What irritates me is when amazon do it. On my own book. There is a sponsored link on my book pointing people at the TCO comparisons -the one we know were so unbiased.
Can I get the links taken down? nope, they pay, they get. All I can do is make sure the next edition has no support for Windows whatsoever.
HP Labs. On Page Mill, East of El Camino. HP's "building 1", and still home to the original H and P offices (not occupied by carly, thank you very much)
Go in the entrance , get past security (you need someone with HP ID with you), turn left, turn left again to the offices, go along to the end of the building, ask around. Then get into his office and ask to see some interesting demos. He does fun things with Bluetooth too:)
the $70 was an anecdotal quote by someone buying 800 of them. I think in that volume you get discount.
Even so, maybe the manager making the quote screwed up; it wouldnt surprise me.
I do like your solution though. I'm involved a lot of grid work and we really need a decent distributed FS there, even for the read-only stuff. It's no good being able to get CPU cycles when all your data is inaccessible.
yes, cost/node becomes everything at volume. I mean, fibre-channel? $70/node -> 35K for 500 of them. Now add windows at $400/node -its just too expensive.
the other is the maintenance costs; if you have to do anything by hand to each node, you cannot admin the full site economically.
Good point about feedback -it's like the publishing industry - Amazon's sales ranking is the only sales metric that takes less than three months round trip. Information is power.
Regarding PC line R&D, laptops are still hard. They are the last bastion of high-QA client side systems. Get something wrong with the HDD mount and your AFR (annualized failure rate) goes from 20% to 60%, and there go your profits and customer happiness. Servers are the other area. But even there, Intel likes to help. Indeed, one way to view Dell is the execution arm of Intel's R&D.
when it comes to cost model, the big diffs are %age R&D spend, and time between parts are paid for and the vendor gets paid. Dell have your credit card cashed before they pay for their parts, whereas HP (and anyone else who sells indirect) has to pay the vendors, have the stock and not reap the benefits of price depreciation.
So to win against Dell its hard -you need to go to direct manufacture, if not direct sales. As well as kill all interesting R&D and instead focus on short term deliverables. If HP does have to go that way, it will hurt the company long term. But perhaps that is what they will have to do to survive at all. Going out of business hurts even more, long term:)
they are very disconcerting to encounter in the mountains. I was driving up in scotland when this hercules crawled overhead at about 100' - it is a lot harder to avoid reacting badly to the sight of something that looks about to land on you, than it is it deal with a fighter plane going above you at that height -they are usually gone before you have time to notice.
yeah, I wonder how it handles hyperthreaded cores too? Still, windows has some atrocious CPU crippling built in too.
I am still not sure about per-machine subscriptions. I know I could get my employer to pay for one for my desktop, but my laptop that dual boots into WinXP and Linux? harder to justify. And the two home boxes are off their radar, so I would lose consistency across platforms. I am currently thinking retail Suse9.1 is for me.
(NB, for winXP I do get support in exchange for running the Beta of SP2. you get your bugs dealt with by somebody compenent then, though not competent enough to deal with laptop-hibernate failures when running a VPN over the WLAN; there are some things no OS vendor can handle)
Yes, a good load balancer is better than anything, as it looks at the health of the entire system before balancing stuff.
A bad one is worse; I had one system where one box wasnt working right; POST requests to render something were failing in 1/10 of a second, instead of rendering in 30s. So the load balancer, dumb little fuck that it was, routed more stuff that way, because its queue was faster.
moral: you need to integrate health into load balance. We ship such a health check into apache axis to make it easy -look for happyaxis.jsp on google, but not enough people use it, I fear.
Yes, but what if a single supercomputer isnt enough?
The interesting work in performance computing is very grid-centric these days, because even, say, CERN's cray room isnt up to handling the amount of data the Large Hadron Collider is going to be generating. Instead the data is going to have to be handed off to umpteen nodes out there to filter it, filter it, filter it and highlight interesting stuff.
The Grid is still a long way from being perfect (very long way), though we are getting there (I work on deployment problems). But I can see a time where it is the right architecture for your work because the data sharing and management is up there with the raw cycles.
That doesnt mean it wont be nice to have a Cray or NEC supercomputer (the latter are really into the Grid, BTW) in your fabric, but you'll be using all those idle AMD boxes around the labs too.
Or, would like to buy one very big tractor now, or slowly buy little tractors at a cost per tractor that decreases at 5% a month.
Of course, after 6 months the little tractors you were buying go off the price list and you end up with a heterogenous mess, but still, scaleable addition of new kit is where clustering wins over mainframs and supercomputers.
i.e. What if you needed slightly more than one big tractor?
One specific optimisation is that on a PIII core and Java1.4.1 or later, the JIT compiler generates SSE instruction codes for FP math, rather than using the x86 part. So the floating point performance is better than what you'd get if you just compiled a single binary for one x86 platform and let people run it.
So doing CPU intensive algorithms in Java does make sense. GUIs, that is a different matter, but things are improving there too:)
We get the 'never' option in WinXP SP2, along with a firewall that finally whines on outbound calls. The latter could raise more support calls than IT depts want at first, but there you go.
However, SP2 still doesnt include a virus scanner or an ad-aware-like app scanner, so if your system is already infested with something, tough.
I have lost the link; it was somewhere on the support page. The problem was in the fancy new booter GUI. Turn that off and it was fixed, but it worried me.
As a Java developer on some Apache projects, I'm not yet rushing out to boycott Java. I'll just deny Sun any of my money. Oh, wait a minute, I wasn't giving them any anyway.
How about I let Sun continue to lose money providing and supporting Java? That would be even worse than boycotting it, wouldnt it?
Also, beware of IBM bearing gifts. They are supporting OSS right now because they see the strategic value. But if their opinions change, watch them change their mind. It's like Oracle: they support linux as a way to keep OS costs down, but are against OSS database solutions. If (when) an OSS database gets to the point that it threatens Oracle or DB2, I could imagine both companies changing their stories about the value of OSS.
I'd argue something different: RedHat's current subscription-only policy is going to kill it if they keep it up.
If the current pricing model of RedHat is relaly such that Solaris or Windows costs *less* that a RedHat sub, then it is in trouble.
I know I will get pointed at fedora, but given that fedora kills the network on a any laptop whose network is on a PC-card, isnt supported by all those binary things I use (nvida, vmware), I dont view it as a broadly tested or stable enough solution to work with.
Sun are just going to put pricing pressure on RedHat; the real test is what penetration Novell/SuSE linux has. If a version backed by the suits gets picked up by the PC vendors and sold mainstream, it can put serious pressure on RedHat's position as "the" north american commercial Linux distro.
Look at the effective encouragement for diesel cars -protects the french car industry against the japanese.
Look at the Common Agricultural Policy, that preserves rural peasant life at the expense of the rest of the world.
The EU is pretty blatantly self-serving at times. But it has embraced the Kyoto agreement, and it is not afraid of slapping around companies that overstep the mark. And it monitors those decisions pretty well.
Maybe it is because there is no president; to buy the EU you have to subvert a majority of countries, and without a big MS presence, none of them can be swayed by the MS creates Jobs story.
Mandrake 10.0 seems to work fairly well on current HP kit,
RH9.0 was troublesome but that is primarily because it was not yet mobile. To get the WLAN working on an omnibook I had to get one of the linux WLAN contributors to spend half an hour fiddling with the configuration settings. Still, I have spent more time than that with regedit on windows.
Can I observe I am one of those people who is (currently) paid to work on an OSS project, a distributed deployment framework called SmartFrog.
I am doing this as part of my day job as a Computing Researcher at HPLabs, an organisation which hasnt (yet) suffered the knife of death in it, but still has to deal with a slow death of many cuts.
Why are we doing OSS work? Because (a) we know we can get something back from the community (academic as well as pure OSS), (b) its what users expect, and (c) as we dont have a software business any more, how else can you ship it.
OSS makes a great platform for doing CS research; I know lots of people who use it. Take Jean "Linux WLAN" Tourhilles, or David Mossberger, one of the leads on the IA64 ports of Linux.
Its good for research as in closed source, MS have such a monopoly that you cannot innovate, and even if you get access to Win2K source, you cannot share your works with others. Research and OSS goes hand in hand, be it industrial or academic. Or personal: there is nothing to stop the reader doing innovative stuff at home, be it community WLAN frameworks, new thread scheduling algorithms or better XML parsers.
Yes, COM stuff that you want under HKEY_CLASSES_ROOT must be admin writeable; which means DX and lots of other stuff. In theory WinXP has better side-by-side stuff, but I've never seen it work properly.
Another one is that there is no isolation between app components. You cannot have two versions of the same app in different file systems, with different registry keys.
the whole thing is built on the premise that that there is one version of MS office; each user has the right to change some options on that version, but not all.
Also, because write access to bits of the registry are restricted to admins, you cannot install anything complex (like a game) without admin access. Which is just silly.
Remember that an HP PC executive was the first person quoted in the MS antitrust findings, "if we had a choice, you'd be second".
There is no love lost between parts of the company, especially the original HP. Compaq, on the other hand, have thrived for years by sucking up to the man, and been very good at it. Yet if you look at the workstation and server lines, they have been certified for linux distros for a while (usually redhat 'premium' stuff), and been orderable with the OS. No retributions yet:)
The biggest risk with MS is that they will cut the company out of some big special, like a new product, like getting so many people on longhorn beta test, etc. They would probably do that today except that MS know they dont have a choice. The HP/Compaq PC line is a big enough chunk of sales that they dont dare walk away.
At the same time, I can imagine a lot of high level voicemails going back and forth:)
too damn right they will. There is some deep irony that this got launched at CES, the same place that carly fiorina got on stage with the RIAA people - I note she didnt announce 'by the way, we will soon be introducing CD burners that can label disks like retail CDs'.
Slashdot Mirror
What irritates me is when amazon do it. On my own book. There is a sponsored link on my book pointing people at the TCO comparisons -the one we know were so unbiased.
Can I get the links taken down? nope, they pay, they get. All I can do is make sure the next edition has no support for Windows whatsoever.
HP Labs. On Page Mill, East of El Camino. HP's "building 1", and still home to the original H and P offices (not occupied by carly, thank you very much)
:)
Go in the entrance , get past security (you need someone with HP ID with you), turn left, turn left again to the offices, go along to the end of the building, ask around. Then get into his office and ask to see some interesting demos. He does fun things with Bluetooth too
the $70 was an anecdotal quote by someone buying 800 of them. I think in that volume you get discount.
Even so, maybe the manager making the quote screwed up; it wouldnt surprise me.
I do like your solution though. I'm involved a lot of grid work and we really need a decent distributed FS there, even for the read-only stuff. It's no good being able to get CPU cycles when all your data is inaccessible.
Jean does work in Palo Alto, BTW.
I only got wireless working on RH9 on my laptop (w/USB 802.11b) by taking it to his cube and refusing to leave till he sorted it out.
yes, cost/node becomes everything at volume. I mean, fibre-channel? $70/node -> 35K for 500 of them. Now add windows at $400/node -its just too expensive.
the other is the maintenance costs; if you have to do anything by hand to each node, you cannot admin the full site economically.
Good point about feedback -it's like the publishing industry - Amazon's sales ranking is the only sales metric that takes less than three months round trip. Information is power.
Regarding PC line R&D, laptops are still hard. They are the last bastion of high-QA client side systems. Get something wrong with the HDD mount and your AFR (annualized failure rate) goes from 20% to 60%, and there go your profits and customer happiness. Servers are the other area. But even there, Intel likes to help. Indeed, one way to view Dell is the execution arm of Intel's R&D.
when it comes to cost model, the big diffs are %age R&D spend, and time between parts are paid for and the vendor gets paid. Dell have your credit card cashed before they pay for their parts, whereas HP (and anyone else who sells indirect) has to pay the vendors, have the stock and not reap the benefits of price depreciation.
:)
So to win against Dell its hard -you need to go to direct manufacture, if not direct sales. As well as kill all interesting R&D and instead focus on short term deliverables. If HP does have to go that way, it will hurt the company long term. But perhaps that is what they will have to do to survive at all. Going out of business hurts even more, long term
they are very disconcerting to encounter in the mountains. I was driving up in scotland when this hercules crawled overhead at about 100' - it is a lot harder to avoid reacting badly to the sight of something that looks about to land on you, than it is it deal with a fighter plane going above you at that height -they are usually gone before you have time to notice.
yeah, I wonder how it handles hyperthreaded cores too? Still, windows has some atrocious CPU crippling built in too.
I am still not sure about per-machine subscriptions. I know I could get my employer to pay for one for my desktop, but my laptop that dual boots into WinXP and Linux? harder to justify. And the two home boxes are off their radar, so I would lose consistency across platforms. I am currently thinking retail Suse9.1 is for me.
(NB, for winXP I do get support in exchange for running the Beta of SP2. you get your bugs dealt with by somebody compenent then, though not competent enough to deal with laptop-hibernate failures when running a VPN over the WLAN; there are some things no OS vendor can handle)
yes, but its pricing model also puts the OSS community off it. $300/year? that is more than Solaris, or WinXP.
I do need ISV support for one thing (VMware), but I am planning on making SuSE9.1 my reference platform for ISV supported software.
Yes, a good load balancer is better than anything, as it looks at the health of the entire system before balancing stuff.
A bad one is worse; I had one system where one box wasnt working right; POST requests to render something were failing in 1/10 of a second, instead of rendering in 30s. So the load balancer, dumb little fuck that it was, routed more stuff that way, because its queue was faster.
moral: you need to integrate health into load balance. We ship such a health check into apache axis to make it easy -look for happyaxis.jsp on google, but not enough people use it, I fear.
Yes, but what if a single supercomputer isnt enough?
The interesting work in performance computing is very grid-centric these days, because even, say, CERN's cray room isnt up to handling the amount of data the Large Hadron Collider is going to be generating. Instead the data is going to have to be handed off to umpteen nodes out there to filter it, filter it, filter it and highlight interesting stuff.
The Grid is still a long way from being perfect (very long way), though we are getting there (I work on deployment problems). But I can see a time where it is the right architecture for your work because the data sharing and management is up there with the raw cycles.
That doesnt mean it wont be nice to have a Cray or NEC supercomputer (the latter are really into the Grid, BTW) in your fabric, but you'll be using all those idle AMD boxes around the labs too.
Or, would like to buy one very big tractor now, or slowly buy little tractors at a cost per tractor that decreases at 5% a month.
Of course, after 6 months the little tractors you were buying go off the price list and you end up with a heterogenous mess, but still, scaleable addition of new kit is where clustering wins over mainframs and supercomputers.
i.e. What if you needed slightly more than one big tractor?
One specific optimisation is that on a PIII core and Java1.4.1 or later, the JIT compiler generates SSE instruction codes for FP math, rather than using the x86 part. So the floating point performance is better than what you'd get if you just compiled a single binary for one x86 platform and let people run it.
So doing CPU intensive algorithms in Java does make sense. GUIs, that is a different matter, but things are improving there too:)
We get the 'never' option in WinXP SP2, along with a firewall that finally whines on outbound calls. The latter could raise more support calls than IT depts want at first, but there you go.
However, SP2 still doesnt include a virus scanner or an ad-aware-like app scanner, so if your system is already infested with something, tough.
I have lost the link; it was somewhere on the support page. The problem was in the fancy new booter GUI. Turn that off and it was fixed, but it worried me.
I will check out Whitebox, thanks.
As a Java developer on some Apache projects, I'm not yet rushing out to boycott Java. I'll just deny Sun any of my money. Oh, wait a minute, I wasn't giving them any anyway.
How about I let Sun continue to lose money providing and supporting Java? That would be even worse than boycotting it, wouldnt it?
Also, beware of IBM bearing gifts. They are supporting OSS right now because they see the strategic value. But if their opinions change, watch them change their mind. It's like Oracle: they support linux as a way to keep OS costs down, but are against OSS database solutions. If (when) an OSS database gets to the point that it threatens Oracle or DB2, I could imagine both companies changing their stories about the value of OSS.
I'd argue something different: RedHat's current subscription-only policy is going to kill it if they keep it up.
:)
If the current pricing model of RedHat is relaly such that Solaris or Windows costs *less* that a RedHat sub, then it is in trouble.
I know I will get pointed at fedora, but given that fedora kills the network on a any laptop whose network is on a PC-card, isnt supported by all those binary things I use (nvida, vmware), I dont view it as a broadly tested or stable enough solution to work with.
Sun are just going to put pricing pressure on RedHat; the real test is what penetration Novell/SuSE linux has. If a version backed by the suits gets picked up by the PC vendors and sold mainstream, it can put serious pressure on RedHat's position as "the" north american commercial Linux distro.
Sun are probably still doomed, either way
Look at the effective encouragement for diesel cars -protects the french car industry against the japanese.
Look at the Common Agricultural Policy, that preserves rural peasant life at the expense of the rest of the world.
The EU is pretty blatantly self-serving at times. But it has embraced the Kyoto agreement, and it is not afraid of slapping around companies that overstep the mark. And it monitors those decisions pretty well.
Maybe it is because there is no president; to buy the EU you have to subvert a majority of countries, and without a big MS presence, none of them can be swayed by the MS creates Jobs story.
Mandrake 10.0 seems to work fairly well on current
HP kit,
RH9.0 was troublesome but that is primarily because it was not yet mobile. To get the WLAN working on an omnibook I had to get one of the linux WLAN contributors to spend half an hour fiddling with the configuration settings. Still, I have spent more time than that with regedit on windows.
Can I observe I am one of those people who is (currently) paid to work on an OSS project, a distributed deployment framework called SmartFrog.
I am doing this as part of my day job as a Computing Researcher at HPLabs, an organisation which hasnt (yet) suffered the knife of death in it, but still has to deal with a slow death of many cuts.
Why are we doing OSS work? Because (a) we know we can get something back from the community (academic as well as pure OSS), (b) its what users expect, and (c) as we dont have a software business any more, how else can you ship it.
OSS makes a great platform for doing CS research; I know lots of people who use it. Take Jean "Linux WLAN" Tourhilles, or David Mossberger, one of the leads on the IA64 ports of Linux.
Its good for research as in closed source, MS have such a monopoly that you cannot innovate, and even if you get access to Win2K source, you cannot share your works with others. Research and OSS goes hand in hand, be it industrial or academic. Or personal: there is nothing to stop the reader doing innovative stuff at home, be it community WLAN frameworks, new thread scheduling algorithms or better XML parsers.
Yes, COM stuff that you want under HKEY_CLASSES_ROOT must be admin writeable; which means DX and lots of other stuff. In theory WinXP has better side-by-side stuff, but I've never seen it work properly.
Another one is that there is no isolation between app components. You cannot have two versions of the same app in different file systems, with different registry keys.
the whole thing is built on the premise that that there is one version of MS office; each user has the right to change some options on that version, but not all.
Also, because write access to bits of the registry are restricted to admins, you cannot install anything complex (like a game) without admin access. Which is just silly.
Remember that an HP PC executive was the first person quoted in the MS antitrust findings, "if we had a choice, you'd be second".
:)
:)
There is no love lost between parts of the company, especially the original HP. Compaq, on the other hand, have thrived for years by sucking up to the man, and been very good at it. Yet if you look at the workstation and server lines, they have been certified for linux distros for a while (usually redhat 'premium' stuff), and been orderable with the OS. No retributions yet
The biggest risk with MS is that they will cut the company out of some big special, like a new product, like getting so many people on longhorn beta test, etc. They would probably do that today except that MS know they dont have a choice. The HP/Compaq PC line is a big enough chunk of sales that they dont dare walk away.
At the same time, I can imagine a lot of high level voicemails going back and forth
too damn right they will. There is some deep irony that this got launched at CES, the same place that carly fiorina got on stage with the RIAA people - I note she didnt announce 'by the way, we will soon be introducing CD burners that can label disks like retail CDs'.