If Skype gets anywhere close to mainstream it will be forced to provide mandatory legal intercept without any need for inventing fantom associations. This will be regardless of its use for drug dealing or not.
And here comes the most interesting bit. In order to provide legal intercept capabilities it will have to provide law enforcement access via a remote control interface to computers serving as supernodes in the P2P network. These computers are not even owned by Skype and may be outside the jurisdiction of the party requesting intercept. In fact intercepting on them may be illegal in the country where they are located. This is bound to get very entertaining at some point sooner or later.
And by the way using specially dedicated nodes for legal intercept only will not work because one of the requirements for legal intercept in telephony is that it should not be noticeable to either party in the conversation. A node located in a strange place will very happily show up in the netstat on both Linux and Windows and writing a utility which shows which address block is the supernode you are connected to is a piece of cake.
Scotland laws differ from England and Wales (you gotta love the UK). The actual enforcement framework and guidelines differ even further. Strange as it may seem it is true. For example the UK law is dated 1999 and the Scottish one is dated 2003 and as far as I know they actually differ.
The real problem is with companies who have a less efficient and more expensive product, but buy off possibly competing patent technology and keep it in their safes until doomsday.
Oil companies own patents on battery and hybrid vehicle technologies and keep them in their safes, chemical companies keep patents on long life tires and do not develop them so that they can sell inferior product, so on so fourth.
These are what really stiffles innovation and should be dealt with. The little trolls are a buyproduct of the rules set by the big players for their own benefit. If the patent system is changed to deal with them the little trolls will die off from natural causes.
UK law is wonderfully vague. It says that it is illegal to put a child at risk without providing a clear definition of risk and responsibility.
In practice the local authorities usually follow these guidelines: http://www.nspcc.org.uk/html/home/needadvice/child renathomealone.htm. In a very narrow reading of these guidelines it is illegal to let a child under 13 travel on their own (by extension of the rule that they should not be left on their own). This has been applied this way across the UK multiple times (according to friends of mine who work as teachers).
Is this application right or wrong - IANAL. All I can say is that it errodes the sense of responsibility in the child.
Is it this or mashing biology, chemistry and physics into a half baked mash called "Science"? Or the complete liquidation of any homework and any home assignments in primary school? Or the idiotic laws that force the parents to babysit their offspring till they are 14 years old removing any sense of reason and responsibility? I remember that at the age of 7 I had to travel across one quarter of a 10 million city alone to school every day. And I was not the only one to do so. In fact there was not a single parent dropping off or picking up children after the first week. Frankly, before we get to the crazy frog, shooting all the MP critters who pushed this stupid law followed by a selective school run cull may be a better place to start.
MTBF is MEAN TIME BEFORE FAILURE not Time before guaranteed failure. It is the Expected Value of your failure distribution. By increasing your MTBF, you are decreasing the probability of failure within your estimated installation lifetime. As a result over your projected installation lifetime the overall number of failures and overall maintenance expenditure decreases. When managing 100+ servers it is guaranteed to have some failures due to normal wear and tear even during the warranty period. Not trying to decrease their rate is a prime case of masohism.
As far as your experience is concerned - 5 years is a long shot for the systems that roll off the production line today. In fact very few of them will survive that length of time without power management. Out of the 1U systems I have dealt with lately not a single one. This was not the case with most old servers from 5+ years ago as they were designed for OSes which did not have power management at the time. As a result their cooling system capacity is way over what they actually need. Just opening a Proliant G1, G2 and G3 side by side is a good enough example. The cooling capacity in them does not increase while the heat emission from the CPU increases more then 4 times.
Same here. But they have both ACPI and CPU frequency scaling (aka centrino, powernow, longrun, longhaul) enabled.
The reasons are fairly simple.
Keeping the servers cooler decreases the electricity bill by up to 20% from ACPI and up to 75% from cpufreq for an average Pentium/Xeon based corporate install when idle. Athlon/Opteron numbers are comparable. This also decreases airconditioner wear and tear.
If your servers are well designed and have temperature controlled fans (all IBMs and Compaqs fall in this category) this increases fan MTBF by up to 2 times. Most importantly if a fan fails the server usually is cold enough to send out a clean alert and even work until you get around to replace it.
Your capacitor deterioration rate (they are the first thing to blow in a server nowdays) is considerably less which means that your overall server MTBF goes up and reliability increases considerably
There is only one case where power management is unnecessary. It is if you are running a computational load which requires your servers to work 365x24x7 flat out. Running without ACPI and cpufreq (if supported) for an average corporate or ISP load is plain stupid. It results in a less reliable server installation.
By the way, do not worry, 99.99 of the server sysadmins out there join you in this stupidity. In fact I did 7 years ago when I knew less about server administration.
While I am an old AMD fanboy and given a choice I will always chose AMD Intel must be given credit where credit is due. If you use CPU frequency scaling on Intel which is supported on both Pentium 4 and Xeon you can drop its consumption into the sub-25W territory when idle and ramp it up in a lot of increments (usually 8) to full as the load arrives. While in theory AMD powernow-k8 should give you similar features, in practice I have yet to see an SMP motherboard that does not have it disabled. It is also not supported properly until 2.6.13 which means that it will not work on the stock RedHat kernel.
As a result of this you will get a considerable power consumption improvement when moving to AMD for things that run full blast all the time like computation. You will not see this improvement in a correctly configured Xeon installation running Linux on recent RedHat ES or Debian (or anything with a post 2.6.7 kernel). In fact there Intel will beat AMD flat out.
The important item here is "correctly configured". Based on my not very scientific observations 99.99% of the corporate IT out there is not competent enough to do this. 1 out of 1000 on average knows that you can turn most of the Centrino-like features on a server. The rest have no clue.
You can use most of these features on a desktop or a server if you:
1. Use Linux 2. Bother to turn it on
Just look at the kernel documentation for cpufreq and the on_demand governor. Alternatively you can use cpufreqd which allows even finer tuning.
Turning it on for a dual CPU Xeon will drop the power consumption from 400W+ to under 100W when idle. In fact this feature works on any Pentium4 class CPU.
Numbers for Opterons are similar, but most dual and quad Opteron motherboards lack proper support for this feature.
I actually use them for servers in my day job and in a number of my own projects. So my 0.02 eu
They are nice, but they have their limitations. On the positive side:
Very low power consumption.
Very high IO speed. In fact considerably higher than expected. I have been getting 2+ times faster IO than from an Intel Xeon from them.
AES acceleration on the higher end models, high quality hardware RNG and RSA acceleration on the models coming up this year
On the negative side:
Very small cache. Much smaller than anything else on the market. As a result it is blazing fast to start and blazing fast on any small data set, but churning large data slows it down considerably. This also causes task switching under higher loads to be considerably slower than expected.
The default factory fans on the higher end models are not temperature controlled. They have an MTBF of around a year and a half which is too low by modern Datacenter standards. On the positive side, the thermal throttle on them is good enough to allow them to run with the fan failed.
Overall they are good enough to run a mail relay, a DNS server a firewall, a QoS conditioner or an Odds-N-Sods. They are definitely slower then I would like to for a news server. CGI performance and MySQL are abissmal and computation or compilation are in the "you might as well forget it" area.
The most painful part is actually the relatively low number of people providing good server class enclosures for them. This http://www.icp-epia.co.uk/ is one of the few sites. Even they do not have everything so you end up assembling most of them yourself which costs time and money.
This will work for an office. It will not work for a datacenter which is the original question. The heat density in a datacenter is quite often in the 16KW per rack range for peak consumption. Precooling this will give at most 5 minutes. No point to bother
Why only peak load reduction? Load reduction in terms of electricity is the actual way to start.
A good example - a Dual CPU Pentium4 Xeon Intel OEM system in a standard Intel OEM chassis eats nearly 400W when idle with no power management. With standard ACPI power management when idle it eats 350-380W. With CPU frequency scaling using the ondemand CPU governor it will eat less then 100W when idle. The numbers for Opteron based systems are not much different. A usual datacenter is designed to cope with full demand after losing up to 25% loss in terms of CPU capacity due to failures. This across a datacenter means a saving of up to 3KW per rack at peal capacity and 9KW+ when idle for fully loaded 42U racks. It also means increasing MTBF on fans by up to 2 times, increasing MTBF on the airconditioner by up to 4 times, improving disk MTBF by 20-30% due to lower heat, so on so fourth.
Now, here is the 1 million KWh question:
How many server sysadmins turn this on? How many server sysadmins actually check server systems during evaluation for working CPU frequency scaling and working power management? How many server sysadmins actually take this into account during capacity planning? As a matter of fact how many perform thermal correction for MTBF when doing datacenter and system design?
Looking around I can say that in the UK it is somewhere between 0 and 0.01%. Dunno about other countries.
The diagram you are pointing to is a lie. If it was true the BES would have been talking to handsets directly and it does not. It talks to something in RIM. That something talks to the provider GPRS network and via it to the handheld.
This is what I am referring to.
No-one knows what happens there. There is no information in the press and all marketing literature is full of lies. All of it carefully omits this RIM side component. Personally, I seriously doubt that they have the computing power in the older handhelds to do 3DES or AES on all messages (as they claim). If they did why did they implement both x509/SMIME and PGP at the server and not at the handheld defeating the entire idea of end-to-end encryption?
You are showing all signs of addiction including detachment from reality and assigning to your addictive substance (crackberry) mythical powers which it does not posses. Encryption is clearly not one of them.
If I recall correctly the Crackberry is the only wireless email system where the PGP and x509 integration are solely at the server. The messages are decrypted at the server and sent to the device unencrypted perusing only the GPRS wireless encryption which is pretty weak by modern standards. They are temporarily stored at the device unencrypted as well, which defeats the idea of PGP and x509 completely.
Not a dupe, but I do not quite understand the article classification.
First of all, cudos to the NetBSD crowd for maintaining an OS that can run even on a Dead Marmot. Over the years, I have run it on several Dead Marmots (TM) like MIPS 3000 DecStations (should not be mistaken with Dead Badgers which can run Linux).
What I do not understand is the article classification. Surely, the more popular of the BSDs should have gotten prime time coverage and the less popular coverage between headlines. What's the deal here? Or the FreeBSD people should port their OS to a Dead Lemming instead of dropping the support for anything pre-80486 alltogether?
Yes you can. It was most likely compiled versus Blow^H^H^H^H^MoTiff. That is at best. At worst versus Sun's old stuff (so old now, that forgot what it was called...). In order for it to be useable on a modern Linux or BSD desktop it will have to be ported to a newer graphic toolkit and widget set. Which brings us back to square one.
As a reference model on how things are supposed to be done by the book with professionally commented and written source and properly written documentation.
What does it do that Linux doesn't?
Correct integration of statefull firewall and QoS framework, full integration of NTP into the kernel, multiple alternative timer sources across the entire kernel not just parts of the network stack, full realtime posix timers, possibility to alter HZ above 2500 without bastardizing the kernel to hell, so on so fourth
Who actually uses FreeBSD?
Anyone who needs a proper working R&D platform with predictable and well documented behaviour. For example I do most of my R&D on BSD because it is written by the book and I can compare what I do with the actual articles and papers written by people. Once I got working what I want I move it to linux because this is what people tend to use. This is also the moment I usually start cursing.
As a matter of fact they had it. Several of the hottest summers on record over the last 10 years. Simply Russia is seeing what all of Europe will see as the gulfstream weakens which is what is supposed to be the effect of Global warming. Hotter summers and colder winters.
The direct data sample is actually smaller. It is less than 30 years. Before that there were no weather satellites and ground stations have never covered the entire globe. 200 years from ground stations are available only for 20-30 locations mostly in Western Europe and Eastern USA..
Indirect data sample - Oxygen isotope distribution, CO2 content, methane content, morphology of some algae and plankton, etc spans back nearly a million years now. All of these can be used to get an estimate for a global or local temperature average. The last 10000 are covered with fairly good precision.
So your 200 years claim is bogus. If you are talking about direct data there is considerably less than that. If you are talking about all data, there is a useable sample going back 10000+ years.
3ware used to be extremely sensitive to bus noise. As a result on at least some motherboards it will lead to memory corruption and outright crashes. If mounting on a riser card ensure that the riser is grounded on all ground contact points. There were discussions of the underlying problem (initially erroneously blamed on driver bugs) all over the FreeBSD mailing lists.
Compared to a decent modern SCSI system like one of the more recent Compaq smartraids or a recent Adaptec, 3ware with SATA is quite slow. From the ones I had to work with, it is the only hardware RAID adapter which is so slow that it triggers the silly OOM killer mistakes in Linux 2.6.10 even on RAID1. As a measure of how slow it is - reading off a 5 year old SCSI based SmartRAID on a P3 600 over 1Gbit NFS and writing out onto a quad Opteron with 3ware congests the 3ware on IO. The system goes into the twighlight zone of 12+ loadavg.
So on, so fourth. The only advantage 3ware has over other systems are the reasonably well supported and well working drivers on both Linux and FreeBSD. If you need performance SCSI is still the answer.
Unless I am mistaken there is no intention to hit the wall as far as AMD is concerned.
AMD has not increased the heat output for quite a while. Their CPUs still produce the same heat. IIRC it is 65, 85 or 110 depending on the submodel for Athlon and Opteron. If their CTO is to be believed they do not intend to change any of these values anytime soon. They will ship 110 for people who do not care, 85 for ones who kind'a care and 65 for blades and small form factor. They intend to increase the performance while keeping to one of these "sweet spots" for all three types.
Simply the market has demanded quieter and quieter PCs lately. As a result Intel went the BTX route which provides lower noise and better cooling due to a new case design. AMD dropped the noise on their coolers by changing the cooler design at least twice over the last 2 years. Possibly more times. This is from looking at upgrade leftovers I have left which are not that many so they do not make a statistically significant sample.
With the minor difference that Windows users can run them in a decent resolution and decent refresh rates for something simple as Microshaft Word. Linux users will not be able to do this for a while. And I mean users, not freaks living on the apt-get unstable edge or developers.
When something really new comes out it takes at least 6 months+ for it to get filtered down at X-whatever level and at least a year for it getting to a distribution level. There are so many branches and fig leaves on the X tree now that even if ATI releases a source for any specific X-whatever it will be bloody useless to anyone but that specific X-whatever.
Just as an example - I looked at merging the changes for the recent crop of PCI Express Radeons from X.org into the debian sarge (Thy Last XFree86) driver and I simply had to set it aside. A few days job at least because in their infinite wisdom the X.org people have started merging things like DRM right into the code instead of separating them out. So as a result the source which was arcane before the split is now outright mindbending in some places. Code maintainability... Yeah... Right... Heard about it...
Someone somewhere has been charging for better QoS for years. Level 3 in the US has been charging for better qos for 4+ years now. It is time for the slashdot crowd to wake up and face reality. It has been done for a very long time now. I used to work in an ISP that made a living off this 8 years ago. It is beeing done and it will be done.
And you are absolutely correct, at the end it all balances out. There is as limit to what you can squeeze out of a network. If you go over it the users will eat your arse. If you do not use it you are forced to invest more in hardware and connectivity.
Slashdot Mirror
If Skype gets anywhere close to mainstream it will be forced to provide mandatory legal intercept without any need for inventing fantom associations. This will be regardless of its use for drug dealing or not.
And here comes the most interesting bit. In order to provide legal intercept capabilities it will have to provide law enforcement access via a remote control interface to computers serving as supernodes in the P2P network. These computers are not even owned by Skype and may be outside the jurisdiction of the party requesting intercept. In fact intercepting on them may be illegal in the country where they are located. This is bound to get very entertaining at some point sooner or later.
And by the way using specially dedicated nodes for legal intercept only will not work because one of the requirements for legal intercept in telephony is that it should not be noticeable to either party in the conversation. A node located in a strange place will very happily show up in the netstat on both Linux and Windows and writing a utility which shows which address block is the supernode you are connected to is a piece of cake.
Scotland laws differ from England and Wales (you gotta love the UK). The actual enforcement framework and guidelines differ even further. Strange as it may seem it is true. For example the UK law is dated 1999 and the Scottish one is dated 2003 and as far as I know they actually differ.
No.
Trolls like NTP are not the worst.
The real problem is with companies who have a less efficient and more expensive product, but buy off possibly competing patent technology and keep it in their safes until doomsday.
Oil companies own patents on battery and hybrid vehicle technologies and keep them in their safes, chemical companies keep patents on long life tires and do not develop them so that they can sell inferior product, so on so fourth.
These are what really stiffles innovation and should be dealt with. The little trolls are a buyproduct of the rules set by the big players for their own benefit. If the patent system is changed to deal with them the little trolls will die off from natural causes.
Not quite.
d renathomealone.htm. In a very narrow reading of these guidelines it is illegal to let a child under 13 travel on their own (by extension of the rule that they should not be left on their own). This has been applied this way across the UK multiple times (according to friends of mine who work as teachers).
UK law is wonderfully vague. It says that it is illegal to put a child at risk without providing a clear definition of risk and responsibility.
In practice the local authorities usually follow these guidelines: http://www.nspcc.org.uk/html/home/needadvice/chil
Is this application right or wrong - IANAL. All I can say is that it errodes the sense of responsibility in the child.
Err...
Is it this or mashing biology, chemistry and physics into a half baked mash called "Science"?
Or the complete liquidation of any homework and any home assignments in primary school?
Or the idiotic laws that force the parents to babysit their offspring till they are 14 years old removing any sense of reason and responsibility? I remember that at the age of 7 I had to travel across one quarter of a 10 million city alone to school every day. And I was not the only one to do so. In fact there was not a single parent dropping off or picking up children after the first week. Frankly, before we get to the crazy frog, shooting all the MP critters who pushed this stupid law followed by a selective school run cull may be a better place to start.
MTBF is MEAN TIME BEFORE FAILURE not Time before guaranteed failure. It is the Expected Value of your failure distribution. By increasing your MTBF, you are decreasing the probability of failure within your estimated installation lifetime. As a result over your projected installation lifetime the overall number of failures and overall maintenance expenditure decreases. When managing 100+ servers it is guaranteed to have some failures due to normal wear and tear even during the warranty period. Not trying to decrease their rate is a prime case of masohism.
As far as your experience is concerned - 5 years is a long shot for the systems that roll off the production line today. In fact very few of them will survive that length of time without power management. Out of the 1U systems I have dealt with lately not a single one. This was not the case with most old servers from 5+ years ago as they were designed for OSes which did not have power management at the time. As a result their cooling system capacity is way over what they actually need. Just opening a Proliant G1, G2 and G3 side by side is a good enough example. The cooling capacity in them does not increase while the heat emission from the CPU increases more then 4 times.
Same here. But they have both ACPI and CPU frequency scaling (aka centrino, powernow, longrun, longhaul) enabled.
The reasons are fairly simple.
There is only one case where power management is unnecessary. It is if you are running a computational load which requires your servers to work 365x24x7 flat out. Running without ACPI and cpufreq (if supported) for an average corporate or ISP load is plain stupid. It results in a less reliable server installation.
By the way, do not worry, 99.99 of the server sysadmins out there join you in this stupidity. In fact I did 7 years ago when I knew less about server administration.
Not quite correct.
While I am an old AMD fanboy and given a choice I will always chose AMD Intel must be given credit where credit is due. If you use CPU frequency scaling on Intel which is supported on both Pentium 4 and Xeon you can drop its consumption into the sub-25W territory when idle and ramp it up in a lot of increments (usually 8) to full as the load arrives. While in theory AMD powernow-k8 should give you similar features, in practice I have yet to see an SMP motherboard that does not have it disabled. It is also not supported properly until 2.6.13 which means that it will not work on the stock RedHat kernel.
As a result of this you will get a considerable power consumption improvement when moving to AMD for things that run full blast all the time like computation. You will not see this improvement in a correctly configured Xeon installation running Linux on recent RedHat ES or Debian (or anything with a post 2.6.7 kernel). In fact there Intel will beat AMD flat out.
The important item here is "correctly configured". Based on my not very scientific observations 99.99% of the corporate IT out there is not competent enough to do this. 1 out of 1000 on average knows that you can turn most of the Centrino-like features on a server. The rest have no clue.
You can use most of these features on a desktop or a server if you:
1. Use Linux
2. Bother to turn it on
Just look at the kernel documentation for cpufreq and the on_demand governor. Alternatively you can use cpufreqd which allows even finer tuning.
Turning it on for a dual CPU Xeon will drop the power consumption from 400W+ to under 100W when idle. In fact this feature works on any Pentium4 class CPU.
Numbers for Opterons are similar, but most dual and quad Opteron motherboards lack proper support for this feature.
They are nice, but they have their limitations. On the positive side:
- Very low power consumption.
- Very high IO speed. In fact considerably higher than expected. I have been getting 2+ times faster IO than from an Intel Xeon from them.
- AES acceleration on the higher end models, high quality hardware RNG and RSA acceleration on the models coming up this year
On the negative side:- Very small cache. Much smaller than anything else on the market. As a result it is blazing fast to start and blazing fast on any small data set, but churning large data slows it down considerably. This also causes task switching under higher loads to be considerably slower than expected.
- The default factory fans on the higher end models are not temperature controlled. They have an MTBF of around a year and a half which is too low by modern Datacenter standards. On the positive side, the thermal throttle on them is good enough to allow them to run with the fan failed.
Overall they are good enough to run a mail relay, a DNS server a firewall, a QoS conditioner or an Odds-N-Sods. They are definitely slower then I would like to for a news server. CGI performance and MySQL are abissmal and computation or compilation are in the "you might as well forget it" area.The most painful part is actually the relatively low number of people providing good server class enclosures for them. This http://www.icp-epia.co.uk/ is one of the few sites. Even they do not have everything so you end up assembling most of them yourself which costs time and money.
This will work for an office. It will not work for a datacenter which is the original question. The heat density in a datacenter is quite often in the 16KW per rack range for peak consumption. Precooling this will give at most 5 minutes. No point to bother
Why only peak load reduction? Load reduction in terms of electricity is the actual way to start.
A good example - a Dual CPU Pentium4 Xeon Intel OEM system in a standard Intel OEM chassis eats nearly 400W when idle with no power management. With standard ACPI power management when idle it eats 350-380W. With CPU frequency scaling using the ondemand CPU governor it will eat less then 100W when idle. The numbers for Opteron based systems are not much different. A usual datacenter is designed to cope with full demand after losing up to 25% loss in terms of CPU capacity due to failures. This across a datacenter means a saving of up to 3KW per rack at peal capacity and 9KW+ when idle for fully loaded 42U racks. It also means increasing MTBF on fans by up to 2 times, increasing MTBF on the airconditioner by up to 4 times, improving disk MTBF by 20-30% due to lower heat, so on so fourth.
Now, here is the 1 million KWh question:
How many server sysadmins turn this on? How many server sysadmins actually check server systems during evaluation for working CPU frequency scaling and working power management? How many server sysadmins actually take this into account during capacity planning? As a matter of fact how many perform thermal correction for MTBF when doing datacenter and system design?
Looking around I can say that in the UK it is somewhere between 0 and 0.01%. Dunno about other countries.
The diagram you are pointing to is a lie. If it was true the BES would have been talking to handsets directly and it does not. It talks to something in RIM. That something talks to the provider GPRS network and via it to the handheld.
This is what I am referring to.
No-one knows what happens there. There is no information in the press and all marketing literature is full of lies. All of it carefully omits this RIM side component. Personally, I seriously doubt that they have the computing power in the older handhelds to do 3DES or AES on all messages (as they claim). If they did why did they implement both x509/SMIME and PGP at the server and not at the handheld defeating the entire idea of end-to-end encryption?
If I recall correctly the Crackberry is the only wireless email system where the PGP and x509 integration are solely at the server. The messages are decrypted at the server and sent to the device unencrypted perusing only the GPRS wireless encryption which is pretty weak by modern standards. They are temporarily stored at the device unencrypted as well, which defeats the idea of PGP and x509 completely.
Not a dupe, but I do not quite understand the article classification.
First of all, cudos to the NetBSD crowd for maintaining an OS that can run even on a Dead Marmot. Over the years, I have run it on several Dead Marmots (TM) like MIPS 3000 DecStations (should not be mistaken with Dead Badgers which can run Linux).
What I do not understand is the article classification. Surely, the more popular of the BSDs should have gotten prime time coverage and the less popular coverage between headlines. What's the deal here? Or the FreeBSD people should port their OS to a Dead Lemming instead of dropping the support for anything pre-80486 alltogether?
Yes you can. It was most likely compiled versus Blow^H^H^H^H^MoTiff. That is at best. At worst versus Sun's old stuff (so old now, that forgot what it was called...). In order for it to be useable on a modern Linux or BSD desktop it will have to be ported to a newer graphic toolkit and widget set. Which brings us back to square one.
As a reference model on how things are supposed to be done by the book with professionally commented and written source and properly written documentation.
What does it do that Linux doesn't?
Correct integration of statefull firewall and QoS framework, full integration of NTP into the kernel, multiple alternative timer sources across the entire kernel not just parts of the network stack, full realtime posix timers, possibility to alter HZ above 2500 without bastardizing the kernel to hell, so on so fourth
Who actually uses FreeBSD?
Anyone who needs a proper working R&D platform with predictable and well documented behaviour. For example I do most of my R&D on BSD because it is written by the book and I can compare what I do with the actual articles and papers written by people. Once I got working what I want I move it to linux because this is what people tend to use. This is also the moment I usually start cursing.
Here went my moderation...
Not for spite.
Check the recent investors in NTP (or licensees - investment SCO style).
Check the owners of alternative solutions to RIM which are at the moment bottom feeding on scraps from the edges of the market.
Compare.
Think.
As a matter of fact they had it. Several of the hottest summers on record over the last 10 years. Simply Russia is seeing what all of Europe will see as the gulfstream weakens which is what is supposed to be the effect of Global warming. Hotter summers and colder winters.
You are incorrect on multiple counts.
The direct data sample is actually smaller. It is less than 30 years. Before that there were no weather satellites and ground stations have never covered the entire globe. 200 years from ground stations are available only for 20-30 locations mostly in Western Europe and Eastern USA..
Indirect data sample - Oxygen isotope distribution, CO2 content, methane content, morphology of some algae and plankton, etc spans back nearly a million years now. All of these can be used to get an estimate for a global or local temperature average. The last 10000 are covered with fairly good precision.
So your 200 years claim is bogus. If you are talking about direct data there is considerably less than that. If you are talking about all data, there is a useable sample going back 10000+ years.
3ware used to be extremely sensitive to bus noise. As a result on at least some motherboards it will lead to memory corruption and outright crashes. If mounting on a riser card ensure that the riser is grounded on all ground contact points. There were discussions of the underlying problem (initially erroneously blamed on driver bugs) all over the FreeBSD mailing lists.
Compared to a decent modern SCSI system like one of the more recent Compaq smartraids or a recent Adaptec, 3ware with SATA is quite slow. From the ones I had to work with, it is the only hardware RAID adapter which is so slow that it triggers the silly OOM killer mistakes in Linux 2.6.10 even on RAID1. As a measure of how slow it is - reading off a 5 year old SCSI based SmartRAID on a P3 600 over 1Gbit NFS and writing out onto a quad Opteron with 3ware congests the 3ware on IO. The system goes into the twighlight zone of 12+ loadavg.
So on, so fourth. The only advantage 3ware has over other systems are the reasonably well supported and well working drivers on both Linux and FreeBSD. If you need performance SCSI is still the answer.
Unless I am mistaken there is no intention to hit the wall as far as AMD is concerned.
AMD has not increased the heat output for quite a while. Their CPUs still produce the same heat. IIRC it is 65, 85 or 110 depending on the submodel for Athlon and Opteron. If their CTO is to be believed they do not intend to change any of these values anytime soon. They will ship 110 for people who do not care, 85 for ones who kind'a care and 65 for blades and small form factor. They intend to increase the performance while keeping to one of these "sweet spots" for all three types.
Simply the market has demanded quieter and quieter PCs lately. As a result Intel went the BTX route which provides lower noise and better cooling due to a new case design. AMD dropped the noise on their coolers by changing the cooler design at least twice over the last 2 years. Possibly more times. This is from looking at upgrade leftovers I have left which are not that many so they do not make a statistically significant sample.
Just run a webserver and post in slashdot. I am sure accounts will be mildly entertained the moment they get the GPRS bill.
With the minor difference that Windows users can run them in a decent resolution and decent refresh rates for something simple as Microshaft Word. Linux users will not be able to do this for a while. And I mean users, not freaks living on the apt-get unstable edge or developers.
When something really new comes out it takes at least 6 months+ for it to get filtered down at X-whatever level and at least a year for it getting to a distribution level. There are so many branches and fig leaves on the X tree now that even if ATI releases a source for any specific X-whatever it will be bloody useless to anyone but that specific X-whatever.
Just as an example - I looked at merging the changes for the recent crop of PCI Express Radeons from X.org into the debian sarge (Thy Last XFree86) driver and I simply had to set it aside. A few days job at least because in their infinite wisdom the X.org people have started merging things like DRM right into the code instead of separating them out. So as a result the source which was arcane before the split is now outright mindbending in some places. Code maintainability... Yeah... Right... Heard about it...
Grgghhhhhhhhhhh....
Someone somewhere has been charging for better QoS for years. Level 3 in the US has been charging for better qos for 4+ years now. It is time for the slashdot crowd to wake up and face reality. It has been done for a very long time now. I used to work in an ISP that made a living off this 8 years ago. It is beeing done and it will be done.
And you are absolutely correct, at the end it all balances out. There is as limit to what you can squeeze out of a network. If you go over it the users will eat your arse. If you do not use it you are forced to invest more in hardware and connectivity.