"It was his baby and he owned it."
He may have designed it, and he may have built the majority of it by himself, but the city owned it, not him.
If I pay a builder to build a house for me, it is mine. Not his, he may win awards for it, but it is mine. As long as I paid him for it is mine and he must had the keys over to me. He can't decide that he wants to keep it.
Actually MS did NOT write the OS for the PC back in the 80's. Bill G. purchased a OS (86-DOS, a.k.a.QDOS) for the PC back then and renamed it.
Typically people don't consider hardware open, they consider software open and MS did not own the hardware.
I remember that, was called departmental computing. I had friends that laughed at me when I went into mainframes in the early 80's and the mainframes were going to be gone in a few years.
Here it is 20+ years later and I still work with mainframes.
I got into the networking side and I still do mainframe networking along with distributed networking.
O.K, a little box with a few CPUs, put as much external connectivity inside the box and say I have saved the world. Blade servers have been around for awhile. Virtualization is old hat.
Mainframes have been doing this for a long time. IBM has been doing virtualization at the OS level since the 60's and at the hardware level since at least the early 80's.
Depending on workload an IBM z10 can have thousands of virtual OS's running in less than 20 sq of floor space. Just about the same space as 3 standard server racks.
I know of company running about about 500 virtual Linux images on a single 10-way z9. The z10 has about twice the capacity of a z9. On average a z9 uses less energy than a single rack of blade servers and the z10 uses less energy than a z9.
If you want you can now run Solaris on an IBM mainframe under z/VM. There is another company that has just released a product that allows you to run virtual Windows systems on an IBM mainframe. Yes, Microsoft Windows running on an IBM mainframe.
Its like people don't realize that this is OLD technology not new. The idea if a few big computers running everything and sharing resources is all what mainframes have been about and are still about.
Yep, thats the answer. Lets setup VPN's between every computer and every web, ftp, chat, and whatever server. Or use https, ftps, sftp, etc. on every connection.
Just who is going to end up paying for all the equipment upgrades to support all that encryption and decryption on the servers or the extra hardware that needs to be purchased to offload the encryption?
Who is going to pay for all of those certificates for SSL encryption? You do realize that the reason most sites don't offer SSL is because they have to pay for certificates? If they use self-signed certificates, they you get those annoying pop-up windows because you browser can't verify the certificate. I for one do NOT want to have to import thousands of self-singed certs.
Yep, entry level is about $300K for the 1st year and $50K per year after that. However what are you comparing that to? That $300K initial cost and $50K on going could be supporting 500-1,000 users. In environments with 1,500 or more end users the avg. mainframe cost per end user is somewhere around $1,500 - $2,000 where the avg. cost per end user in the distributed server world is $5,000+ per end user.
You can't just look at the cost, you must look at the cost per something and the total cost for your enviroment. Example: We had somebody pricing out memory on zSeries it was $10,000 per GB, so 8GB was $80K and they quickly pointed out that for our high end servers it was only $1,000 per GB, so only $8,000 per 8GB. I quickly pointed out that we only had to spend $160,000 to get both our mainframes an additional 8GB of memory each, whereas he was going to spend $320K to get the 40 servers in the server farm upgraded with an additional 8GB each. That did not matter, the fact was per GB the distributed servers were less expensive.
Have you see IBM's new z6 processor? Quad core 4.4GHz, 2 I/O interfaces running 17GB ps each, 2 interface for SMP function at 48GB ps each, and 4 memory interface at 13 GB ps each. Each core has a decimal math accelerator, each pair of cores share a compression/decompression accelerator and a encryption/decryption accelerator.
Their new z10 mainframe can have up to 20 of these (80 cores) of which 64 of the cores can be configured for the customer to use. The other 16 cores are used for spares and special processing.
The z6 (or z10 CPU) is very closely related to IBM's Power6.
Are you implying that Windows is just as reliable as z/OS? I don't think so.
Unfortunately IBM will not publish any standard benchmarks for zSeries systems running z/OS. The problem is they don't want to publish the list price as each system sold is basically a custom packaged and priced system.
What I will say is that there are Intel based systems that can run z/OS using special software. The biggest one out there is based on NEC's newest 32-way dual-core (64 total cores) XEON box. According to the company that sells the system it is rated at about 3,500 z/OS MIPS. The new z/10 would only need about 4 CPU's to get to the 3,000 MIPS range, even the older z9 would only need 6 CPU's, and the even older z990 would only need about 10 CPU's.
I can't believe anybody in their right mind would spend millions of dollars on a Windows server. A Linux server, xxxxBSD server, or a Unix server, sure.
The actual upgrade of the hardware to add more processing power is simple and easy. The processors come in "books", each box (depending on the model you have: z990, z9, z10) will have anywhere from 8 to 20 user assignable CPU's already installed. If you are not using all of the CPU's, the IBM CE just dials into the the mainframe service element remotely and turns on another CPU. If you are using all of the CPU's in the currently installed book(s), he then must come out and install a new book, which on the z9 and z10 can be done concurrently and non-distributively. For the most part upgrading memory can be just as easy.
The long part is the paper work on purchasing the upgrade, getting software licenses upgraded, and installing any new license codes that may be required.
Even better. Say you finally run out of horse power on your z9 and you must upgrade to a z10. With a push pull upgrade, you disconnect all your I/O cables, unplug the z9, roll it out of the way, roll in the z10, plug in all I/O cables, plug in the power, IPL, and your done. Image, the hardware for hundreds or even thousands of "servers" upgraded that easily. Total time 3-4 hours. Oh and generally the new hardware has less energy requirements, so you reduce your environmental costs also.
We had a new director for our data center that had not worked with mainframes for 20 years. He was amazed that we swapped out our mainframe in 3 hours. That was 3 hours from the time we started to shut the system down, until it was back up and running. No re-install of the OS, no re-deployment of any applications. He just imaged what we could do if we had all of our distributed systems running on one of these in virtual environments.
The z9 supported 16 2.7 GBps (yes gigabytes) I/O interfaces for a total of 43 GBps of I/O. The z10 supports up to 16 InfiniBand I/O inter connects running at 6 GBps each, for 96 GBps of I/O bandwidth.
Not sure what type of data center you are in, but we still need 3-phase power for some of our non-zSeries boxes. Most blade servers require it. Our non-zSeries boxes generate more heat and draw more power than our 2 zSeries servers combined and we only have about 70 small Intel/AMD based system.
True you can't run x86 binaries on a zSeries, but you can't run zSeries binaries on a x86 and didn't seem to be a problem when everybody started to migrate away from the mainframe. Why should it be a problem to migrate back?
You can run most Linux/POSIX based code by just re-compiling it under Linux on zSeries. Windows code will need to be changed, maybe. I am assuming that you could run WINE under Linux on zSeries and thus that would allow you to run anything that can run under WINE. But I would not want to see what the performance would look like.:)
Now, if you want to run your program under z/OS (one of the operating systems that you can run on zSeries) then you have some code re-writing to do, unless you are running Java code. Which should run with very little if any changes under z/OS.
Unless you are running Linux only, and only truly free Linux based code, and you are running with out support (which you can do on zSeries also) you still need to license software on x86. The licensing fees for z/OS include support also and in some cases upgrades. Whereas Windows based software (especially Windows itself) only includes the ability to run the software and sometimes software upgrades. If you want support you must pay extra and for most Windows based software you have to pay to upgrade.
As for DASD, if you are running Linux only you can connect any SAN that support fiber channel to the mainframe. You only need "special" DASD if you plan to run z/OS.
The System Programming books are for z/OS, not necessarily zSeries. You don't need a z/OS system programmer to run Linux on zSeries.
In some instances you seem to be getting hardware (zSeries) and software (z/OS) requirements mixed up.
Based on some reading the NYSE was running on IBM mainframes running z/OS as the back end system, the clearing system and billing system.. But early last year they announce they were replacing all with pSeries server running AIX. This was to save money. I am not sure if they have migrated, but one article stated they had 1600 zSeries MIPS, which although is quite a bit, is no where near the capacity of the top end z9.
Yes, z800 was replaced by z890, which was replaced by the z9BC. I actually forgot just how powerful the z9 is. We just got z990's mid 2006, IBM wanted to get our z900 off the floor, z990 out of the pipeline and we did not need z9s (we wanted them, but we didn't need them).
It is weird, the mainframe couldn't handled the workload, but they are still keeping it and running things on it. My guess is that the workload moved off was low priorty work and basically got what was left over from other, more critical, workloads. Also they don't state what DASD system they were using, if it was ESCON or FICON and how many channels they had to it.
Just read the article. The "mainframe" was a 2066-002, which is a crippled/baby mainframe with two CPUs (limted to 4), total MIPS is 384. The lastest and greatest mainframe has over 450 MIPS on a single CPU. Not what I would consider a powerfull mainframe.
You really can't compare MIPS on any platform to MIPS on any other platform. What my company was informally told was that for every 1 mainframe MIPS you would need anywhere between 10 and 400 x86 MIPS depending on the profile of your application. Big range there. Where mainframes typically excel is at accessing large volumes of data and doing it quickly. The top of the range mainframe today can go to 54 processors, which I know is not that impressive as there are other platforms that have more, have 512GB of "RAM", which other can have more, but it does have a total system capacity of 192 GB (yes GIGA BYTES) of I/O throughput. Now you can also hotswap any and every component on todays mainframe: CPU, "RAM, and I/O cards. It has up to 96 seperate 2 GB (yes GIGA BYTES) I/O buses.
Mainframes and super computers are two different types of computers and are targeted and two different markets.
In fact the super computer market is gone. I'm not even sure when the last "super comptuer" was built. Today's "super computers" are nothing but clustered smaller computers, either Intel/AMD x86 or PowerPC based, have been for at least 5 years.
Mainframes are not targeted for compute intensive only work. They are targeted as a general purpose computer that can run multiple types of workloads at the same time. I know of sites that are using the virtualization capabilities of IBM's mainframes to run thousands of virtual Linux images that replaced or prevented the installation of thousands of Intel based boxes. One company has estimated that they will save $16 million dollars over the next 3 years by running 350 virtual Linux images on the mainframe vs. running them on real hardware or even virtual images using VMWare. This was after they purchased two IBM mainframes just for the virtual Linux enviroment.
Not sure how current your mainframe experience is, but today's mainframes only single point of failure is power. It must have electricity to run.
Decentralization can have it problems. I know of sites that have tried to get rid of their mainframe, but after trying to create a database cluster of a few hundred boxes, they realized that updates are a pain, performance was no where near what the mainframe had, software costs were way more than they ever thought they would be (who new MS and Oracle charged by the CPU?) and the people needed to run 500-1000 servers were much more expensive than the number of people needed to run two mainframes.
Now mainframes are not for everybody. I would not suggest every company run mainframes, but mainframes do have their places. They still run 60-70 of all business critical transactions in the world.
If you are in a mainframe (excuse me zSeries Server) enviroment, the HMC, the ESCON director, MOSS/E console for the 3746-900, the APPN Network Node Processor on the 3746-900, and the console for older 2105's (Shark's) are all still OS/2. The newer sharks (2150-800) use Linux for the consols and I hear that the z9's HMC is Linux.
Mainframes do use PowerPC processors. However, they use them in their I/O Processor cards, not as their central processing units. They use special processors for that. To show you just how small mainframes are, the current z900 has 20 CPU's on a card that is about 8"x8", the z990 has 12 CPU's on a card that is about 6"x6". I am not sure about the size on the just announced z9. The bulk of the size in the current mainframe is for I/O connections. The CPU's and memory take up less that 1/4 of the box, that includes the cooling using. The box about the size of two racks, 1/4 is CPU in memory, 3/4 is for I/O connections.
They are using CMOS based processors and have been for about 10 years now. However it took them about 3-4 years to get the CMOS chips to have as much processing power as the older bi-polar based chips.
They do use different CPU archictures and even memory access and protection methods. On a mainframe box the hardware/microcode, not the OS, is responsible to make sure that one task does not access memory that it is not authorized to.
Because of the way the memory is shared, there is no penalty for excuting a task on a different processor.
Each CPU has two instruction units. Each instruction is executed twice on each CPU. The results are compared and if the are the same, then the resutlts are used. If they are different, the instruction is re-exected and results compared. If they are the same, the results are used. If they are not the same the CPU is marked as bad, taken offline and backup CPU is enabled.
You do not get a bad execution result because of sun flares or magents.
At some time in the past MS was having some performance problems and they use Akamai's web caching services to handle the load. This was noticed because Akamai uses Linux/Apache boxes for web caching and some of MS sites were showing up as running Linux and Apache
Slashdot Mirror
"It was his baby and he owned it." He may have designed it, and he may have built the majority of it by himself, but the city owned it, not him. If I pay a builder to build a house for me, it is mine. Not his, he may win awards for it, but it is mine. As long as I paid him for it is mine and he must had the keys over to me. He can't decide that he wants to keep it.
Actually MS did NOT write the OS for the PC back in the 80's. Bill G. purchased a OS (86-DOS, a.k.a.QDOS) for the PC back then and renamed it. Typically people don't consider hardware open, they consider software open and MS did not own the hardware.
I remember that, was called departmental computing. I had friends that laughed at me when I went into mainframes in the early 80's and the mainframes were going to be gone in a few years. Here it is 20+ years later and I still work with mainframes. I got into the networking side and I still do mainframe networking along with distributed networking.
O.K, a little box with a few CPUs, put as much external connectivity inside the box and say I have saved the world. Blade servers have been around for awhile. Virtualization is old hat. Mainframes have been doing this for a long time. IBM has been doing virtualization at the OS level since the 60's and at the hardware level since at least the early 80's. Depending on workload an IBM z10 can have thousands of virtual OS's running in less than 20 sq of floor space. Just about the same space as 3 standard server racks. I know of company running about about 500 virtual Linux images on a single 10-way z9. The z10 has about twice the capacity of a z9. On average a z9 uses less energy than a single rack of blade servers and the z10 uses less energy than a z9. If you want you can now run Solaris on an IBM mainframe under z/VM. There is another company that has just released a product that allows you to run virtual Windows systems on an IBM mainframe. Yes, Microsoft Windows running on an IBM mainframe. Its like people don't realize that this is OLD technology not new. The idea if a few big computers running everything and sharing resources is all what mainframes have been about and are still about.
Yep, thats the answer. Lets setup VPN's between every computer and every web, ftp, chat, and whatever server. Or use https, ftps, sftp, etc. on every connection. Just who is going to end up paying for all the equipment upgrades to support all that encryption and decryption on the servers or the extra hardware that needs to be purchased to offload the encryption? Who is going to pay for all of those certificates for SSL encryption? You do realize that the reason most sites don't offer SSL is because they have to pay for certificates? If they use self-signed certificates, they you get those annoying pop-up windows because you browser can't verify the certificate. I for one do NOT want to have to import thousands of self-singed certs.
Yep, entry level is about $300K for the 1st year and $50K per year after that. However what are you comparing that to? That $300K initial cost and $50K on going could be supporting 500-1,000 users. In environments with 1,500 or more end users the avg. mainframe cost per end user is somewhere around $1,500 - $2,000 where the avg. cost per end user in the distributed server world is $5,000+ per end user. You can't just look at the cost, you must look at the cost per something and the total cost for your enviroment. Example: We had somebody pricing out memory on zSeries it was $10,000 per GB, so 8GB was $80K and they quickly pointed out that for our high end servers it was only $1,000 per GB, so only $8,000 per 8GB. I quickly pointed out that we only had to spend $160,000 to get both our mainframes an additional 8GB of memory each, whereas he was going to spend $320K to get the 40 servers in the server farm upgraded with an additional 8GB each. That did not matter, the fact was per GB the distributed servers were less expensive.
I guess you missed the recent announcement of the z10 mainframe, 4.4 Ghz. Seem fairly fast to me.
Have you see IBM's new z6 processor? Quad core 4.4GHz, 2 I/O interfaces running 17GB ps each, 2 interface for SMP function at 48GB ps each, and 4 memory interface at 13 GB ps each. Each core has a decimal math accelerator, each pair of cores share a compression/decompression accelerator and a encryption/decryption accelerator.
Their new z10 mainframe can have up to 20 of these (80 cores) of which 64 of the cores can be configured for the customer to use. The other 16 cores are used for spares and special processing.
The z6 (or z10 CPU) is very closely related to IBM's Power6.
http://en.wikipedia.org/wiki/IBM_z6
http://www2.hursley.ibm.com/decimal/IBM-z6-mainframe-microprocessor-Webb.pdf
Are you implying that Windows is just as reliable as z/OS? I don't think so. Unfortunately IBM will not publish any standard benchmarks for zSeries systems running z/OS. The problem is they don't want to publish the list price as each system sold is basically a custom packaged and priced system. What I will say is that there are Intel based systems that can run z/OS using special software. The biggest one out there is based on NEC's newest 32-way dual-core (64 total cores) XEON box. According to the company that sells the system it is rated at about 3,500 z/OS MIPS. The new z/10 would only need about 4 CPU's to get to the 3,000 MIPS range, even the older z9 would only need 6 CPU's, and the even older z990 would only need about 10 CPU's. I can't believe anybody in their right mind would spend millions of dollars on a Windows server. A Linux server, xxxxBSD server, or a Unix server, sure.
The actual upgrade of the hardware to add more processing power is simple and easy. The processors come in "books", each box (depending on the model you have: z990, z9, z10) will have anywhere from 8 to 20 user assignable CPU's already installed. If you are not using all of the CPU's, the IBM CE just dials into the the mainframe service element remotely and turns on another CPU. If you are using all of the CPU's in the currently installed book(s), he then must come out and install a new book, which on the z9 and z10 can be done concurrently and non-distributively. For the most part upgrading memory can be just as easy. The long part is the paper work on purchasing the upgrade, getting software licenses upgraded, and installing any new license codes that may be required.
Even better. Say you finally run out of horse power on your z9 and you must upgrade to a z10. With a push pull upgrade, you disconnect all your I/O cables, unplug the z9, roll it out of the way, roll in the z10, plug in all I/O cables, plug in the power, IPL, and your done. Image, the hardware for hundreds or even thousands of "servers" upgraded that easily. Total time 3-4 hours. Oh and generally the new hardware has less energy requirements, so you reduce your environmental costs also. We had a new director for our data center that had not worked with mainframes for 20 years. He was amazed that we swapped out our mainframe in 3 hours. That was 3 hours from the time we started to shut the system down, until it was back up and running. No re-install of the OS, no re-deployment of any applications. He just imaged what we could do if we had all of our distributed systems running on one of these in virtual environments.
Sounds more like a poorly planned/setup disk enviroment.
The z9 supported 16 2.7 GBps (yes gigabytes) I/O interfaces for a total of 43 GBps of I/O. The z10 supports up to 16 InfiniBand I/O inter connects running at 6 GBps each, for 96 GBps of I/O bandwidth.
Not sure what type of data center you are in, but we still need 3-phase power for some of our non-zSeries boxes. Most blade servers require it. Our non-zSeries boxes generate more heat and draw more power than our 2 zSeries servers combined and we only have about 70 small Intel/AMD based system. True you can't run x86 binaries on a zSeries, but you can't run zSeries binaries on a x86 and didn't seem to be a problem when everybody started to migrate away from the mainframe. Why should it be a problem to migrate back? You can run most Linux/POSIX based code by just re-compiling it under Linux on zSeries. Windows code will need to be changed, maybe. I am assuming that you could run WINE under Linux on zSeries and thus that would allow you to run anything that can run under WINE. But I would not want to see what the performance would look like. :)
Now, if you want to run your program under z/OS (one of the operating systems that you can run on zSeries) then you have some code re-writing to do, unless you are running Java code. Which should run with very little if any changes under z/OS.
Unless you are running Linux only, and only truly free Linux based code, and you are running with out support (which you can do on zSeries also) you still need to license software on x86. The licensing fees for z/OS include support also and in some cases upgrades. Whereas Windows based software (especially Windows itself) only includes the ability to run the software and sometimes software upgrades. If you want support you must pay extra and for most Windows based software you have to pay to upgrade.
As for DASD, if you are running Linux only you can connect any SAN that support fiber channel to the mainframe. You only need "special" DASD if you plan to run z/OS.
The System Programming books are for z/OS, not necessarily zSeries. You don't need a z/OS system programmer to run Linux on zSeries.
In some instances you seem to be getting hardware (zSeries) and software (z/OS) requirements mixed up.
Based on some reading the NYSE was running on IBM mainframes running z/OS as the back end system, the clearing system and billing system.. But early last year they announce they were replacing all with pSeries server running AIX. This was to save money. I am not sure if they have migrated, but one article stated they had 1600 zSeries MIPS, which although is quite a bit, is no where near the capacity of the top end z9.
Yes, z800 was replaced by z890, which was replaced by the z9BC. I actually forgot just how powerful the z9 is. We just got z990's mid 2006, IBM wanted to get our z900 off the floor, z990 out of the pipeline and we did not need z9s (we wanted them, but we didn't need them). It is weird, the mainframe couldn't handled the workload, but they are still keeping it and running things on it. My guess is that the workload moved off was low priorty work and basically got what was left over from other, more critical, workloads. Also they don't state what DASD system they were using, if it was ESCON or FICON and how many channels they had to it.
Just read the article. The "mainframe" was a 2066-002, which is a crippled/baby mainframe with two CPUs (limted to 4), total MIPS is 384. The lastest and greatest mainframe has over 450 MIPS on a single CPU. Not what I would consider a powerfull mainframe.
You really can't compare MIPS on any platform to MIPS on any other platform. What my company was informally told was that for every 1 mainframe MIPS you would need anywhere between 10 and 400 x86 MIPS depending on the profile of your application. Big range there. Where mainframes typically excel is at accessing large volumes of data and doing it quickly. The top of the range mainframe today can go to 54 processors, which I know is not that impressive as there are other platforms that have more, have 512GB of "RAM", which other can have more, but it does have a total system capacity of 192 GB (yes GIGA BYTES) of I/O throughput. Now you can also hotswap any and every component on todays mainframe: CPU, "RAM, and I/O cards. It has up to 96 seperate 2 GB (yes GIGA BYTES) I/O buses.
Mainframes and super computers are two different types of computers and are targeted and two different markets. In fact the super computer market is gone. I'm not even sure when the last "super comptuer" was built. Today's "super computers" are nothing but clustered smaller computers, either Intel/AMD x86 or PowerPC based, have been for at least 5 years. Mainframes are not targeted for compute intensive only work. They are targeted as a general purpose computer that can run multiple types of workloads at the same time. I know of sites that are using the virtualization capabilities of IBM's mainframes to run thousands of virtual Linux images that replaced or prevented the installation of thousands of Intel based boxes. One company has estimated that they will save $16 million dollars over the next 3 years by running 350 virtual Linux images on the mainframe vs. running them on real hardware or even virtual images using VMWare. This was after they purchased two IBM mainframes just for the virtual Linux enviroment. Not sure how current your mainframe experience is, but today's mainframes only single point of failure is power. It must have electricity to run. Decentralization can have it problems. I know of sites that have tried to get rid of their mainframe, but after trying to create a database cluster of a few hundred boxes, they realized that updates are a pain, performance was no where near what the mainframe had, software costs were way more than they ever thought they would be (who new MS and Oracle charged by the CPU?) and the people needed to run 500-1000 servers were much more expensive than the number of people needed to run two mainframes. Now mainframes are not for everybody. I would not suggest every company run mainframes, but mainframes do have their places. They still run 60-70 of all business critical transactions in the world.
The natural home is z/VM where it was orignally developed. It was then ported to all other of IBM's OS's.
If you are in a mainframe (excuse me zSeries Server) enviroment, the HMC, the ESCON director, MOSS/E console for the 3746-900, the APPN Network Node Processor on the 3746-900, and the console for older 2105's (Shark's) are all still OS/2. The newer sharks (2150-800) use Linux for the consols and I hear that the z9's HMC is Linux.
I think you are getting OS/2 confused with OS/VS V2. OS/2 did not come out until the 80's.
Mainframes do use PowerPC processors. However, they use them in their I/O Processor cards, not as their central processing units. They use special processors for that. To show you just how small mainframes are, the current z900 has 20 CPU's on a card that is about 8"x8", the z990 has 12 CPU's on a card that is about 6"x6". I am not sure about the size on the just announced z9. The bulk of the size in the current mainframe is for I/O connections. The CPU's and memory take up less that 1/4 of the box, that includes the cooling using. The box about the size of two racks, 1/4 is CPU in memory, 3/4 is for I/O connections. They are using CMOS based processors and have been for about 10 years now. However it took them about 3-4 years to get the CMOS chips to have as much processing power as the older bi-polar based chips. They do use different CPU archictures and even memory access and protection methods. On a mainframe box the hardware/microcode, not the OS, is responsible to make sure that one task does not access memory that it is not authorized to. Because of the way the memory is shared, there is no penalty for excuting a task on a different processor. Each CPU has two instruction units. Each instruction is executed twice on each CPU. The results are compared and if the are the same, then the resutlts are used. If they are different, the instruction is re-exected and results compared. If they are the same, the results are used. If they are not the same the CPU is marked as bad, taken offline and backup CPU is enabled. You do not get a bad execution result because of sun flares or magents.
At some time in the past MS was having some performance problems and they use Akamai's web caching services to handle the load. This was noticed because Akamai uses Linux/Apache boxes for web caching and some of MS sites were showing up as running Linux and Apache
Does this mean that instead TCP/IP MS is just giving us TP?