The pay as a contractor was quite a bit higher for me than salaried even though I now work for the same company in a similar position (other department though). The rate charged however was a little over double (~$180/hour) yet the place I worked at still had to provide me with office space etc.
I never understood why you wouldn't just hire somebody instead of paying more than double for the same person to come and waste a lot of time because they're not actually working for YOU. They're working for someone else, someone who doesn't have your best interests at heart and they get commissioned on top of that for selling other products to you (ever wondered why those contracting companies are Microsoft Gold Partners and Cisco, Oracle etc. certified if they're just basically an HR replacement?) instead of looking for the best fit for your project.
What you don't understand is how the economy works. If a government invests in the science and technology for this, jobs will be created (in that country) and the valuation of that country will be higher.
Currently, the US is more bent on destroying other countries (effectively reverting education and science into the stone age) than advancing technologically and building an economy. Why do you think China is booming? Not because every business moved there because it was cheap but because the government actively funded schooling, infrastructure, healthcare etc. in order to accommodate businesses to move there. Off course, their ideology might not fit with yours and that might actually be holding them back more than it should but you can't deny that they haven't basically funded their own economy.
Say the US finds a viable, cost-effective way to get into space and travel long distances maybe even start returning resources. Don't you think China, Europe and other countries won't be interested in sending their own missions along.
Some of their projects were simply amateur-level or research-only or trying to replace something that didn't need replacing. ZFS was and is currently being monetized by a host of smaller companies in it's open-source state. Many companies would've loved to pay for specific enhancements to ZFS or support for what was considered the unstable branch but the business-side of Sun was still stuck on the "let's see how we can sell the hardware with it" and even though the hardware was nice, it was overpriced and overqualified for most users. "Supported by Sun" ZFS pools were stuck several version numbers behind their competition. Also "Paid Solaris" was not fully open sourced and the community was developing alternatives to many technologies in Solaris and thus the development was done on OpenSolaris or a derivative but couldn't easily be ported to a paid Solaris.
IBM uses Linux and x86 boxes, not so much proprietary stuff anymore. They have nice hardware but also really good service and r&d into their products which is why people in eg. supercomputing and mainframe replacements choose IBM over cheaper white boxes with the same software.
Google relies on open source. The application is not open source but then again, most business applications aren't. Google didn't invest in Open Source when it came along like the big names like IBM and HP who were selling calculators and typewriters before computers, Google came along because of Open Source - if they would have to pay for every piece of software initially and couldn't modify it they probably wouldn't even have existed.
Back in the day, some Linux distro's included them in the non-free repositories maintained outside the US. Why don't we find a legal loophole like that and just let the customer (who usually doesn't have to pay for the license) decide?
It may be an outrage to you but is it justifiable if you look at the big picture? Sure some may die because of these decisions but how many more die indirectly from producing and releasing these gasses?
Humans deteriorate, a single human doesn't evolve just like a single non-biological chip doesn't evolve into a faster or better thing. The human race does evolve. Chip baking processes don't (usually) deteriorate. It's like saying that the human race is no longer evolving because we're good enough as we are or we haven't seen any major changes in the last 200 years or we can't see what part of us will evolve next. It's a fact of nature that we're going to continue evolving but nobody knows how (yet).
We'll eventually go into quantum processing and there are already techniques being developed like in-chip optical communications and other compounds that are much better at scaling down. The fact that the writer can't see it (or bothers researching it) doesn't mean nobody is working on it.
The 'theoretical limit' has always been 10 years away. We know we are going to get into some issues (such as quantum physics and molecular issues) but they have said the same about SiO2 since the invention of the microprocessor in every magazine I've read (it's too big, it can't be purified enough, the electrical current is going to be too low, we need to start supercooling,...). There is always somebody fixing the issues well in advance in the surrounding fields of chemistry, physics and electronics.
Not really, the several HBA's take forever to boot (press any key within x seconds to get to the BIOS). I could take out their BIOS loaders but then I can't boot from the disks connected to them.
Expensive and hard as far as chip design goes. You need to use transistors, memory and code to do so and as soon as the system is booted in any non-DOS OS (even OS/2), it's usage is limited. You can skip the whole thing saving costs, power and real estate on the chip and in the firmware.
They've killed it by only supporting the features necessary to re-share existing NFS services using SMB and AD. Integration of Windows with non-AD LDAP and Kerberos is virtually non-existent and requires a ton of work and 3rd party utilities to get it working. I don't think NFSv4 is even supported.
How is that comparing apples and oranges? My workstation is supposed to be much faster than my iMac but the BIOS eats up all the boot time. In a server it's because several peripheral BIOS are hooking into the main BIOS and the only way to change settings is manually in the BIOS but with EFI you could re-program your cards while booted and then do a quick reboot when necessary. It just goes to show the absurdity of the BIOS and how it can be done with EFI.
BIOS has a LOT of limitations. >2TB hard drives, network boot, disk controllers, GPU's, IPMI,... everything has to subvert the BIOS in some way which makes it mightily slow. My iMac boots with Lion in 7 seconds. My Linux machine takes 15 seconds just getting to Grub, my servers take up to 45 seconds to get to the boot loader.
BIOS is ALWAYS hooked into 8086 mode (real mode) so at boot time you are limited by it's calls (such as 13h for disks) and that's hard and expensive to emulate on a non-x86 system (such as most Intel/AMD processors).
People use way too simple passwords too. I was recently informed that one of my passwords was too simple (although it's 8 characters long and I thought pretty unique).
They had took their own password database to analyze it (the passwords were encrypted) as well as a multi-lingual dictionary and fed it into a GPU system. In a few seconds they had cracked all the standard dictionary words for about 300,000 passwords. They then went through all the combinations and variations (1337 speak etc.) of their dictionary words (about half a billion variations) and in a few minutes had cracked half of their accounts.
Who does the government hire to build them then? Government contractors and employees are more likely to skip safety procedures (see the BP oil rig disaster and just about any other environmental disaster out there) for profit or out of laziness. They also know they can't get fired or reap any consequences so why would they care?
What we need is companies that want to be liable when stuff goes wrong. Companies need to be liable for their coal plants and liable for their nuclear plants. Right now our taxes through health care are soaking up those costs. For every death or illness caused by the byproducts of the energy production (whether that be smoke, fuel leaks or nuclear radiation), an energy plant needs to pay into the health care system. Also, the cost of energy should be regulated so they can't just raise their prices to offset the cost of their dirtier plants, there needs to be incentive to go to a cheaper (and thus more environmental friendly) solution. You may say you can't just replace a coal plant but you could replace the individual engines one by one with self-contained (container size) reactors - they're not that expensive, some data centers have them as main power and use the grid for backup.
I don't see Bing advertising Google nor Microsoft advertising Linux. It took many, many years and literally millions of dollars in fines for them to simply remove Windows Media Center from EU versions of Windows.
I think Google has explained before how part of their algorithm works - if the site is faster, it's higher ranked. Since Google -> Google crawling is probably in the sub 10ms delay range, it will be higher ranked.
Google does not have a monopoly, get over it already.
It is indeed possible if you have a constant or recurring draw from other sinks (like resistive lights, capacitive motors etc.) but I guess if you have even one of those malfunctioning with a random draw (such as an off-center aquarium pump or an AC unit) or you add signals (like X10 or Ethernet-over-Powerline) that this kind of 'attack' is quickly trumped unless you can get right at the circuit where the TV is on. For that matter, I think an optical attack would be much more reliable (where you measure the light output reflection through someone's windows as they watch TV) but then you might just be find that a spy cam and/or a telelens is much more reliable. If you have those CableCo DVR boxes, they can also be queried remotely.
I don't. I just put on DD-WRT. The fact of the matter is that most of the devices are not built to be at above average household loads (I do have a lot of devices and stream audio through the house) for a long time. The D-Link router wasn't even updated with DD-WRT, it was a stock desktop router with 4 gigabit ports that simply melted because of the gaming and file sharing.
Many recommend putting heat sinks and fans in even if you're not over clocking it. I eventually stopped relying on those routers for all server work (DNS, DHCP) and just put up one of my computers as a permanent router to the Internet.
I used to recommend the Asus RT-N16 and had mine running DD-WRT but after a week out of warranty it up and died (totally dead, no lights, nothing). It seems a common issue with the power supply. Netgear stuff has issues keeping a wireless connection, Linksys would just crash and need a reboot once in a while, one gigabit D-Link router literally melted during a LAN party.
I've had Apple Airport Express, Airport Extreme and Time Capsule and they're all still in working order with uptime's of hundreds of days, my Airport Express attached to a UPS has 700 days of uptime and only goes out in case I'm moving it or extended power outages.
There's a sucker born every minute and most suckers get to management somehow. Those manager will try to cover their asses and thus implement some expensive solution from someone which is promoted in one of those free CIO magazines but in the end does nothing.
Once it's legislated it's usually too late. The law is there and hard (if not impossible) to remove. Those that want these laws are not going to go for the big companies, they're going to go for the small ones that don't have the money to put up a fight and thus have to pay into the racket. Once that happens, they have precedent for ever larger companies and eventually individuals.
Hire somebody to infiltrate the lobbyists for those laws offices. Have them download your company's stuff which you do not license to them and report it. Do the same for any politician that voted this law into office.
Some companies never learn. You can't overdo it. Look at Apple - they had the chance to do everything on the Internet and in the beginning, iPhone apps were ONLY allowed on the Internet and have a local HTML5 storage unit. IT DIDN'T WORK. We can't and don't want to be permanently connected to the Internet. It's also the reason Google Apps is not taking off. We want to be able to continue working if we're somewhere else - in the middle of the woods even.
I do appreciate the efforts Apple and others (like some OpenOffice plugins) is making to make sharing and publishing materials easier, to synchronize faster and easier, to always have your music library or your photo's or your bookmarks available wherever you are in the world or when your house burns down or when your computer crashes. But I also would like to keep them on my hard drive so I can look at them and listen to music when I'm not connected.
I WANT: my data in online storage for free or cheap as a backup, remote login and synchronization tool I DO NOT WANT: my productivity to be affected when the service I WANT is not available or is slowed down for whatever reason
4000 pounds is like what, $6000? Not much you can get with that. Do you really want to build a cluster? As other said, you can rent a cluster but HPC in science is hopelessly inefficient and comes in with large datasets (bandwidth is usually metered) and spits out even larger datasets (bandwidth is metered + online storage is metered) and you'll be paying through the nose because somebody (the researcher) doesn't know how to program correctly. First of all, I would definitely recommend you look around in your institution or in peer institutions whether or not there is already a cluster you can rent (or use for free). Most large institutions have a supercomputer and even smaller departments (like Physics, Astronomy and Imaging or Visual Sciences) have their own small cluster that is not 100% used. You may need to work within the framework of your local politics about that and make concessions as far as time allotments and constraints go but it's cheaper (or free).
If you want to go the DIY route, why don't you just buy a machine from Supermicro (their 3U's are both towers and rack mount) and fill it with a good amount (at least 2GB/core) RAM and 2-4 processors (eg. 6-core or 8-core AMD Opterons), a couple of 2TB hard drives (mirror) and you're pretty much through your budget especially if you want to throw in an nVidia Tesla card. If you want to, you could use virtual machines on those system with Xen or VirtualBox (whichever fits better)
The advantages of this approach is this: - Much less maintenance even if you go the virtual machine route - Much faster interprocess and device (storage etc.) communications - interprocess communications over gigabit kills performance unnecessarily on small cluster. Larger clusters have InfiniBand. - You can still expand it later with another machine like it and use cluster software then. - Footprint is smaller. You can fit these machines in 3 or 4U and they come with a 1.5kW power supply (to support the GPU's). You can buy about 6 1U systems for roughly the same price but you've doubled your rack usage and the power supplies are together about 3 kW because now you've got to power the motherboards and peripherals of 6 devices. - GPU computing cannot be done in cheap 1U devices. And even if you can (and spend a little more and get only 4 machine), only 1 fits, rarely 2. The 3U solutions fit 4 (sometimes 5) GPU's perfectly and are built (power supply, cooling) for it. Even if you don't do GPU computing right now, even MATLAB can offload certain functions to gaming GPU's. They have a little less memory than the Tesla's but they only cost $150-250 (compared to the $700-1200 for a Tesla (EDU discount)) - No need to maintain cluster software (and it can be a pain in the neck) - It fits under a desk and doesn't take up rack space if you don't want it to. No need to pay for hosting or cooling, no extra noise.
As far as distro goes, usually a Red Hat is chosen if you want to pay the money for support, Debian because it's stable or Scientific Linux because it has a certain packet you need, Gentoo or something similar if you need to run a very specific application and need the bottom of the can for performance. Ask your users what they're using and go with the platform that has the widest support.
As far as platform, that's harder to answer. Are they batch jobs - SGE (Sun Grid Engine) is very good and very extensive but I'm unsure what the status is as far as Oracle. Torque and Maui are also widely used. Does it need to look like a single machine then openMosix (or whatever it's called these days) or OpenSSI.
Networking is hard to answer. They both have their up and downs and really depends on the use cases. I would recommend InfiniBand if you can bear the costs but a well-implemented 10GbE will do very well too depending on the jobs (is there a lot of node-to-node communications, then go with IB, an existing or cheaper cluster can more easily be extended with Ethernet).
GPU - hands down nVidia. Sorry ATi/AMD but most researchers use CUDA and both of them do OpenCL. There are also 1U GPU cards from nVidia that are widely supported. ATi on Linux is still unstable, even for compute (although my experience is purely anecdotal with a small cluster). Do you have to buy Tesla cards? Not necessarily. Maybe you should have some for the amount of memory they have but the gaming gear will do just as well with properly programmed or small jobs - also, all GPU memory is mapped into RAM so make sure if you have 3GB on your card you have at least 3GB + however much your base configuration is.
Slashdot Mirror
The pay as a contractor was quite a bit higher for me than salaried even though I now work for the same company in a similar position (other department though). The rate charged however was a little over double (~$180/hour) yet the place I worked at still had to provide me with office space etc.
I never understood why you wouldn't just hire somebody instead of paying more than double for the same person to come and waste a lot of time because they're not actually working for YOU. They're working for someone else, someone who doesn't have your best interests at heart and they get commissioned on top of that for selling other products to you (ever wondered why those contracting companies are Microsoft Gold Partners and Cisco, Oracle etc. certified if they're just basically an HR replacement?) instead of looking for the best fit for your project.
What you don't understand is how the economy works. If a government invests in the science and technology for this, jobs will be created (in that country) and the valuation of that country will be higher.
Currently, the US is more bent on destroying other countries (effectively reverting education and science into the stone age) than advancing technologically and building an economy. Why do you think China is booming? Not because every business moved there because it was cheap but because the government actively funded schooling, infrastructure, healthcare etc. in order to accommodate businesses to move there. Off course, their ideology might not fit with yours and that might actually be holding them back more than it should but you can't deny that they haven't basically funded their own economy.
Say the US finds a viable, cost-effective way to get into space and travel long distances maybe even start returning resources. Don't you think China, Europe and other countries won't be interested in sending their own missions along.
Some of their projects were simply amateur-level or research-only or trying to replace something that didn't need replacing. ZFS was and is currently being monetized by a host of smaller companies in it's open-source state. Many companies would've loved to pay for specific enhancements to ZFS or support for what was considered the unstable branch but the business-side of Sun was still stuck on the "let's see how we can sell the hardware with it" and even though the hardware was nice, it was overpriced and overqualified for most users. "Supported by Sun" ZFS pools were stuck several version numbers behind their competition. Also "Paid Solaris" was not fully open sourced and the community was developing alternatives to many technologies in Solaris and thus the development was done on OpenSolaris or a derivative but couldn't easily be ported to a paid Solaris.
IBM uses Linux and x86 boxes, not so much proprietary stuff anymore. They have nice hardware but also really good service and r&d into their products which is why people in eg. supercomputing and mainframe replacements choose IBM over cheaper white boxes with the same software.
Google relies on open source. The application is not open source but then again, most business applications aren't. Google didn't invest in Open Source when it came along like the big names like IBM and HP who were selling calculators and typewriters before computers, Google came along because of Open Source - if they would have to pay for every piece of software initially and couldn't modify it they probably wouldn't even have existed.
Back in the day, some Linux distro's included them in the non-free repositories maintained outside the US. Why don't we find a legal loophole like that and just let the customer (who usually doesn't have to pay for the license) decide?
It may be an outrage to you but is it justifiable if you look at the big picture? Sure some may die because of these decisions but how many more die indirectly from producing and releasing these gasses?
Humans deteriorate, a single human doesn't evolve just like a single non-biological chip doesn't evolve into a faster or better thing. The human race does evolve. Chip baking processes don't (usually) deteriorate. It's like saying that the human race is no longer evolving because we're good enough as we are or we haven't seen any major changes in the last 200 years or we can't see what part of us will evolve next. It's a fact of nature that we're going to continue evolving but nobody knows how (yet).
We'll eventually go into quantum processing and there are already techniques being developed like in-chip optical communications and other compounds that are much better at scaling down. The fact that the writer can't see it (or bothers researching it) doesn't mean nobody is working on it.
The 'theoretical limit' has always been 10 years away. We know we are going to get into some issues (such as quantum physics and molecular issues) but they have said the same about SiO2 since the invention of the microprocessor in every magazine I've read (it's too big, it can't be purified enough, the electrical current is going to be too low, we need to start supercooling, ...). There is always somebody fixing the issues well in advance in the surrounding fields of chemistry, physics and electronics.
Not really, the several HBA's take forever to boot (press any key within x seconds to get to the BIOS). I could take out their BIOS loaders but then I can't boot from the disks connected to them.
Expensive and hard as far as chip design goes. You need to use transistors, memory and code to do so and as soon as the system is booted in any non-DOS OS (even OS/2), it's usage is limited. You can skip the whole thing saving costs, power and real estate on the chip and in the firmware.
Not anymore. The extended instruction set is more RISC than anything else. It's effectively an x86 emulator running on top of a RISC architecture.
They've killed it by only supporting the features necessary to re-share existing NFS services using SMB and AD. Integration of Windows with non-AD LDAP and Kerberos is virtually non-existent and requires a ton of work and 3rd party utilities to get it working. I don't think NFSv4 is even supported.
How is that comparing apples and oranges? My workstation is supposed to be much faster than my iMac but the BIOS eats up all the boot time. In a server it's because several peripheral BIOS are hooking into the main BIOS and the only way to change settings is manually in the BIOS but with EFI you could re-program your cards while booted and then do a quick reboot when necessary. It just goes to show the absurdity of the BIOS and how it can be done with EFI.
BIOS has a LOT of limitations. >2TB hard drives, network boot, disk controllers, GPU's, IPMI, ... everything has to subvert the BIOS in some way which makes it mightily slow. My iMac boots with Lion in 7 seconds. My Linux machine takes 15 seconds just getting to Grub, my servers take up to 45 seconds to get to the boot loader.
BIOS is ALWAYS hooked into 8086 mode (real mode) so at boot time you are limited by it's calls (such as 13h for disks) and that's hard and expensive to emulate on a non-x86 system (such as most Intel/AMD processors).
People use way too simple passwords too. I was recently informed that one of my passwords was too simple (although it's 8 characters long and I thought pretty unique).
They had took their own password database to analyze it (the passwords were encrypted) as well as a multi-lingual dictionary and fed it into a GPU system. In a few seconds they had cracked all the standard dictionary words for about 300,000 passwords. They then went through all the combinations and variations (1337 speak etc.) of their dictionary words (about half a billion variations) and in a few minutes had cracked half of their accounts.
Who does the government hire to build them then? Government contractors and employees are more likely to skip safety procedures (see the BP oil rig disaster and just about any other environmental disaster out there) for profit or out of laziness. They also know they can't get fired or reap any consequences so why would they care?
What we need is companies that want to be liable when stuff goes wrong. Companies need to be liable for their coal plants and liable for their nuclear plants. Right now our taxes through health care are soaking up those costs. For every death or illness caused by the byproducts of the energy production (whether that be smoke, fuel leaks or nuclear radiation), an energy plant needs to pay into the health care system. Also, the cost of energy should be regulated so they can't just raise their prices to offset the cost of their dirtier plants, there needs to be incentive to go to a cheaper (and thus more environmental friendly) solution. You may say you can't just replace a coal plant but you could replace the individual engines one by one with self-contained (container size) reactors - they're not that expensive, some data centers have them as main power and use the grid for backup.
I don't see Bing advertising Google nor Microsoft advertising Linux. It took many, many years and literally millions of dollars in fines for them to simply remove Windows Media Center from EU versions of Windows.
I think Google has explained before how part of their algorithm works - if the site is faster, it's higher ranked. Since Google -> Google crawling is probably in the sub 10ms delay range, it will be higher ranked.
Google does not have a monopoly, get over it already.
It is indeed possible if you have a constant or recurring draw from other sinks (like resistive lights, capacitive motors etc.) but I guess if you have even one of those malfunctioning with a random draw (such as an off-center aquarium pump or an AC unit) or you add signals (like X10 or Ethernet-over-Powerline) that this kind of 'attack' is quickly trumped unless you can get right at the circuit where the TV is on. For that matter, I think an optical attack would be much more reliable (where you measure the light output reflection through someone's windows as they watch TV) but then you might just be find that a spy cam and/or a telelens is much more reliable. If you have those CableCo DVR boxes, they can also be queried remotely.
I don't. I just put on DD-WRT. The fact of the matter is that most of the devices are not built to be at above average household loads (I do have a lot of devices and stream audio through the house) for a long time. The D-Link router wasn't even updated with DD-WRT, it was a stock desktop router with 4 gigabit ports that simply melted because of the gaming and file sharing.
Many recommend putting heat sinks and fans in even if you're not over clocking it. I eventually stopped relying on those routers for all server work (DNS, DHCP) and just put up one of my computers as a permanent router to the Internet.
I used to recommend the Asus RT-N16 and had mine running DD-WRT but after a week out of warranty it up and died (totally dead, no lights, nothing). It seems a common issue with the power supply. Netgear stuff has issues keeping a wireless connection, Linksys would just crash and need a reboot once in a while, one gigabit D-Link router literally melted during a LAN party.
I've had Apple Airport Express, Airport Extreme and Time Capsule and they're all still in working order with uptime's of hundreds of days, my Airport Express attached to a UPS has 700 days of uptime and only goes out in case I'm moving it or extended power outages.
There's a sucker born every minute and most suckers get to management somehow. Those manager will try to cover their asses and thus implement some expensive solution from someone which is promoted in one of those free CIO magazines but in the end does nothing.
Once it's legislated it's usually too late. The law is there and hard (if not impossible) to remove. Those that want these laws are not going to go for the big companies, they're going to go for the small ones that don't have the money to put up a fight and thus have to pay into the racket. Once that happens, they have precedent for ever larger companies and eventually individuals.
These days those arrangements are thwarted by Facebook and Twitter.
Hire somebody to infiltrate the lobbyists for those laws offices. Have them download your company's stuff which you do not license to them and report it. Do the same for any politician that voted this law into office.
Some companies never learn. You can't overdo it. Look at Apple - they had the chance to do everything on the Internet and in the beginning, iPhone apps were ONLY allowed on the Internet and have a local HTML5 storage unit. IT DIDN'T WORK. We can't and don't want to be permanently connected to the Internet. It's also the reason Google Apps is not taking off. We want to be able to continue working if we're somewhere else - in the middle of the woods even.
I do appreciate the efforts Apple and others (like some OpenOffice plugins) is making to make sharing and publishing materials easier, to synchronize faster and easier, to always have your music library or your photo's or your bookmarks available wherever you are in the world or when your house burns down or when your computer crashes. But I also would like to keep them on my hard drive so I can look at them and listen to music when I'm not connected.
I WANT: my data in online storage for free or cheap as a backup, remote login and synchronization tool
I DO NOT WANT: my productivity to be affected when the service I WANT is not available or is slowed down for whatever reason
4000 pounds is like what, $6000? Not much you can get with that. Do you really want to build a cluster? As other said, you can rent a cluster but HPC in science is hopelessly inefficient and comes in with large datasets (bandwidth is usually metered) and spits out even larger datasets (bandwidth is metered + online storage is metered) and you'll be paying through the nose because somebody (the researcher) doesn't know how to program correctly. First of all, I would definitely recommend you look around in your institution or in peer institutions whether or not there is already a cluster you can rent (or use for free). Most large institutions have a supercomputer and even smaller departments (like Physics, Astronomy and Imaging or Visual Sciences) have their own small cluster that is not 100% used. You may need to work within the framework of your local politics about that and make concessions as far as time allotments and constraints go but it's cheaper (or free).
If you want to go the DIY route, why don't you just buy a machine from Supermicro (their 3U's are both towers and rack mount) and fill it with a good amount (at least 2GB/core) RAM and 2-4 processors (eg. 6-core or 8-core AMD Opterons), a couple of 2TB hard drives (mirror) and you're pretty much through your budget especially if you want to throw in an nVidia Tesla card. If you want to, you could use virtual machines on those system with Xen or VirtualBox (whichever fits better)
The advantages of this approach is this:
- Much less maintenance even if you go the virtual machine route
- Much faster interprocess and device (storage etc.) communications - interprocess communications over gigabit kills performance unnecessarily on small cluster. Larger clusters have InfiniBand.
- You can still expand it later with another machine like it and use cluster software then.
- Footprint is smaller. You can fit these machines in 3 or 4U and they come with a 1.5kW power supply (to support the GPU's). You can buy about 6 1U systems for roughly the same price but you've doubled your rack usage and the power supplies are together about 3 kW because now you've got to power the motherboards and peripherals of 6 devices.
- GPU computing cannot be done in cheap 1U devices. And even if you can (and spend a little more and get only 4 machine), only 1 fits, rarely 2. The 3U solutions fit 4 (sometimes 5) GPU's perfectly and are built (power supply, cooling) for it. Even if you don't do GPU computing right now, even MATLAB can offload certain functions to gaming GPU's. They have a little less memory than the Tesla's but they only cost $150-250 (compared to the $700-1200 for a Tesla (EDU discount))
- No need to maintain cluster software (and it can be a pain in the neck)
- It fits under a desk and doesn't take up rack space if you don't want it to. No need to pay for hosting or cooling, no extra noise.
As far as distro goes, usually a Red Hat is chosen if you want to pay the money for support, Debian because it's stable or Scientific Linux because it has a certain packet you need, Gentoo or something similar if you need to run a very specific application and need the bottom of the can for performance. Ask your users what they're using and go with the platform that has the widest support.
As far as platform, that's harder to answer. Are they batch jobs - SGE (Sun Grid Engine) is very good and very extensive but I'm unsure what the status is as far as Oracle. Torque and Maui are also widely used. Does it need to look like a single machine then openMosix (or whatever it's called these days) or OpenSSI.
Networking is hard to answer. They both have their up and downs and really depends on the use cases. I would recommend InfiniBand if you can bear the costs but a well-implemented 10GbE will do very well too depending on the jobs (is there a lot of node-to-node communications, then go with IB, an existing or cheaper cluster can more easily be extended with Ethernet).
GPU - hands down nVidia. Sorry ATi/AMD but most researchers use CUDA and both of them do OpenCL. There are also 1U GPU cards from nVidia that are widely supported. ATi on Linux is still unstable, even for compute (although my experience is purely anecdotal with a small cluster). Do you have to buy Tesla cards? Not necessarily. Maybe you should have some for the amount of memory they have but the gaming gear will do just as well with properly programmed or small jobs - also, all GPU memory is mapped into RAM so make sure if you have 3GB on your card you have at least 3GB + however much your base configuration is.