I have seen dell 2650s hit over 800 Megabytes (6.4 Gb) per second running MPI over InfiniBand using large buffer sizes. The limit is pretty much the PCI-X 133 Mhz interface we are on. I suspect that with PCI-X DDR and PCI Express, we will be able to get a lot closer to 10 Gbit.
Really? Seems odd that my company keeps selling it. Sun is using it in their next systems. Network Appliance is using it today. Intel is selling systems with it. Chips are made by Mellanox, Agilent, and Fujitsu. Sun and IBM are doing chips for internal use. There are still a good number of IB vendors as well. See InfiniCon Systems (where I work), TopSpin, InfiniSwitch, Voltaire, and JNI. There are also a bunch of folks using IB in embedded environments. The high performance computing market and database clustering market are really interested in IB. 10 Gb ethernet is many times more expensive and without a TOE that can do 10 Gb, no one can really use it except for switch to switch. 10 Gb fibre channel isn't quite out yet. Also, many companies are still paying people to spend a big chunck of their time working on InfiniBand Trade Association Working Group issues.
Finally, none of the technologies you mentioned solve clustering issues because they are for inside the box, not from box to box.
I work at InfiniCon Systems and we do a lot of InfiniBand related things. We have switching, adapters, and connections from IB to gigabit ethernet and IB to fibre channel. See http://www.infinicon.com/.
The real next steps for IB is 12X (30 Gb) and on mother board IB. 12 X is in development. Currently, IB adapters are limited by the PCI-X slot they sit in. PCI-X DDR and PCI Express should help, but just having it on the mother board and throwing PCI out would be interesting. Small form factor clusters would really benefit to just having memory, CPU, and an HCA.
Not exactly. Most NICs today due DMA without CPU assistance. This talks about writing to another host's memory. Much of it is about removing the TCPI/IP stack to a NIC but also allowing a caller to specify where the data should go in remote memory. Offloading the TCP/IP stack will be needed for current servers to push 10 Gb over TCP/IP. It also becomes a big deal for latency reduction and for iSCSI performance. It makes a big difference. Most of todays dual CPU Intel based boxes have trouble going too much over 1Gb. 10 Gb is out of the question without offloading. Using RDMA over a Send/Recv type model also give you more flexibility over interrupt rate. Finally, offloading the stack will include offloading segmentation, a big CPU hog. Although APIs exist for this today on some OSs, this API would also include it. InfiniBand does a lot of this already. Sockets Direct Protocol is pretty similar in the way it works. In fact, many of the folks that worked on SDP worked on this spec. MS is very into this and should be talking about it at this years WinHEC.
Actually, IB is up to 10 Gigabit. I have seen performanc at 800 Megabytes per second using MPI. I work over at InfiniCon Systems on InfiniBand related software. Interesting uses of this include database clustering and high performance computing clusters. Think 4000 node clusters. APIs include MPI for HPC stuff and DAPL for database clustering and RDMA file systems. You can use Sockets Direct Protocol to offload your TCP/IP traffic. IPoIB handles other IP traffic. There are also protocols for connecting to ethernet and fibre channel.
This is common issue in the embedded world and sometimes is the main reason linux isn't used. I have been in a situation where linux was considered and one of the reasons it lost out was that we felt the amount of real intellectual property we could put into it was limited. We make our systems from parts that other vendors could also buy and so our software really makes a big difference. In these cases, we felt that we had to use either a BSD based system or a proprietary one that allowed us rights to change the full source. We are currently with the proprietary model but the licensing charges are keeping us looking BSD again.
Wow...I had no idea it had risen so high. I wonder how that number changes world wide. More importantly for this issue, what precentage of the CD buying public use a computer for music? It is these users that are going to care, not someone who buys CDs but only uses his computer for e-mail and word processing.
I agree with you although I suspect that in the end most people will not care. The vast majority of people play music on standard CDs, cassettes, and yes even records. Digital rights only matter to PC owners/users and they make a small portion of the population. Just what percentage of people who buy music regularly (say more than 2 CDs a year) activly copy their music in any fasion. Most just are happy to play their CDs. As a technology lover this drives me crazy but I know it to be true. What really seems silly is how none of these companies realize that they are their own worst enemy with some of these issues. If CDs were priced reasonably (cost less than tapes to make but cost more than tapes for the consumer?) and online music was sold by them at a reasonable rate, they wouldn't have to worry about this. Instead they intend to blead the average person dry and block any attempt to spread the word that it doesn't have to be that way. I can't stand when companies choose to block innovation out of fear instead of embracing it.
Actually, Intel is still pushing IB. They sponsered a multicompany IB demo that is being taken around from show to show (IDF, SNW, NetInterop, etc). They discontinued their 1X products because everyone wanted 4X and other companies started to deliver 4X while all Intel had was plain old 1X. The ironic thing is that they hit the market so early that almost every IB company has some Intel IB equipment, especially HCAs. The party isn't over either. A lot of big companies (Dell, IBM, Microsoft, HP, and Sun) still have IB on their mind. My company is enjoying lots of interest in our IB products. 3GIO solves some problems but it doesn't provide for host to host communication, and I/O device sharing. These are two really big plusses for InfiniBand.
I used to write financial software in C++. We has objects for Interest Rates that handled conversions to and from various types (Annualized, Continuous, Bond Equivalent Yield, etc) and could work as normal double floating point values. We had a class hierarchy for option evaluation functions. At the base a parent class had the features common to all options and their evaluation formulas. Children classes then implmented specific evaluation formulas. Sometimes tweaks were needed to those formulas for specific situations and another level of inheritance was used. We found that the OO C++ code was much easier to maintain and add on to then the previous functional C version.
As other posters have said, to really use OO you have to break the problem down into objects. What are the features of these objects? What do they do? How do they interact? Another good idea is to see what certain objects have in common. This allows you push these common features and functions into a parent class that both objects inherit from.
Many companies are planning to move to IB based blades. Dell for one; they are calling them bricks. Here the blade is a standard IB form factor module. This lets vendors do some really nice things. Get rid of PCI for one. Next get rid of internal I/O (storage, ethernet). The blade uses the IB backplane to connect to the IB fabric and thus to other blades for IPC and to I/O modules for ethernet and storage connectivity. With speeds at 2.5 Gb/s, 10 Gb/s, and 30 Gb/s you can come up with some really nice clustering applications. And you get to use a standard that many companies are backing. Now the blade just houses processors, memory, and an InfiniBand Host Channel Adapter chip or two. Moving the I/O out leaves you a lot more room. You could probably fit 8 blades or so in 3U of space. And these blades can use top shelf I/O like Gb Ethernet and 2Gb Fibre Channel where most blades today are 100 Mb ethernet and IDE or SCSI.
I am curious about where clustering will go for the connection between nodes. Ethernet, Fibre Channel and various proprietary formats are around but all have issues. InfiniBand is also on the horizon. While I work with InfiniBand development, I am not involved with any kind of clustering work. I see Sockets over InfiniBand as interesting method for inter-node communication. What do those of you who do work in the field think?
Back before Compaq purchased DEC, Intel and DEC were in a lawsuit over Intel keeping a bunch of Alpha technical specs (including details about there technology to run 32-bit code) to help Intel (and HP) with their IA-64 work. Intel responded to DEC's lawsuit by threatening to withhold Intel processors from DEC and issuing a counter lawsuit. This put DEC up the creek without a paddle since they made the bulk of their money from Intel based systems. DEC responded by crying, "monopolist!" If I remember correctly, at one point it looked like the government was really going to get involved. It only got settled when Compaq swallowed DEC up. The lawsuits were dropped or settled out of court and the government's investigation of Intel quietly died. I don't even know if Intel even got a slap on the wrist.
So Compaq purchased DEC, sold StrongArm to Intel, sold AltaVist to CMGI, and is discontinuing Alpha and selling some of its IP to Intel. Why did they buy DEC again? It seems a lot of effort and money for DEC's services division.
This king of leads to the MS 3 rule; no MS product works at all well until version 3. MS can just throw more and more money at a problem until it goes away.
Very few other companies in the world could have afforded or would have wanted to keep MSN going. MS is different though. Once they attack a space, they just keep fighting (and spending). They will not allow defeat.
Take a look at embedded software. Another version of WinCE, 3.0 (aka PocketPC) is trying to push forward and staring to do better. At the same time, NT Embedded has finally spawned Windows XP Embedded (Win2k Embedded never made it out of the gate).
MSNBC keeps pushing forward. The mighty CNN (backed by AOLTime) is now struggling to fight off this (and Fox). Spend enough and keep trying and people watched.
Enterprise software? They have Win2k running on 32 processor intel based systems with 64 gigs of RAM. Exchange is all of the place. SQLServer use is growing.
As a business they do some many things to make sure they win. Every piece is tied to all the others. They tell you that "If you run windows and office at home, you should use a PocketPC! It runs all the same stuff!" They say that "If you run Win2k, than Exchange, SQLServer, and IIS run the best!" They want everything to tie to them. Your windows login becomes your "Passport". Now they own part of your identity. Pretty soon you will have to pay to use your own passport. Just a penny a login...
The fact is that they are just too damn good at this capitalist game. In order to protect the people and not stiffle innovation, the playing field has to be made a bit more fair. The government no longer seems up to the task. Our only hope is that MS's enemies gang up on them. Can even AOLTimeWarner, Sony, Sun, Oracle, and IBM combined beat them? I don't know. I sure hope so because I would have rather have a bunch of powerful companies in specific sectors than one all powerful company in all sectors of the economy.
I am not expecting/. to send investigative reporters to far off techie conventions and the like. I just want them to read through an article before they post it. I don't think I am asking too much here. This doesn't really require too much extra money. While I would love real technical editors, I understand that/. may not be able to afford it. I find it hard to believe they can't afford to read an article before they post it though.
What really bothers me is that often it seems that/. is trying to manipulate readers with outrageous headlines just to get people to read on. I can't stand when traditional news sources do this and/. is no different.
Perhaps I am just too pessimistic and/. doesn't do it on purpose. Perhaps it is just that their editors don't have the time or money to check stories before rushing them out the door. I just don't understand how someone can read a post enough to believe it is worth an AskSlashdot but not bother to read the article sourced in the post.
Don't get me wrong, I really like/. on the whole. I just think that it could be even better. The only way the powers that be will know that people want change is if they say it. So I am saying it.
First about this post. As others have commented, this only applies to Windows XP/2002 running on IA-64 processors. All operating systems wishing to boot on these systems will need changes to work with Intel's firmware for IA64 based systems.
This brings a fact about/. that keeps raising its ugly head. All too often stories are posted that are not checked up on. I understand that those who run/. can not be expected to be experts in everything but stories are often posted that only required a simple reading of a FAQ or two to see that the post is just plain wrong. I had high hopes of/. getting much more professional when it was purchased by Andover. But nothing has changed. The staff seems about the same size and quality that it was before the corporate move./. really has such great potential. If it could start using expert editors who only post stories that have been properly checked. One reason I love magazines like Discover and Popular Science because the stories are usually checked out pretty well. I understand the need to post things that generate interest but the site also claims to be a news site. No self respecting news site would allow such a low level of checking into the background of the stories they present. Is/. a news site at its heart or just entertainment? It has the potential to be both a high quality and entertaining news site and discussion forum. I understand that/. is a business and the object of a business is to make profits for their shareholders but I don't understand why this has to be in direct opposition to high quality.
While WinHEC is a "hardware" conference, they were lots of companies there that were not traditional hardware engineering companies. Not only did AOL not show up in large way themselves (they had a few people there as observers) they also were not mentioned at all. I was surprised to notice that while MS railed against companies like Sun when talking about Servers and WindRiver when talking about Embedded stuff, they failed to mention AOL when talking about Net services. I suspect MS is really a bit worried about AOLTimeWarner. They don't even have any kind of marketing ready to show how hailstorm riding on.NET will be a better choice than AOL. MSN seems to not have the ability to get even close to AOL. MS is now buying customers and services just to get what AOL signs up through simple word of mouth and cd distribution. MS has a two big fights on its hands: AOLTimeWarner in the net serverices space and Sony and Nintendo in the computer game console space. Wait, make that three because you can't forget the gonverment and many class actions in the legal space. Have fun MS!
A true sense of the confidence of AOL was shown some time ago when Case appeared on Charlie Rose. Rose got to asking Case about MS's offer to buy AOL or get crushed by Microsoft. Case tried to avoid the issue but Rose pushed. I paraphrase:
Rose: "Come on, you have to feel good about that decision [not to agree to an MS purchase].
Case (finally smiling): "Well...yes. We felt then and that it definitly was the right way to go. Today, I think it is fair to say that we were right."
I wouldn't want to be on Case's sh*t list...makes you almost feel bad for MS.
I find it interesting that much of this talk about embedded devices fails to mention i/o devices. Some may say that an i/o device's software should be written bare metal with little os at all. While this can be a great option for performance reasons, it requires a very talented staff and a lot of time. This may not be an option, especially for a first generation product from a young company. As the I/O workload is pushed away from the main CPU to add on cards, I/O devices become more complex. The choice of an embedded operating system becomes more and more important as developers look at the laundry list of features and performance requirements. Fibre Channel cards and NICs keep moving faster and faster. PCI 64-bit 66 Mhz will be yielding to PCI-X and double data rate PCI-X. InfiniBand 4x and 12X links will not only push data faster but create many new configuration and manageability issues for i/o devices. All these things make creating high quality i/o device very difficult. A report on what embedded OS's work best for things like Gigabit NICs, Fibre Channel RAID, Network Attached Storage, etc. would be very interesting. Why don't we see more of this? I guess it's no fun to report on something that doesn't have a pretty screen with a web browser on it.
I used to work at Unisys, writing device drivers. It is actually a really neat place in terms of technology. The first time I toured their lab I was amazed at all the different types of technology in there. They had some old mainframes that had uptimes in years. Products that were over a decade old and were still in use and still solved the problem they were purchased to solve. As mentioned by some others, the ES7000 and its CMP architecture is pretty cool. By the way, although it does run Windows 2000 Data Center Server (heck, the release of Win2kDCS waited for Unisys to complete various improvements on Win2kDCS), it also runs SCO Unix. There was talk of other ports in the future. It is neat to have a windows/unix machine that supports hot swap of nearly everything (memory, processors, i/o devices) and dynamic partitioning. Can't be that bad when you consider that Compaq, Dell, and HP all have decided to OEM it rather than build a box of their own in that class. Plus it comes with IntelliFIBRE, a FibreChannel card that I wrote device drivers for! (Yeah, I am bit biased there...but who isn't about their work?) Making big computers, software for them, and services wrapped around all of that may not be sexy, but is profitable. Something that all those.coms can't claim. Still had $6.8 billion in revenue with $334 million in profit for 2000. Not too shabby for a bad year in tech.
NTFS and I believe FAT 32 (don't know about FAT 16) partitions retain memory of what their letter was. Win2k and NT4 will simply try that letter, if it is in use it will move up until it runs out. This keeps applications in line that reference the c: style notation. The problem is that c: is really just a symbolic link to a disk number and partition number. This link is not a UNIX one that is made through the file system but made at the device driver level.
The problem occurs when some software does not use a file system and does direct disk access. Some databases do this for example. This type of software usually does not use a drive lettered partition. It relies on disk and partition numbers and may or may not get broken when the numbering changes. Win2k disk and partition numbering is simpy based on discovery order. This in turn is based on the order in which drivers are loaded and where hardware is located on PCI and storage buses (SCSI/FibreChannel/FireWire). Put a new FibreChannel card in front of your old scsi and suddenly the database doesn't work. Write a special driver for you superduper disks that loads before the standard win2k driver and you may screw things up.
As others of noted Win2k supports mounting now. This lets you just have C as a sort of "/".
As for driver letters on firewire things should go like this in the win2k world:
PCI (if firewire pci card) or root bus driver (built into mother board) finds firewire bus, loads firewire bus driver.
Standard firewire bus driver finds device.
An additional driver or two will load some or all of which would be vendor specific. It could run like the standard scsi port driver/ vendor scsi miniport combonation. It would present an upword interface the same as presented by the normal scsi port/miniport combo but send firewire out the bottom.
This would let the standard pnp disk, partition, and file system drivers load on top.
They would talk scsi ops down the stack where the firewire scsi port/miniport would translate and encapsulate as needed and send it out, using the Firewire bus driver. Numbering occurs as things are discovered. Letters are assigned to partitions trying to use what was stored, moving up letters if a conflict occurs.
In the fc/scsi world, you can send reserve, release, and reset signals. Server 1 goes down, server 2 trys reserve, if server 1 died too fast to release, reserve fails, server 2 resets ownership (blows reserve away), server 2 reserves, server 2 up and running with storage. I am not sure how firewire storage works (are the ops scsi like, the same fc ops are?), but I would geuss you would have something similar. Not it just depends if the host OS supports an application sending this ops accross. Then a cluster manager application can handle it. Perhaps the cluster manager needs to run at or in conjunction with some clustering device drivers but you get the picture. Should work...
If things are getting easier, then why does it seem that code is getting worse and more bugs are turning up? Things are still hard to do. As we figure out how to do one thing easily, another problem comes along. Saying that computer science is easier due to advancements is like saying any other science is easier. Physics has come a long way in 50 years but is it easier? Some would say it's a lot harder.
Someday, sentient computer intelligences may be able to write their own code. Someday humans may be able to do the same with genetics. Those days are far off, though so until then programming will be tricky, pointers will be a pain, and schools will have far less graduating CS majors than freshmen ones.
I work in embedded systems (firmware, device drivers, etc.). This world is almost unaffected by the DotCom cruch. I think it is for exactly the reasons you state. This area requires a real engineering background. Experience in it is not easy to come by in the begining and people don't always stick with it. This keeps the pool of workers small, while the demand remains as high as ever. Sometimes, the lower level you go, the better you do. Nothing worth doing is easy and nothing easy is worth paying a lot for.
I think it will help make computer science (and other similar) degrees more important. It will also increase the value of experience. Companies will now be a lot more carefull on who they hire. They are going to want to see more proof that the prospective developer can get the job done well. Things like college GPA (especially in you major) and performance at previous jobs will matter. Interviews will become more technical.
In the end it will probably be better for people that are good at what they do. The industry will see them as more important, find that they are rare, and be willing to pay even more for them. Although companies are going out of business, there are new opportunities elsewhere. DotComs may be dying off but all software development isn't. Code doesn't write itself and it will not for a long, long time. There are still problems that people want solved that require someone getting some kind of computer hardware to do things it wouldn't do all on its own.
Slashdot Mirror
I have seen dell 2650s hit over 800 Megabytes (6.4 Gb) per second running MPI over InfiniBand using large buffer sizes. The limit is pretty much the PCI-X 133 Mhz interface we are on. I suspect that with PCI-X DDR and PCI Express, we will be able to get a lot closer to 10 Gbit.
Really? Seems odd that my company keeps selling it. Sun is using it in their next systems. Network Appliance is using it today. Intel is selling systems with it. Chips are made by Mellanox, Agilent, and Fujitsu. Sun and IBM are doing chips for internal use. There are still a good number of IB vendors as well. See InfiniCon Systems (where I work), TopSpin, InfiniSwitch, Voltaire, and JNI. There are also a bunch of folks using IB in embedded environments.
The high performance computing market and database clustering market are really interested in IB. 10 Gb ethernet is many times more expensive and without a TOE that can do 10 Gb, no one can really use it except for switch to switch. 10 Gb fibre channel isn't quite out yet.
Also, many companies are still paying people to spend a big chunck of their time working on InfiniBand Trade Association Working Group issues.
Finally, none of the technologies you mentioned solve clustering issues because they are for inside the box, not from box to box.
I work at InfiniCon Systems and we do a lot of InfiniBand related things. We have switching, adapters, and connections from IB to gigabit ethernet and IB to fibre channel. See http://www.infinicon.com/.
The real next steps for IB is 12X (30 Gb) and on mother board IB. 12 X is in development. Currently, IB adapters are limited by the PCI-X slot they sit in. PCI-X DDR and PCI Express should help, but just having it on the mother board and throwing PCI out would be interesting. Small form factor clusters would really benefit to just having memory, CPU, and an HCA.
Not exactly. Most NICs today due DMA without CPU assistance. This talks about writing to another host's memory. Much of it is about removing the TCPI/IP stack to a NIC but also allowing a caller to specify where the data should go in remote memory.
Offloading the TCP/IP stack will be needed for current servers to push 10 Gb over TCP/IP. It also becomes a big deal for latency reduction and for iSCSI performance. It makes a big difference. Most of todays dual CPU Intel based boxes have trouble going too much over 1Gb. 10 Gb is out of the question without offloading. Using RDMA over a Send/Recv type model also give you more flexibility over interrupt rate. Finally, offloading the stack will include offloading segmentation, a big CPU hog. Although APIs exist for this today on some OSs, this API would also include it.
InfiniBand does a lot of this already. Sockets Direct Protocol is pretty similar in the way it works. In fact, many of the folks that worked on SDP worked on this spec. MS is very into this and should be talking about it at this years WinHEC.
Actually, IB is up to 10 Gigabit. I have seen performanc at 800 Megabytes per second using MPI. I work over at InfiniCon Systems on InfiniBand related software. Interesting uses of this include database clustering and high performance computing clusters. Think 4000 node clusters. APIs include MPI for HPC stuff and DAPL for database clustering and RDMA file systems. You can use Sockets Direct Protocol to offload your TCP/IP traffic. IPoIB handles other IP traffic. There are also protocols for connecting to ethernet and fibre channel.
Sun acquired Pirus Networks to help them on a chassis with FibreChannel, iSCSI, and perhaps InfiniBand.2 1423
0 919S0076
http://www.byteandswitch.com/document.asp?doc_id=
Before that they picked up Dolphin Interconnect to help them make a 4x (30 Gigabit/sec) InfiniBand Host Channel Adapter.
Here is an article from an EETimes Network site, CommsDesign with some details.
http://www.commsdesign.com/news/tech_beat/OEG2002
It is definitly interesting stuff. Everyone is trying to do Shared I/O and I/O Virtualization; maybe Sun can get it right.
This is common issue in the embedded world and sometimes is the main reason linux isn't used. I have been in a situation where linux was considered and one of the reasons it lost out was that we felt the amount of real intellectual property we could put into it was limited. We make our systems from parts that other vendors could also buy and so our software really makes a big difference. In these cases, we felt that we had to use either a BSD based system or a proprietary one that allowed us rights to change the full source. We are currently with the proprietary model but the licensing charges are keeping us looking BSD again.
Wow...I had no idea it had risen so high. I wonder how that number changes world wide. More importantly for this issue, what precentage of the CD buying public use a computer for music? It is these users that are going to care, not someone who buys CDs but only uses his computer for e-mail and word processing.
I agree with you although I suspect that in the end most people will not care. The vast majority of people play music on standard CDs, cassettes, and yes even records. Digital rights only matter to PC owners/users and they make a small portion of the population. Just what percentage of people who buy music regularly (say more than 2 CDs a year) activly copy their music in any fasion. Most just are happy to play their CDs. As a technology lover this drives me crazy but I know it to be true.
What really seems silly is how none of these companies realize that they are their own worst enemy with some of these issues. If CDs were priced reasonably (cost less than tapes to make but cost more than tapes for the consumer?) and online music was sold by them at a reasonable rate, they wouldn't have to worry about this. Instead they intend to blead the average person dry and block any attempt to spread the word that it doesn't have to be that way. I can't stand when companies choose to block innovation out of fear instead of embracing it.
Actually, Intel is still pushing IB. They sponsered a multicompany IB demo that is being taken around from show to show (IDF, SNW, NetInterop, etc). They discontinued their 1X products because everyone wanted 4X and other companies started to deliver 4X while all Intel had was plain old 1X. The ironic thing is that they hit the market so early that almost every IB company has some Intel IB equipment, especially HCAs.
The party isn't over either. A lot of big companies (Dell, IBM, Microsoft, HP, and Sun) still have IB on their mind. My company is enjoying lots of interest in our IB products. 3GIO solves some problems but it doesn't provide for host to host communication, and I/O device sharing. These are two really big plusses for InfiniBand.
I used to write financial software in C++. We has objects for Interest Rates that handled conversions to and from various types (Annualized, Continuous, Bond Equivalent Yield, etc) and could work as normal double floating point values. We had a class hierarchy for option evaluation functions. At the base a parent class had the features common to all options and their evaluation formulas. Children classes then implmented specific evaluation formulas. Sometimes tweaks were needed to those formulas for specific situations and another level of inheritance was used. We found that the OO C++ code was much easier to maintain and add on to then the previous functional C version.
As other posters have said, to really use OO you have to break the problem down into objects. What are the features of these objects? What do they do? How do they interact? Another good idea is to see what certain objects have in common. This allows you push these common features and functions into a parent class that both objects inherit from.
Many companies are planning to move to IB based blades. Dell for one; they are calling them bricks. Here the blade is a standard IB form factor module. This lets vendors do some really nice things. Get rid of PCI for one. Next get rid of internal I/O (storage, ethernet). The blade uses the IB backplane to connect to the IB fabric and thus to other blades for IPC and to I/O modules for ethernet and storage connectivity. With speeds at 2.5 Gb/s, 10 Gb/s, and 30 Gb/s you can come up with some really nice clustering applications. And you get to use a standard that many companies are backing. Now the blade just houses processors, memory, and an InfiniBand Host Channel Adapter chip or two. Moving the I/O out leaves you a lot more room. You could probably fit 8 blades or so in 3U of space. And these blades can use top shelf I/O like Gb Ethernet and 2Gb Fibre Channel where most blades today are 100 Mb ethernet and IDE or SCSI.
I am curious about where clustering will go for the connection between nodes. Ethernet, Fibre Channel and various proprietary formats are around but all have issues. InfiniBand is also on the horizon. While I work with InfiniBand development, I am not involved with any kind of clustering work. I see Sockets over InfiniBand as interesting method for inter-node communication. What do those of you who do work in the field think?
Back before Compaq purchased DEC, Intel and DEC were in a lawsuit over Intel keeping a bunch of Alpha technical specs (including details about there technology to run 32-bit code) to help Intel (and HP) with their IA-64 work. Intel responded to DEC's lawsuit by threatening to withhold Intel processors from DEC and issuing a counter lawsuit. This put DEC up the creek without a paddle since they made the bulk of their money from Intel based systems. DEC responded by crying, "monopolist!" If I remember correctly, at one point it looked like the government was really going to get involved. It only got settled when Compaq swallowed DEC up. The lawsuits were dropped or settled out of court and the government's investigation of Intel quietly died. I don't even know if Intel even got a slap on the wrist.
So Compaq purchased DEC, sold StrongArm to Intel, sold AltaVist to CMGI, and is discontinuing Alpha and selling some of its IP to Intel. Why did they buy DEC again? It seems a lot of effort and money for DEC's services division.
This king of leads to the MS 3 rule; no MS product works at all well until version 3. MS can just throw more and more money at a problem until it goes away.
Very few other companies in the world could have afforded or would have wanted to keep MSN going. MS is different though. Once they attack a space, they just keep fighting (and spending). They will not allow defeat.
Take a look at embedded software. Another version of WinCE, 3.0 (aka PocketPC) is trying to push forward and staring to do better. At the same time, NT Embedded has finally spawned Windows XP Embedded (Win2k Embedded never made it out of the gate).
MSNBC keeps pushing forward. The mighty CNN (backed by AOLTime) is now struggling to fight off this (and Fox). Spend enough and keep trying and people watched.
Enterprise software? They have Win2k running on 32 processor intel based systems with 64 gigs of RAM. Exchange is all of the place. SQLServer use is growing.
As a business they do some many things to make sure they win. Every piece is tied to all the others. They tell you that "If you run windows and office at home, you should use a PocketPC! It runs all the same stuff!" They say that "If you run Win2k, than Exchange, SQLServer, and IIS run the best!" They want everything to tie to them. Your windows login becomes your "Passport". Now they own part of your identity. Pretty soon you will have to pay to use your own passport. Just a penny a login...
The fact is that they are just too damn good at this capitalist game. In order to protect the people and not stiffle innovation, the playing field has to be made a bit more fair. The government no longer seems up to the task. Our only hope is that MS's enemies gang up on them. Can even AOLTimeWarner, Sony, Sun, Oracle, and IBM combined beat them? I don't know. I sure hope so because I would have rather have a bunch of powerful companies in specific sectors than one all powerful company in all sectors of the economy.
I am not expecting /. to send investigative reporters to far off techie conventions and the like. I just want them to read through an article before they post it. I don't think I am asking too much here. This doesn't really require too much extra money. While I would love real technical editors, I understand that /. may not be able to afford it. I find it hard to believe they can't afford to read an article before they post it though.
/. is trying to manipulate readers with outrageous headlines just to get people to read on. I can't stand when traditional news sources do this and /. is no different.
/. doesn't do it on purpose. Perhaps it is just that their editors don't have the time or money to check stories before rushing them out the door. I just don't understand how someone can read a post enough to believe it is worth an AskSlashdot but not bother to read the article sourced in the post.
/. on the whole. I just think that it could be even better. The only way the powers that be will know that people want change is if they say it. So I am saying it.
What really bothers me is that often it seems that
Perhaps I am just too pessimistic and
Don't get me wrong, I really like
First about this post. As others have commented, this only applies to Windows XP/2002 running on IA-64 processors. All operating systems wishing to boot on these systems will need changes to work with Intel's firmware for IA64 based systems. /. that keeps raising its ugly head. All too often stories are posted that are not checked up on. I understand that those who run /. can not be expected to be experts in everything but stories are often posted that only required a simple reading of a FAQ or two to see that the post is just plain wrong. I had high hopes of /. getting much more professional when it was purchased by Andover. But nothing has changed. The staff seems about the same size and quality that it was before the corporate move. /. really has such great potential. If it could start using expert editors who only post stories that have been properly checked. One reason I love magazines like Discover and Popular Science because the stories are usually checked out pretty well. I understand the need to post things that generate interest but the site also claims to be a news site. No self respecting news site would allow such a low level of checking into the background of the stories they present. Is /. a news site at its heart or just entertainment? It has the potential to be both a high quality and entertaining news site and discussion forum. I understand that /. is a business and the object of a business is to make profits for their shareholders but I don't understand why this has to be in direct opposition to high quality.
/.er who hopes things get better...
This brings a fact about
from a
While WinHEC is a "hardware" conference, they were lots of companies there that were not traditional hardware engineering companies. Not only did AOL not show up in large way themselves (they had a few people there as observers) they also were not mentioned at all. I was surprised to notice that while MS railed against companies like Sun when talking about Servers and WindRiver when talking about Embedded stuff, they failed to mention AOL when talking about Net services. I suspect MS is really a bit worried about AOLTimeWarner. They don't even have any kind of marketing ready to show how hailstorm riding on .NET will be a better choice than AOL. MSN seems to not have the ability to get even close to AOL. MS is now buying customers and services just to get what AOL signs up through simple word of mouth and cd distribution. MS has a two big fights on its hands: AOLTimeWarner in the net serverices space and Sony and Nintendo in the computer game console space. Wait, make that three because you can't forget the gonverment and many class actions in the legal space. Have fun MS!
A true sense of the confidence of AOL was shown some time ago when Case appeared on Charlie Rose. Rose got to asking Case about MS's offer to buy AOL or get crushed by Microsoft. Case tried to avoid the issue but Rose pushed. I paraphrase:
Rose: "Come on, you have to feel good about that decision [not to agree to an MS purchase].
Case (finally smiling): "Well...yes. We felt then and that it definitly was the right way to go. Today, I think it is fair to say that we were right."
I wouldn't want to be on Case's sh*t list...makes you almost feel bad for MS.
I find it interesting that much of this talk about embedded devices fails to mention i/o devices. Some may say that an i/o device's software should be written bare metal with little os at all. While this can be a great option for performance reasons, it requires a very talented staff and a lot of time. This may not be an option, especially for a first generation product from a young company. As the I/O workload is pushed away from the main CPU to add on cards, I/O devices become more complex. The choice of an embedded operating system becomes more and more important as developers look at the laundry list of features and performance requirements. Fibre Channel cards and NICs keep moving faster and faster. PCI 64-bit 66 Mhz will be yielding to PCI-X and double data rate PCI-X. InfiniBand 4x and 12X links will not only push data faster but create many new configuration and manageability issues for i/o devices. All these things make creating high quality i/o device very difficult. A report on what embedded OS's work best for things like Gigabit NICs, Fibre Channel RAID, Network Attached Storage, etc. would be very interesting. Why don't we see more of this? I guess it's no fun to report on something that doesn't have a pretty screen with a web browser on it.
I used to work at Unisys, writing device drivers. It is actually a really neat place in terms of technology. The first time I toured their lab I was amazed at all the different types of technology in there. They had some old mainframes that had uptimes in years. Products that were over a decade old and were still in use and still solved the problem they were purchased to solve. As mentioned by some others, the ES7000 and its CMP architecture is pretty cool. By the way, although it does run Windows 2000 Data Center Server (heck, the release of Win2kDCS waited for Unisys to complete various improvements on Win2kDCS), it also runs SCO Unix. There was talk of other ports in the future. It is neat to have a windows/unix machine that supports hot swap of nearly everything (memory, processors, i/o devices) and dynamic partitioning. Can't be that bad when you consider that Compaq, Dell, and HP all have decided to OEM it rather than build a box of their own in that class. Plus it comes with IntelliFIBRE, a FibreChannel card that I wrote device drivers for! (Yeah, I am bit biased there...but who isn't about their work?) Making big computers, software for them, and services wrapped around all of that may not be sexy, but is profitable. Something that all those .coms can't claim. Still had $6.8 billion in revenue with $334 million in profit for 2000. Not too shabby for a bad year in tech.
NTFS and I believe FAT 32 (don't know about FAT 16) partitions retain memory of what their letter was. Win2k and NT4 will simply try that letter, if it is in use it will move up until it runs out. This keeps applications in line that reference the c: style notation. The problem is that c: is really just a symbolic link to a disk number and partition number. This link is not a UNIX one that is made through the file system but made at the device driver level.
The problem occurs when some software does not use a file system and does direct disk access. Some databases do this for example. This type of software usually does not use a drive lettered partition. It relies on disk and partition numbers and may or may not get broken when the numbering changes. Win2k disk and partition numbering is simpy based on discovery order. This in turn is based on the order in which drivers are loaded and where hardware is located on PCI and storage buses (SCSI/FibreChannel/FireWire). Put a new FibreChannel card in front of your old scsi and suddenly the database doesn't work. Write a special driver for you superduper disks that loads before the standard win2k driver and you may screw things up.
As others of noted Win2k supports mounting now. This lets you just have C as a sort of "/".
As for driver letters on firewire things should go like this in the win2k world:
PCI (if firewire pci card) or root bus driver (built into mother board) finds firewire bus, loads firewire bus driver.
Standard firewire bus driver finds device.
An additional driver or two will load some or all of which would be vendor specific. It could run like the standard scsi port driver/ vendor scsi miniport combonation. It would present an upword interface the same as presented by the normal scsi port/miniport combo but send firewire out the bottom.
This would let the standard pnp disk, partition, and file system drivers load on top.
They would talk scsi ops down the stack where the firewire scsi port/miniport would translate and encapsulate as needed and send it out, using the Firewire bus driver. Numbering occurs as things are discovered. Letters are assigned to partitions trying to use what was stored, moving up letters if a conflict occurs.
In the fc/scsi world, you can send reserve, release, and reset signals. Server 1 goes down, server 2 trys reserve, if server 1 died too fast to release, reserve fails, server 2 resets ownership (blows reserve away), server 2 reserves, server 2 up and running with storage. I am not sure how firewire storage works (are the ops scsi like, the same fc ops are?), but I would geuss you would have something similar. Not it just depends if the host OS supports an application sending this ops accross. Then a cluster manager application can handle it. Perhaps the cluster manager needs to run at or in conjunction with some clustering device drivers but you get the picture. Should work...
If things are getting easier, then why does it seem that code is getting worse and more bugs are turning up? Things are still hard to do. As we figure out how to do one thing easily, another problem comes along. Saying that computer science is easier due to advancements is like saying any other science is easier. Physics has come a long way in 50 years but is it easier? Some would say it's a lot harder.
Someday, sentient computer intelligences may be able to write their own code. Someday humans may be able to do the same with genetics. Those days are far off, though so until then programming will be tricky, pointers will be a pain, and schools will have far less graduating CS majors than freshmen ones.
I work in embedded systems (firmware, device drivers, etc.). This world is almost unaffected by the DotCom cruch. I think it is for exactly the reasons you state. This area requires a real engineering background. Experience in it is not easy to come by in the begining and people don't always stick with it. This keeps the pool of workers small, while the demand remains as high as ever. Sometimes, the lower level you go, the better you do. Nothing worth doing is easy and nothing easy is worth paying a lot for.
I think it will help make computer science (and other similar) degrees more important. It will also increase the value of experience. Companies will now be a lot more carefull on who they hire. They are going to want to see more proof that the prospective developer can get the job done well. Things like college GPA (especially in you major) and performance at previous jobs will matter. Interviews will become more technical.
In the end it will probably be better for people that are good at what they do. The industry will see them as more important, find that they are rare, and be willing to pay even more for them. Although companies are going out of business, there are new opportunities elsewhere. DotComs may be dying off but all software development isn't. Code doesn't write itself and it will not for a long, long time. There are still problems that people want solved that require someone getting some kind of computer hardware to do things it wouldn't do all on its own.