I was under the impression that designing a modern CPU took a unique combination of a lot of skillful engineers and an extremely expensive modern fab. HP probably has plenty of manufacturing plants, and heck, maybe a few fabs for CCDs and the like for cameras and scanners and other optics... But I doubt they have a CPU fab.
Find me a place that sells a twenty five year old tape reader that can read twenty five year old tapes, and tell me that it's going to be cost effective to keep having to use that technology.
And I never said hard disks were for long term storage. What I'm saying is, nothing is anymore. You can't count on being able to buy any of the necessary components for a technology that exists today in fifty years, or a hundred years. You'll be lucky if you can find it on eBay in 25 years or that some company might cater to your business by then. But it's just a terrible idea to think that you can set your data aside in some media and forget about it.
We're talking about two different use cases. For large amounts of data, a hard drive by courier can be a lot secure and reliable.
For long-term archival, hard drives are just now becoming popular as a temporary means of data storage. The problem is archivists seem to think that if you get just the right media and store it in just the right conditions, it'll last forever because that's how archival has always been done. But technology advances so quickly that even if you do everything right, you can end up with something no one has a reader for in ten, twenty, fifty years. Do you think fifty years from now many people will have Blu-Ray readers? Really? Do you think they'll be anything but niche products sold at extravagant prices (or on eBay) to people who bought into the BD-ROM archival marketing?
No one has an answer for long term media over twenty five years. No one. CD-ROM has barely been around that long, tapes that old cannot be read in any current players, and hard drives back then used IDE, which I can still get adapters for.
The only realistic way to archive digital media is to have a planned rotation policy. So, if I were to start today I'd start with 1 or 1.5TB disks in bulk, in 3 years consolidate those 2:1 to 2 or 3TB disks, etc. And keep consolidating (reducing the number of disks while also storing the data at multiple sites) perpetually. That's the only solution that keeps your data yours, and not at the mercy of a technology that you know won't be supported in twenty five years.
And of course, just because I recommend hard drives don't mean I recommend throwing everything else out the window. Judicious use of ECC, storing archive data at multiple sites and even biting the bullet and storing data multiple times at a single site are all options that should be explored when determining your archive policy.
Why? Hard drives are fantastic, 1TB for $100 and have superior read/write characteristics. Why worry about WORM when you can digitally sign the data and replicate it offsite cheaply without having to invest in niche burning and changing equipment that would be necessary to switch away from hard disks?
Switching to optical media is like switching to tape. Unless you're already invested, I don't see why you'd want to get involved there.
But that's the point, this is a really, really rare error that would be difficult to audit for.
Maybe I should give up and realize my point isn't being made well by this back and forth. Let me lay it out:
1. Someone said that a Colorado woman won $42,000,000 from a defective slot machine 2. Someone said well that's just a testament to how good slot machines are, "Hundreds of thousands of slot 'pulls' an hour across the United Sates - one significant newsworthy error. Sounds pretty damm reliable to me." 3. I said that it was just one newsworthy error and that there was a pretty significant selection bias. While that 42,000,000 pull was heavily investigated, an erroneously low or just average win amount would not receive equal attention, and if rare enough would not be easily caught. It would still be an error in the code giving invalid output.
That's all I'm saying. Someone mentioned a real case of a slot machine performing outside of its specification, and someone else said that the frequency of something like that happening is a testament to the quality of the code. I suggested the code could be defective and it could have less noticeable errors with greater frequency and no one would be the wiser.
No, I'm saying that suppose there is a fluke error in the slot machines. Very rarely gives incorrect amounts. Well, $42 million is a rather large number and would see prompt investigation, especially if that's not an amount the slot machine should be able to give out (I don't know the particulars of the case.)
But in Vegas at least, if there was a random fluke error that gave someone $10000 instead of $10, the casino probably wouldn't even notice. Like I said, nearly random error. Obviously the $42 million payout is not a recurring thing either. Likewise, if some guy gets $10 when the machine should have connected to homebase and noticed he was the 1,000,000th pull and given him an extra $10000, no one would have noticed. He would never notice, the casino would not notice, and the company would never be notified about that bug.
What a joke, those aren't SANs, they're storage servers. Single point of failure, and all that. I want to see a SAN. And Nexenta is just an OS for a storage server.
While SUN does not have a true "SAN" with hybrid storage it does support clustering the storage servers in an active-active config.
Technologies. Solaris and OpenSolaris are full of things that geeks, Windows and *nix, would love to see in their OS of choice but Sun invented first. ZFS, Dtrace, and dozens of other features languished in Solaris covered by patents or from a just plain lack of ability and motivation to recreate those features in other OSes.
Even now the comparable features in other OSes are just now starting to approach a release candidate quality, and Sun has already started building new technologies and completely unique solutions based on stuff only Solaris has. Look at Oracle/Sun's new hybrid storage SAN for example. It uses a bunch of spinning disks (which everyone knows are so passé now) in a huge ZFS pool combined with 100-200GB of very fast SSD storage as an active logging and cache system. The result is that even very nearly random writes, when done to a small enough area on disk, can be done almost linearly once fully cached by the SSDs. You thought your RAID card was clever being able to cache 256MB-1GB. These things cache ten or a hundred times as much.
Really clever stuff which is hard to duplicate on other platforms. You certainly can't get a supported solution for something like that from anyone but Sun/Oracle.
For a large casino, a $10k payout wouldn't seem out of the ordinary and warrant any significant investigation. For every $10k they payout, they take $11k in. No problem.
One newsworthy error, you mean. Not "one significant". How many slot machines gave out erroneous amounts but were small enough that the casino didn't think twice? 42,000,000 is a lottery level payout, and would be intensely examined whether legit or not. I doubt most casinos would bat an eye at $10,000, and no one would be the wiser if a seven digit payout turned into a three digit one.
No, the end user or device is what is licensed when buying CALs from Microsoft.
For example:
You have a database application that is accessed by some users directly (query editor or application that directly talks to SQL server.) These users or the devices they use need to be licensed.
You have a database application that a web server talks to, and through this, end users query your database indirectly (that is, when hitting the website, queries are generated indirectly to hit the database.) The end users, because they are still talking to the database albeit indirectly and causing queries to be performed will require CALs.
Lastly, suppose you have a report generator application that creates only predetermined reports (nothing on-demand) at specified times and saves the data to a file (CSV, PDF, XLS, etc) or sends an email. This requires only one CAL regardless of how many people receive the output (the file) because only one user or device is directly or indirectly hitting the database.
There's a real simple reason the licensing is like this, because otherwise what you could do is set up two SQL servers, one running SQL Server Enterprise (licensed by Server + 1 CAL as opposed to per processor, which is much more expensive) and one running SQL Server Express (free.) The Express instance would then have the Enterprise server as a linked server, and all queries would be redirected through that connection. This would effectively multiplex all of your users through one node. Microsoft doesn't want to allow this, so they've set it up so you can't.
tl;dr: the webserver is not the end user, it is not generating queries on its own, but only on demand from end users. It is those end users that need to be licensed.
There's SQL Server Web edition of course, and you could always use something like PostgreSQL, MySQL, or whatever. I'm not endorsing SQL Server in any way, just fighting licensing FUD.
Interesting that your accounting software requires DEP turned off.
And four and eight socket servers are typically on unique hardware which comes with unique support concerns. Everyone in the business does stuff like this, even on the *nix side of things.
Actually I'm a little surprised that Standard supports 4 socket servers but only 32GB of memory, and that Enterprise goes up to 8 sockets. Datacenter goes up to 64 though, and I'm not aware of any servers with that many sockets. I think the biggest you can get is a 16 socket server from IBM? Supposedly Nehalem-EX will support up to 32 sockets in a similar manner.
SQL Server is per server and CAL (that is, you license your server and then buy 50 CALs for the end users) or per socket/CPU. My advice is to always use the latter option because multiplexing doesn't count as one use, so if your website uses SQL Server, you better have CALs for everyone that visits... or just license it by socket.
The CAL model is probably on its way out, it just doesn't make sense that if you have a public facing website that hits the database for authentication, even if you only have 50 real users, you still have over three billion potential users that you need to license.
The bigger problem is write fragmentation. I have a one year old SSD (Intel X25-M, first generation) with about 15 gigabytes of writes per day and earlier this year I had to do a secure wipe on it to reset the fragmentation tables. With TRIM support this problem should go away.
I would really like to know when RAID controllers start supporting TRIM though, and when it'll be possible to hook up a readzilla/logzilla SSD like Sun has to a server to serve as a large second level cache and non-volatile write log.
Slashdot Mirror
Always have a backup chute.
They have 45nm CPU fabs? Really?
I was under the impression that designing a modern CPU took a unique combination of a lot of skillful engineers and an extremely expensive modern fab. HP probably has plenty of manufacturing plants, and heck, maybe a few fabs for CCDs and the like for cameras and scanners and other optics... But I doubt they have a CPU fab.
No one said anything about using hard disk for long-term offline archival. I certainly wouldn't recommend it.
Find me a place that sells a twenty five year old tape reader that can read twenty five year old tapes, and tell me that it's going to be cost effective to keep having to use that technology.
And I never said hard disks were for long term storage. What I'm saying is, nothing is anymore. You can't count on being able to buy any of the necessary components for a technology that exists today in fifty years, or a hundred years. You'll be lucky if you can find it on eBay in 25 years or that some company might cater to your business by then. But it's just a terrible idea to think that you can set your data aside in some media and forget about it.
We're talking about two different use cases. For large amounts of data, a hard drive by courier can be a lot secure and reliable.
For long-term archival, hard drives are just now becoming popular as a temporary means of data storage. The problem is archivists seem to think that if you get just the right media and store it in just the right conditions, it'll last forever because that's how archival has always been done. But technology advances so quickly that even if you do everything right, you can end up with something no one has a reader for in ten, twenty, fifty years. Do you think fifty years from now many people will have Blu-Ray readers? Really? Do you think they'll be anything but niche products sold at extravagant prices (or on eBay) to people who bought into the BD-ROM archival marketing?
I work full time IT, part time as a movie theatre projectionist.
Guess how all our digital movies arrive?
Hard disk.
Are you serious?
No one has an answer for long term media over twenty five years. No one. CD-ROM has barely been around that long, tapes that old cannot be read in any current players, and hard drives back then used IDE, which I can still get adapters for.
The only realistic way to archive digital media is to have a planned rotation policy. So, if I were to start today I'd start with 1 or 1.5TB disks in bulk, in 3 years consolidate those 2:1 to 2 or 3TB disks, etc. And keep consolidating (reducing the number of disks while also storing the data at multiple sites) perpetually. That's the only solution that keeps your data yours, and not at the mercy of a technology that you know won't be supported in twenty five years.
And of course, just because I recommend hard drives don't mean I recommend throwing everything else out the window. Judicious use of ECC, storing archive data at multiple sites and even biting the bullet and storing data multiple times at a single site are all options that should be explored when determining your archive policy.
HP has fabs and/or competent CPU designers?
I doubt Intel really cares who they sell it to, as long as someone keeps buying. When HP moves on from Itanium, it's done for.
Why? Hard drives are fantastic, 1TB for $100 and have superior read/write characteristics. Why worry about WORM when you can digitally sign the data and replicate it offsite cheaply without having to invest in niche burning and changing equipment that would be necessary to switch away from hard disks?
Switching to optical media is like switching to tape. Unless you're already invested, I don't see why you'd want to get involved there.
Since you didn't specify a unit, we shall send you the sum in Zimbabwean Dollars promptly.
But that's the point, this is a really, really rare error that would be difficult to audit for.
Maybe I should give up and realize my point isn't being made well by this back and forth. Let me lay it out:
1. Someone said that a Colorado woman won $42,000,000 from a defective slot machine
2. Someone said well that's just a testament to how good slot machines are, "Hundreds of thousands of slot 'pulls' an hour across the United Sates - one significant newsworthy error. Sounds pretty damm reliable to me."
3. I said that it was just one newsworthy error and that there was a pretty significant selection bias. While that 42,000,000 pull was heavily investigated, an erroneously low or just average win amount would not receive equal attention, and if rare enough would not be easily caught. It would still be an error in the code giving invalid output.
That's all I'm saying. Someone mentioned a real case of a slot machine performing outside of its specification, and someone else said that the frequency of something like that happening is a testament to the quality of the code. I suggested the code could be defective and it could have less noticeable errors with greater frequency and no one would be the wiser.
What are the obvious reasons?
I didn't know the Library of Congress was named after someone? Was there an ancient librarian named Mr. Congress?
No, I'm saying that suppose there is a fluke error in the slot machines. Very rarely gives incorrect amounts. Well, $42 million is a rather large number and would see prompt investigation, especially if that's not an amount the slot machine should be able to give out (I don't know the particulars of the case.)
But in Vegas at least, if there was a random fluke error that gave someone $10000 instead of $10, the casino probably wouldn't even notice. Like I said, nearly random error. Obviously the $42 million payout is not a recurring thing either. Likewise, if some guy gets $10 when the machine should have connected to homebase and noticed he was the 1,000,000th pull and given him an extra $10000, no one would have noticed. He would never notice, the casino would not notice, and the company would never be notified about that bug.
What a joke, those aren't SANs, they're storage servers. Single point of failure, and all that. I want to see a SAN. And Nexenta is just an OS for a storage server.
While SUN does not have a true "SAN" with hybrid storage it does support clustering the storage servers in an active-active config.
Technologies. Solaris and OpenSolaris are full of things that geeks, Windows and *nix, would love to see in their OS of choice but Sun invented first. ZFS, Dtrace, and dozens of other features languished in Solaris covered by patents or from a just plain lack of ability and motivation to recreate those features in other OSes.
Even now the comparable features in other OSes are just now starting to approach a release candidate quality, and Sun has already started building new technologies and completely unique solutions based on stuff only Solaris has. Look at Oracle/Sun's new hybrid storage SAN for example. It uses a bunch of spinning disks (which everyone knows are so passé now) in a huge ZFS pool combined with 100-200GB of very fast SSD storage as an active logging and cache system. The result is that even very nearly random writes, when done to a small enough area on disk, can be done almost linearly once fully cached by the SSDs. You thought your RAID card was clever being able to cache 256MB-1GB. These things cache ten or a hundred times as much.
Really clever stuff which is hard to duplicate on other platforms. You certainly can't get a supported solution for something like that from anyone but Sun/Oracle.
For a large casino, a $10k payout wouldn't seem out of the ordinary and warrant any significant investigation. For every $10k they payout, they take $11k in. No problem.
One newsworthy error, you mean. Not "one significant". How many slot machines gave out erroneous amounts but were small enough that the casino didn't think twice? 42,000,000 is a lottery level payout, and would be intensely examined whether legit or not. I doubt most casinos would bat an eye at $10,000, and no one would be the wiser if a seven digit payout turned into a three digit one.
No, the end user or device is what is licensed when buying CALs from Microsoft.
For example:
You have a database application that is accessed by some users directly (query editor or application that directly talks to SQL server.) These users or the devices they use need to be licensed.
You have a database application that a web server talks to, and through this, end users query your database indirectly (that is, when hitting the website, queries are generated indirectly to hit the database.) The end users, because they are still talking to the database albeit indirectly and causing queries to be performed will require CALs.
Lastly, suppose you have a report generator application that creates only predetermined reports (nothing on-demand) at specified times and saves the data to a file (CSV, PDF, XLS, etc) or sends an email. This requires only one CAL regardless of how many people receive the output (the file) because only one user or device is directly or indirectly hitting the database.
There's a real simple reason the licensing is like this, because otherwise what you could do is set up two SQL servers, one running SQL Server Enterprise (licensed by Server + 1 CAL as opposed to per processor, which is much more expensive) and one running SQL Server Express (free.) The Express instance would then have the Enterprise server as a linked server, and all queries would be redirected through that connection. This would effectively multiplex all of your users through one node. Microsoft doesn't want to allow this, so they've set it up so you can't.
tl;dr: the webserver is not the end user, it is not generating queries on its own, but only on demand from end users. It is those end users that need to be licensed.
There's SQL Server Web edition of course, and you could always use something like PostgreSQL, MySQL, or whatever. I'm not endorsing SQL Server in any way, just fighting licensing FUD.
Interesting that your accounting software requires DEP turned off.
And four and eight socket servers are typically on unique hardware which comes with unique support concerns. Everyone in the business does stuff like this, even on the *nix side of things.
Actually I'm a little surprised that Standard supports 4 socket servers but only 32GB of memory, and that Enterprise goes up to 8 sockets. Datacenter goes up to 64 though, and I'm not aware of any servers with that many sockets. I think the biggest you can get is a 16 socket server from IBM? Supposedly Nehalem-EX will support up to 32 sockets in a similar manner.
SQL Server is per server and CAL (that is, you license your server and then buy 50 CALs for the end users) or per socket/CPU. My advice is to always use the latter option because multiplexing doesn't count as one use, so if your website uses SQL Server, you better have CALs for everyone that visits... or just license it by socket.
The CAL model is probably on its way out, it just doesn't make sense that if you have a public facing website that hits the database for authentication, even if you only have 50 real users, you still have over three billion potential users that you need to license.
Huh? Windows Server is licensed:
Standard and Enterprise: per server (motherboard?)
Datacenter: per CPU socket
The bigger problem is write fragmentation. I have a one year old SSD (Intel X25-M, first generation) with about 15 gigabytes of writes per day and earlier this year I had to do a secure wipe on it to reset the fragmentation tables. With TRIM support this problem should go away.
I would really like to know when RAID controllers start supporting TRIM though, and when it'll be possible to hook up a readzilla/logzilla SSD like Sun has to a server to serve as a large second level cache and non-volatile write log.
That's a little disappointing, the bit about not being able to test more interesting deployment topologies.
However I appreciate your help here, this will help me as it gives me something I can validate code against and say "Yes I can support Oracle."