Check out Scheduled Transfer Protocol. It is a protocol for talking to storage devices over "standard" network hardware. SGI was able to get 790MByte/second over Gigabyte System Network. They have better network hardware than I do...
STP also works over gigabit ethernet hardware (but only at gigabit speeds). It will probably work over 10gigabit ethernet, when that is available in quantity.
Why use special disk interface hardware, if network hardware has better bandwidth, latency, and is cheaper?
I want to tag my tools, so my toolbox will remind
me that I forgot to put away my hammer. If I ignore that reminder, my database can tell me I last used the hammer in attic.
How much does a reader cost? How much do tags cost, in quantity 10, not 100,000?
For obvious privacy reasons, I don't want to have to register my hammer's tag with any outside entity.
The VAX hardware is finally going away, but the
ancient binaries will still run on an Alpha running OpenVMS. And it will run faster.
Without a trusted emulator, they wouldn't
retire the hardware.
My first VAX was an 11/750 running Ultrix 1.0, which was 4.2 BSD with some strings changed in scripts and man pages. Sun's were cheaper and faster back then. It took 27 years for the market
place to realize that VAXen are obsolete! That is market inertia.
With less than 4gbytes of RAM and only 4 processors, this machine is a baby, but it can grow.
Put up a web page listing what hardware you can add. As machines like it get upgraded, you may be able to get more processors, more memory, more disk, tape drives, etc... If you can grow the machine to many processors and more than 4 gbytes of RAM, it will be able to do things that PC class hardware can't. It'll also be quite a learning experience.
Create a grant application for people who want time on your CPU cluster. Try to turn the CPU cluster into a unique resource with a world wide reputation.
The backup is just one step in making your restores work.
I use RCS on source code and configuration files so that I can restore any version ever archived. RCS lets me record a comment on each change, so years later, I can tell what I was thinking when I did something stupid.
I rsync the whole system nightly to another host. In theory, if the primary host dies, I could reboot the secondary host with the backup root partition, and be up 2 minutes later. This has not been tested recently. Mostly, I use this like any other nightly backup, but restores happen at disk/network speed, not tape speeds.
I also use rsync to copy the whole system before doing upgrades, going on vacation, or other times when I want the option to restore from a particular point in time.
I "archive" things to CDR. That is for anything I want to be able to restore more than 18 months from now.
Disk drives are getting cheaper, so I'll be buying more so that I can do more types of restores. (weekly, monthly, etc...)
If a service keeps running when an individual server dies, then a rebooting server will have to refresh its database from a running server. For some types of databases, it is unacceptable to use stale data, even if that database only missed one transaction.
If the database goes stale while a server reboots, it doesn't matter if it is volatile or not.
It might be a much cheaper way to get a large amount of RAMdisk, than buying more system memory.
Motherboards have only so many memory slots. 256MByte DIMMs are the cheapest per MByte, 512MByte DIMMs are still reasonably priced per MByte. Gbyte DIMMs are way more expensive. 2GByte DIMMs are insanely expensive (SGI sells them for their Octanes, I don't know who else does).
If my phone doesn't work, I call the phone company, but not from the broken phone. If my computer won't boot, I go to the web for help, but not from the broken computer.
Technical support for large Beowolf Clusters is not handled the same way as support for my mom's PC. My mom doesn't just need a different level of support, she wants a different style of support.
If her computer is working, she knows how to use the web to get technical help.
There are many ways to accidentally destroy files. Overwriting a file with the wrong stuff is just as bad as removing the file.
It isn't a backup system, it is a restore system. All people care about is if the restore works. Backups are important, because restore can't work without them.
I use different mechanisms for recovering old files. I use RCS for some things, because that gives me a changelog back to the birth of the file. I only use it on files that I knew were important when I created or changed them. I use rsync to keep a current copy of the whole system. This backs up things that I didn't know were important until I destroyed them.
Some media outlets may treat LinuxOne as a serious player in the market. They will be embarrassed and learn to verify their sources in the future.
I have seen zero positive press on LinuxOne aside from their own press releases. The Linux community should respond strongly, quickly and accurately to positive LinuxOne press. This will prevent LinuxOne from growing into a significant problem.
We used to say Linux is good now we must qualify the statement to Good Linux is good.
They are really neat for the first few hours. Then you start to see the limitations.
Regular solids work well. More complex things don't expand and contract so well.
If you get the little set, you think, gosh I can make some cool things with the big set. You need the big set to see that you can only build things so large before they just don't work at all.
There are also little playability issues with the extra swivel joints. They are hard to get together. Sometimes when taking things apart, they come apart in the wrong places.
If you really like construction toys like Lego Technics, assume you'll get 4 to 8 hours of fun out these things. Then it will sit on the shelf.
If everything works right, the fuel cell will consume 3 oxygen molecules for every 2 molecules of methanol and it will produce 2 molecules of carbon dioxide and 2 molecules of water. It is lighter than batteries because you don't count the weight of the oxygen.
Your laptop will produce visible steam when the humidity is high enough. Don't try to use it in a closed box, it will suffocate.
The methanol fuel is toxic, but the by-products aren't. see the EPA's chemical summary This isn't much different from batteries which are generally also toxic.
The methanol must be very pure, or the fuel cell will stop working.
Making methanol is a lot easier than making batteries, so it should be a lot cheaper eventually. Safe packaging and purity requirements will make it more expensive at first.
Slashdot Mirror
STP also works over gigabit ethernet hardware (but only at gigabit speeds). It will probably work over 10gigabit ethernet, when that is available in quantity.
Why use special disk interface hardware, if network hardware has better bandwidth, latency, and is cheaper?
How much does a reader cost? How much do tags cost, in quantity 10, not 100,000?
For obvious privacy reasons, I don't want to have to register my hammer's tag with any outside entity.
This will hurt open source if people take it seriously. The rules aren't well defined, so there is sure to be controversy over who wins.
It will be worse if the contest folks don't show up with the money in a timely way.
Without a trusted emulator, they wouldn't retire the hardware.
My first VAX was an 11/750 running Ultrix 1.0, which was 4.2 BSD with some strings changed in scripts and man pages. Sun's were cheaper and faster back then. It took 27 years for the market place to realize that VAXen are obsolete! That is market inertia.
Put up a web page listing what hardware you can add. As machines like it get upgraded, you may be able to get more processors, more memory, more disk, tape drives, etc... If you can grow the machine to many processors and more than 4 gbytes of RAM, it will be able to do things that PC class hardware can't. It'll also be quite a learning experience.
Create a grant application for people who want time on your CPU cluster. Try to turn the CPU cluster into a unique resource with a world wide reputation.
I use RCS on source code and configuration files so that I can restore any version ever archived. RCS lets me record a comment on each change, so years later, I can tell what I was thinking when I did something stupid.
I rsync the whole system nightly to another host. In theory, if the primary host dies, I could reboot the secondary host with the backup root partition, and be up 2 minutes later. This has not been tested recently. Mostly, I use this like any other nightly backup, but restores happen at disk/network speed, not tape speeds.
I also use rsync to copy the whole system before doing upgrades, going on vacation, or other times when I want the option to restore from a particular point in time.
I "archive" things to CDR. That is for anything I want to be able to restore more than 18 months from now.
Disk drives are getting cheaper, so I'll be buying more so that I can do more types of restores. (weekly, monthly, etc...)
Verify the MD5 checksum with the author or with someone who has checked the binary.
If the database goes stale while a server reboots, it doesn't matter if it is volatile or not.
Motherboards have only so many memory slots. 256MByte DIMMs are the cheapest per MByte, 512MByte DIMMs are still reasonably priced per MByte. Gbyte DIMMs are way more expensive. 2GByte DIMMs are insanely expensive (SGI sells them for their Octanes, I don't know who else does).
Technical support for large Beowolf Clusters is not handled the same way as support for my mom's PC. My mom doesn't just need a different level of support, she wants a different style of support.
If her computer is working, she knows how to use the web to get technical help.
Since devfs is a file system, you can add it to your backup rotation. You might choose to dump it to tape nightly or rsync it to a remote host hourly.
If you want higher reliability, add a journal to the filesystem. If you want high availability mirror changes to the other servers.
Linus made it so you can add as much catastrophic failure tolerance as you feel is necessary.
If the cooler could be scaled up, it would take 3 kilowatts to maintain the temperature of a 30 watt processor.
It's yet to be seen if XFS will be stable enough to actually be included in the 2.4 kernel.
It isn't a backup system, it is a restore system. All people care about is if the restore works. Backups are important, because restore can't work without them.
I use different mechanisms for recovering old files. I use RCS for some things, because that gives me a changelog back to the birth of the file. I only use it on files that I knew were important when I created or changed them. I use rsync to keep a current copy of the whole system. This backs up things that I didn't know were important until I destroyed them.
I have seen zero positive press on LinuxOne aside from their own press releases. The Linux community should respond strongly, quickly and accurately to positive LinuxOne press. This will prevent LinuxOne from growing into a significant problem.
We used to say Linux is good now we must qualify the statement to Good Linux is good.
Regular solids work well. More complex things don't expand and contract so well.
If you get the little set, you think, gosh I can make some cool things with the big set. You need the big set to see that you can only build things so large before they just don't work at all.
There are also little playability issues with the extra swivel joints. They are hard to get together. Sometimes when taking things apart, they come apart in the wrong places.
If you really like construction toys like Lego Technics, assume you'll get 4 to 8 hours of fun out these things. Then it will sit on the shelf.
What is 4 hours of construction fun worth to you?
Your laptop will produce visible steam when the humidity is high enough. Don't try to use it in a closed box, it will suffocate.
The methanol fuel is toxic, but the by-products aren't. see the EPA's chemical summary This isn't much different from batteries which are generally also toxic.
The methanol must be very pure, or the fuel cell will stop working.
Making methanol is a lot easier than making batteries, so it should be a lot cheaper eventually. Safe packaging and purity requirements will make it more expensive at first.
It might have been a failed transaction for Hello Direct, but I found what I needed. So it wasn't a failed purchase for me.