)
-bloo
Wow danny dunn!! I had completely forgotten about that series, I had read this entire series when I was back in england (~9 yrs old) and had thought this was some brit series that america hadn't gotten. The other brit. series like this was something called the "famous five" which was a series of 5 kids on adventures. I remember when I came to america the other series that I thought was quite good (though very 50s genre sci fi) was the "tom swift" series that our local library had. I've never heard of this alfred series but i'll definitely look it up (i'm actually a college level physics prof now so it'll have to be a secret guilty pleasure :))
-bloo
I use pine! I'm not sure what you mean by "default" but what is wrong w/ doing something like imap? There is nothing wrong w/ picking a default for people i.e. mozilla but let people choose/configure what they would like. (that is is it switchable?) Personally I only know a few people who use mozilla.
-bloo
Just so you know, xeons are pretty cheap if you scale back to 2.66 dual xeons (these are xeon not xeon mp chips) you can get them for ~1600 a 1U node w/ a gig of memory.
-avi
Are you in europe? The De/vision one hit me as well. My understanding is w/ de/vision they switched (in germany) to BMG so they are having "major label woes". The one that hit us right before then was VNV nation singles (the genesis ones). I am actually now really wary of buying european cds anymore w/out first staring at the package for a long time to verify that this isn't an issue. Incidentally, the domestic american releases of these albums do not have this issue, that is the domestic de/visions were not copyprotected (two,remixed, devolution etc).
:
I happened to be on the alt.synthpop and got a response from de/visions manager actually a
while back on this issue
"From lorenz"
as far as i know, i can't influence it, because there are BMG
guidelines.
all the BMG-cds are printed at the same place in germany.
Don't know what will happened during the next 12 month, because the next
DE/VISION CD is sceduled for 2003 and then they maybe will have new
guidelines or
copyright protections.
greetings
lorenz
I'm a bit surprised by the blanket sun v*80 machines w/ sun chips run faster than the intel/xeon chips. The code I run is numerically intensive and memory intensive. I have a v880 based on the 933cu chip (4 proc/8 gigs of ram) as well as a "graphic workstation" w/ 2 gigs of ram and 2 of the same chip. I also have bought some cheapy dell "xterminals" that cost $650. Two of these machines are sdram based PIV motherboards w/ 512mb of ram and a 2.4 ghz chip. Using intel's compilers for *our code* the intel chips are wallclocking at twice as fast as our sun chips. (Non MPI single processor runs that are probably using currently about 200 mb of memory) I also have an old rdram based PIV machine also running at 2.4ghz and get another 50% speedup over the sdram based mb (slightly more expensive glorified xterminal). Some of the things we noticed that was a bit surprising is that intel's compilers did really incredible things w/ our code in contrast to gnu while the sun compiler gave us hardly any speedup. Quite honestly, the only case when the sun machine will be worth using is if we need to use (stablely) > 4 gigs of ram.
I've seen a few comments of the hotswapping redundancy issue. I think this is important but I think that sun oversells this. If you have a critical datastore or somesuch yes it is worth that sort of money, but there are a lot of things that are "computationally" intensive that doesn't need the immediate restore losslessness. i.e. the who cares if the machine dies, resubmit the job on a different node, google philosophy. The is afterall the underlying truth behind all the distributed projects and the "beowulf" cluster issue. Needless to say I see a lot of people saying the sun boxes at 1ghz are equivalent to 2.4 ghz xeons. This is how sun is trying to sell their blade servers. I can currently build xeon machines at ~1500 a dual and amd duals at ~1100 a dual while the sun blades are going at (I believe) $999 a processor/blade and those chips are way slower than my v880. I have say $80k to get the most cpu/$ I can afford. Given the fact I can reasonably distribute my code (i.e. don't need "large memory machines > 16gb") and I don't need redundancy and/or terabyte level fiber storage etc (i.e. job dies I resubmit it, blade philosophy) I honestly find it hard to believe that my money is better spent on buying single processor blades than buying dual processor xeons at not much more.
-bloo
I have a v880 w/ the 933mhz cu chips w/ 8gigs of ram and 4 processors and fiber drives as well as a 2 processor workstation based on the same chip. For our code which is pretty memory intensive and cpu intensive (numerical ode integration) running on *old* sdram based 2.4 ghz pIV using intel's compilers the intel code runs about twice as fast. The intel processor
machines were cheap $650 dell boxes (at the time) that we bought to use as glorified "xterminals". I also have a PIV box that is based on rdram which was about a (at the time) a $1600 dell box and I get about another 50% speedup on that machine in comparison to the old dell machines. Its only a single point of data and should be taken as such but thats how the numbers played out for us. Incidentally, one of the biggest surprises were intel's compilers gave a drastic improvement on the intel chips over gnu, while we got practically no speedup going from gnu to suns own compilers on the v880.
-bloo
Have you checked the power draw on the machine? Admittedly its a low end machine but I would guess that the power draw is significantly more than a generic ~$50 firewall/switch/router to the point where the ~$50 switch would pay for itself w/in a year or two..
-bloo
Check out this. Its a masters thesis from the comparative media studies dept at MIT.
-bloo