OS/2 advocates have been bringing up this ATM thing more or less forever. I've never understood how that is supposed to have anything to do with me as a desktop PC user.
I understand what that suggests about its reliability and stability, but that doesn't have anything to do with IBM's desire to get rid of it as a desktop operating system.
I've seen a number of messages suggesting that Sun is shooting itself in the foot with SCSL, since the 'bazaar' model really needs "Open Source" to work.
All these posters seems to be misinterpreting what Sun hopes to gain through this. They aren't hoping that some huge assortment of hackers are going to leap in and do massive amounts of development. They fully expect to continue doing the vast majority of new development in StarOffice, Java, Jini, ClusterTools, etc. Since they aren't really trying for bazaar-style development, they don't need a bazaar-facilitating license.
Once you understand that Sun isn't trying to offload their development work on the hacker community, SCSL seems like a pretty good compromise between Sun's interests and the end users'.
Good things in SCSL for end users:
The source is available.
You can fix bugs that are more important to you than to Sun. Sure, they would probably fix your pet bug someday, but they might have better-paying customers who want other bugs fixed first.
You can add features that Sun isn't interested in, or isn't interested enough in.
You can port the software to non-Sun hardware.
Good things in SCSL for Sun (since no company does this stuff out of the goodness in their hearts):
They get bug fixes done more quickly than if they had to do all the work themselves.
They get software ported to other machines that they may not have the resources to support. This seems like an odd benefit for a hardware company, but Java and Jini become stronger the more widely ported they are. The benefit to seeing ClusterTools and StarOffice ported isn't as clear.
They get good press.
They can stick it to Microsoft. Since Sun makes its money on hardware, they have less to lose than Microsoft, which has nothing to sell but the secrets in their source.
Since SCSL is more restricted than GPL, BSD, etc, they can prevent Microsoft from co-opting the source, adding a whizzy new MS-specific feature, and becoming the de facto 'standard'.
Oh yeah - maybe somebody will implement a new feature or two.
It's like the flat earthers, who continue to insist the earth is not round.
This is one of the funnier lines in the article, since it so clearly applies to the writer rather than the folks he disagrees with.
In a couple of places, he refers to SGI being profitable. According to the SEC, SGI hasn't posted an annual profit in years. In the latest quarterly report, they posted another huge loss. They have 'announced' a $22 million profit for the latest quarter, but I'll reserve judgement until the 10Q is filed.
They are on their third CEO in three years.
They are on their third 'strategy' in three years.
They are selling off their NT and Cray divisions. The last time they did this, they sold the "Business Systems Division" piece of Cray to Sun for a pittance. Sun used that division to create the Starfire (actually, it was almost finished at SGI), which has already sold more than 1500 units, for a total of (guessing) more than $1.5 billion.
Not only are they losing money every quarter, their top line (ie, sales) have been shrinking for three years.
The value of their stock has dropped by about 50% in the last 9 months.
They just laid off more than 15% of their workforce.
How much do you need before acknowledging that a once-great company is in real trouble?
I don't know whether they used the word, but they did talk about the issue:
In applications such as electronic commerce, where uptime is measured in months and years as opposed to days and weeks, Solaris 7 should be the first [NOS] that administrators turn to. [...] To put things in perspective, while NetWare 5.0 and the forthcoming Windows 2000 have the capability to hot-swap PCI boards, Solaris supports hot-swapping of processors on larger SPARC systems.
If Linux can adopt Sun's DR process on Sun hardware, it would be one hell of a trophy to run four seperate instances of Linux on a single E4500.
I wouldn't be surpised if the hardware wouldn't support this.
In any case, I think you're getting a little ahead of yourself. If the hardware does support it, and if people actually want to do it, don't you think it's likely that Sun could get it done first? They've already got all the infrastructure in place, from the E10k work. Linux would be starting from scratch.
how is this better than running 12 completely separate machines?
Performance and ease-of-use. By having them all in the same box, you get to use shared memory instead of network connections for inter-process communication. In many cases, you can avoid inter-process communication altogether - just spawn multiple threads in the same process.
The cool thing about the multiple domains, is that you can repartition them more-or-less on the fly. So, if you have a board that is acting flaky, you can remove it from the main domain for testing. While you're testing, you can reboot just that board, you can run a different version of the OS, you can add and remove devices from it, etc. When you have isolated and fixed the problem, you can add it back into the main domain.
Sun's future clustering solutions include Beowulf type clusters, as they are going for the HPC market as well.
Calling this a Beowulf-type cluster is a little silly. People were putting these things together for years before the Beowuld project started. That's like calling all browsers "Internet Explorer type applications".
The system header files are there for all the 'system' libraries (libc.so, libadm.so, libaio.so, libcurses.so, libm.so, etc). I think the static libraries are there as well - although their use is discouraged. I don't see any 'debug' libraries anywhere - either in/usr/lib, or in the compiler package.
I think the only libraries that don't come with the machine are optimized math libraries (libsunperf.so, for example). Those have a whole separate API and everything, so it's not something you're likely to miss.
I believe a Beowulf could be constructed that could beat an E10000 in raw calculating power for a fraction of the cost.
Sure. Just rack up a couple dozen 4-way P2s. The problem is that 'raw calculating power' is rarely enough. Unless you have an embarrassingly parallel application, communication costs will kill you. Remember that the interprocess communication on a SMP is orders of magnitude faster than between nodes.
Also, if you have an autoparallelizing compiler, you can turn a single-processor sequential code into a multi-processor multithreaded code with a simple command-line option. This can be a huge win for some codes - probably more 'real world' codes than are embarrassingly parallel. This way, you don't need to learn a new programming model - just recompile your old codes.
What follows is strictly hearsay, and may be totally incorrect.
My understanding is that the powers that be decided that the compiler group was not delivering a competitive product. So, they decided to force them to compete with the Portland Group, and others. In essence, they had to start paying their own way - by delivering a product that people would pay for. Maybe it's a coincidence, but since they unbundled the compiler, it has been improving by leaps and bounds. For people who don't want to pay for a compiler, gcc and gdb both work fine with Solaris.
As for the libraries, I'm not sure what you're talking about. They ship with the OS - otherwise even 'cp' wouldn't work.
For the thousand people in the insurance company who don't need full powered pc's, just give them a machine with a p200, a NIC and a 4 gig HD. How much would that cost 200-300?
Hardware is cheap. What's expensive is training the users, maintaining the machine, and downtime when the machine is unavailable due to crashes or upgrades. For what they are doing, there is no reason to have something as complicated as Windows or Linux running.
FWIW, here's here's the web page for free Solaris 7. It looks like StarOffice is also being given away for it.
Though they should make more of the OS available if they want to appeal to CS geeks. I mean, Solaris isn't nearly the educational tool that Linux is because the implementation isn't freely available to look at and tinker with.
Really? I thought the source was available to people at.edu sites. I remember a bunch of hullaballoo when they made that anouncement for Solaris 2.6. I'll bet it's still available.
I think he meant that Sun would not be selling support contracts for Linux - unlike SGI and HP. Just as a point of interest, if you look at the Unix vendors who are selling support for Linux, I think you'll see that all of them also started selling support for NT a few years back. Make of that what you will.
Hmmm. They should, it would show off the machines far better than Solaris.
Not to put too fine a point on it: you have no idea what you're talking about.
Linux will run on small-scale SMP boxes. Currently, Sun's biggest machine has 64 CPUs. Even if Linux can boot on an E10000, trying to compare Linux and Solaris performance would be hugely embarrassing to Linux. Sun has a ton of engineers optimizing Solaris for these huge SMP boxes. How many Linux kernel hackers even have access to an SMP 1/8th this size?
Sun storage is now mainly fiber channel based. Their 14-disk A5000 array can deliver 140 MB/s (using two fiber loops), using Veritas Volume Manager to manage the individual disks. Does Linux support fiber channel at all? Does Veritas run on Linux?
Linux is great, but it's not the answer to everything. Again, to be blunt, it's Linux "advocates" like you who make the rest of us look bad.
Everything he said was taken out of context. That doesn't mean that it's false, but I for one would love to know what he meant by "a great way to get to the wrong answer"...
Actually, he said that Linux was "the right way to get the wrong answer". He meant that supporting open interfaces was absolutely the Right Thing to do. He was less convinced about open implementation, but he was at least warm to the idea.
By the "wrong answer" he meant that the Linux mindset is to have fat, powerful clients. He believes that the right model is to have fat, powerful servers, and thin clients.
Several times he made analogies to the phone system. Your phone is a thin client, connecting to fat servers. You don't buy software for your phone, the phone company supplies features. Etcetera.
Most of it wouldn't apply to the way your average/. reader uses computers, but it does make sense when you start thinking about the throngs of less sophisticated users out there.
How do you make the internet accessible to those people who can't program their VCRs? Asking them to install and configure Windows or Linux isn't going to cut it. For those people, a WebTV-like solution probably makes sense.
If you have a thousand people in an insurance company doing little but data entry and retrieval, does it really make sense to put a full powered PC or workstation on each desk? It would be a hell of a lot cheaper and more manageable to have some thin, stupid clients communicating with the fat servers in the back room.
Assuming that the right answer is Sun's Slowlaris? Maybe for multiprocessor boxes, but definitely not for the price.
The price? Solaris 7 was available for just the cost of the media. I don't know if that's still true or not.
Sun makes its money on hardware. They don't care whether the customer is running Linux or Solaris, as long as they are doing it on Sun hardware.
Why hasn't Sun done this?
on
Java for EGCS
·
· Score: 1
a) If you really believe that Java is "so unappealing to everyone", I expect you haven't been paying attention.
b) The concept of a VM is what makes Java very appealing to _many_ people. Even though WORA is still largely a pipe dream for complicated apps, a lot of people are still working towards it.
Sun's journalling file system is not from Veritas (although VxFS does run on Solaris). It's an extension of UFS. I could be wrong, but I think the source to all of Solaris (including logging UFS) is available to educational customers.
I've seen mentions in ZDnews and elsewhere that all of Solaris is going to be released under Sun's Community Source License (the Java/Jini license) in the near future.
Slashdot Mirror
I understand what that suggests about its reliability and stability, but that doesn't have anything to do with IBM's desire to get rid of it as a desktop operating system.
All these posters seems to be misinterpreting what Sun hopes to gain through this. They aren't hoping that some huge assortment of hackers are going to leap in and do massive amounts of development. They fully expect to continue doing the vast majority of new development in StarOffice, Java, Jini, ClusterTools, etc. Since they aren't really trying for bazaar-style development, they don't need a bazaar-facilitating license.
Once you understand that Sun isn't trying to offload their development work on the hacker community, SCSL seems like a pretty good compromise between Sun's interests and the end users'.
Good things in SCSL for end users:
- The source is available.
- You can fix bugs that are more important to you than to Sun. Sure, they would probably fix your pet bug someday, but they might have better-paying customers who want other bugs fixed first.
- You can add features that Sun isn't interested in, or isn't interested enough in.
- You can port the software to non-Sun hardware.
Good things in SCSL for Sun (since no company does this stuff out of the goodness in their hearts):This is one of the funnier lines in the article, since it so clearly applies to the writer rather than the folks he disagrees with.
In a couple of places, he refers to SGI being profitable. According to the SEC, SGI hasn't posted an annual profit in years. In the latest quarterly report, they posted another huge loss. They have 'announced' a $22 million profit for the latest quarter, but I'll reserve judgement until the 10Q is filed.
They are on their third CEO in three years.
They are on their third 'strategy' in three years.
They are selling off their NT and Cray divisions. The last time they did this, they sold the "Business Systems Division" piece of Cray to Sun for a pittance. Sun used that division to create the Starfire (actually, it was almost finished at SGI), which has already sold more than 1500 units, for a total of (guessing) more than $1.5 billion.
Not only are they losing money every quarter, their top line (ie, sales) have been shrinking for three years.
The value of their stock has dropped by about 50% in the last 9 months.
They just laid off more than 15% of their workforce.
How much do you need before acknowledging that a once-great company is in real trouble?
What are you, nuts? If anything there is too much you can configure in Solaris. For starters:
A collection of tuning papers and resources
In applications such as electronic commerce, where uptime is measured in months and years as opposed to days and weeks, Solaris 7 should be the first [NOS] that administrators turn to. [...] To put things in perspective, while NetWare 5.0 and the forthcoming Windows 2000 have the capability to hot-swap PCI boards, Solaris supports hot-swapping of processors on larger SPARC systems.
I wouldn't be surpised if the hardware wouldn't support this.
In any case, I think you're getting a little ahead of yourself. If the hardware does support it, and if people actually want to do it, don't you think it's likely that Sun could get it done first? They've already got all the infrastructure in place, from the E10k work. Linux would be starting from scratch.
Performance and ease-of-use. By having them all in the same box, you get to use shared memory instead of network connections for inter-process communication. In many cases, you can avoid inter-process communication altogether - just spawn multiple threads in the same process.
The cool thing about the multiple domains, is that you can repartition them more-or-less on the fly. So, if you have a board that is acting flaky, you can remove it from the main domain for testing. While you're testing, you can reboot just that board, you can run a different version of the OS, you can add and remove devices from it, etc. When you have isolated and fixed the problem, you can add it back into the main domain.
Calling this a Beowulf-type cluster is a little silly. People were putting these things together for years before the Beowuld project started. That's like calling all browsers "Internet Explorer type applications".
I think the only libraries that don't come with the machine are optimized math libraries (libsunperf.so, for example). Those have a whole separate API and everything, so it's not something you're likely to miss.
Sure. Just rack up a couple dozen 4-way P2s. The problem is that 'raw calculating power' is rarely enough. Unless you have an embarrassingly parallel application, communication costs will kill you. Remember that the interprocess communication on a SMP is orders of magnitude faster than between nodes.
Also, if you have an autoparallelizing compiler, you can turn a single-processor sequential code into a multi-processor multithreaded code with a simple command-line option. This can be a huge win for some codes - probably more 'real world' codes than are embarrassingly parallel. This way, you don't need to learn a new programming model - just recompile your old codes.
My understanding is that the powers that be decided that the compiler group was not delivering a competitive product. So, they decided to force them to compete with the Portland Group, and others. In essence, they had to start paying their own way - by delivering a product that people would pay for. Maybe it's a coincidence, but since they unbundled the compiler, it has been improving by leaps and bounds. For people who don't want to pay for a compiler, gcc and gdb both work fine with Solaris.
As for the libraries, I'm not sure what you're talking about. They ship with the OS - otherwise even 'cp' wouldn't work.
For the thousand people in the insurance company who don't need full powered pc's, just give them a
machine with a p200, a NIC and a 4 gig HD. How much would that cost 200-300?
Hardware is cheap. What's expensive is training the users, maintaining the machine, and downtime when the machine is unavailable due to crashes or upgrades. For what they are doing, there is no reason to have something as complicated as Windows or Linux running.
Though they should make more of the OS available if they want to appeal to CS geeks. I mean, Solaris isn't nearly the educational tool that Linux is because the implementation isn't freely available to look at and tinker with.
Really? I thought the source was available to people at .edu sites. I remember a bunch of hullaballoo when they made that anouncement for Solaris 2.6. I'll bet it's still available.
I think he meant that Sun would not be selling support contracts for Linux - unlike SGI and HP. Just as a point of interest, if you look at the Unix vendors who are selling support for Linux, I think you'll see that all of them also started selling support for NT a few years back. Make of that what you will.
Hmmm. They should, it would show off the machines far better than Solaris.
Not to put too fine a point on it: you have no idea what you're talking about.
Linux will run on small-scale SMP boxes. Currently, Sun's biggest machine has 64 CPUs. Even if Linux can boot on an E10000, trying to compare Linux and Solaris performance would be hugely embarrassing to Linux. Sun has a ton of engineers optimizing Solaris for these huge SMP boxes. How many Linux kernel hackers even have access to an SMP 1/8th this size?
Sun storage is now mainly fiber channel based. Their 14-disk A5000 array can deliver 140 MB/s (using two fiber loops), using Veritas Volume Manager to manage the individual disks. Does Linux support fiber channel at all? Does Veritas run on Linux?
Linux is great, but it's not the answer to everything. Again, to be blunt, it's Linux "advocates" like you who make the rest of us look bad.
Actually, he said that Linux was "the right way to get the wrong answer". He meant that supporting open interfaces was absolutely the Right Thing to do. He was less convinced about open implementation, but he was at least warm to the idea.
By the "wrong answer" he meant that the Linux mindset is to have fat, powerful clients. He believes that the right model is to have fat, powerful servers, and thin clients.
Several times he made analogies to the phone system. Your phone is a thin client, connecting to fat servers. You don't buy software for your phone, the phone company supplies features. Etcetera.
Most of it wouldn't apply to the way your average /. reader uses computers, but it does make sense when you start thinking about the throngs of less sophisticated users out there.
How do you make the internet accessible to those people who can't program their VCRs? Asking them to install and configure Windows or Linux isn't going to cut it. For those people, a WebTV-like solution probably makes sense.
If you have a thousand people in an insurance company doing little but data entry and retrieval, does it really make sense to put a full powered PC or workstation on each desk? It would be a hell of a lot cheaper and more manageable to have some thin, stupid clients communicating with the fat servers in the back room.
Assuming that the right answer is Sun's Slowlaris? Maybe for multiprocessor boxes, but definitely not for the price.
The price? Solaris 7 was available for just the cost of the media. I don't know if that's still true or not.
Linux is a direct competitor to Solaris.
Sun makes its money on hardware. They don't care whether the customer is running Linux or Solaris, as long as they are doing it on Sun hardware.
a) If you really believe that Java is "so unappealing to everyone", I expect you haven't been paying attention.
b) The concept of a VM is what makes Java very appealing to _many_ people. Even though WORA is still largely a pipe dream for complicated apps, a lot of people are still working towards it.
Sun's journalling file system is not from Veritas (although VxFS does run on Solaris). It's an extension of UFS. I could be wrong, but I think the source to all of Solaris (including logging UFS) is available to educational customers.
I've seen mentions in ZDnews and elsewhere that all of Solaris is going to be released under Sun's Community Source License (the Java/Jini license) in the near future.