Right. Some old toolchains had the convention of using exactly 32 bits for each symbol name, with 3 characters in each 16 bits using RADIX50 encoding (where the 50 is octal, meaning 40 decimal). This makes symbol comparison very simple, but of course means symbols must be truncated to 6 characters and case-folded. The C90 standard specifies undefined behaviour for programs using identifiers that would collide under this mapping.
Twisted pair cable can do pretty well and give you a good few gigabits per second if it's good, high quality copper and has a decent amount of shielding and good insulated and grounded splices and connectors.
Actually it's possible to transfer 10 gigabits over cat 5e. It requires fairly heavy DSP though, and cat 6a for a 100 metre run. That's still unshielded though.
I'm sure they're interested in selling music from the most popular artists. However, their business model involves high volume, low price and no DR, which isn't to the liking of the major labels which most of those artists are signed to. I certainly wouldn't want them to compromise on that. The majors will probably have to deal on their terms eventually.
Some manufacturers still run DOS-based manufacturing tests, perhaps because DOS boots so quickly. The major reason for HP and Dell to offer FreeDOS is that their contracts with MS forbid them to sell "naked PCs". As for why FreeDOS and not Linux... at a guess, it avoids the question of which Linux, and those who immediately replace it with Linux can't claim technical support for either OS.
IP and TCP have checksums in their respective packet headers. NICs that perform hardware checksum generation need to buffer each whole packet before transmission. (Ethernet has its checksum at the end.) The NIC also needs to buffer on both the receive and send sides to allow for DMA latency.
The next major source of latency is the need for system calls for I/O and to copy received data from kernel buffers to user-space buffers. There are various systems that allow processes to do network I/O directly, though most of them are proprietary (rDMA protocols) or insecure (TCP/IP and network driver in user-space with a standard NIC).
Finally, signal processing for physical layer signalling can take significant time. That depends on what the physical layer is. Short-range shielded links don't need so much signal processing as twisted-pair.
Out of these, using some sort of PCI-PCI bridge in place of Ethernet could save you a bit of buffering and a bit of signal processing. But it's the kernel calls that are the real killer. The company I work for has done work on user-level TCP/IP networking with hardware support (now called OpenOnload). Verari reported on our first NIC that "[t]he streaming latency is sub-3 microseconds, which is competitive with specialized high-speed networks, such as InfiniBand and Myrinet".
That's stretching matters a bit. Trespass is not normally a criminal offence, but it is a tort - that is, the property owner can be awarded damages against a trespasser. Also, a trespasser in a house might be considered to threaten the occupant (that could be assault) or to be preparing for burglary (breaking and entering). (IANAL.)
It seems to me that Windows 2000 was a bigger change than Vista. The reason Vista took so damn long was that development was "reset" (i.e. the development branch was abandoned) half way along.
The SSH file transfer protocol supports retrieving byte ranges, so it is possible to implement reget. I made a patch to add it to OpenSSH some time ago, though it has not been merged.
No, the GPL says this option "...is allowed only for noncommercial distribution and only if you received the program in object code or executable form with such an offer, in accord with Subsection b above."
ZFS is licensed under CDDL (and probably GPLv3 in future). Neither of these is compatible with GPLv2, so an in-kernel implementation would be of dubious legality. Nothing Apple does with it is going to change that.
DVD+R discs have more precise positioning information which allows a burner to continue burning after a buffer under-run and still make a valid disc. They can also be marked as DVD-ROM. These both help to improve compatibility. Still, I agree the capacity difference here is more obvious and significant.
There's no requirement to distribute source at the same time. As an alternative, you can offer to distribute source using common media formats or protocols, which Debian does. Some DVD build configurations do include source though.
/* The C standard says both 0 and EXIT_SUCCESS are allowed as indication of success. It's up to the implementation to convert the return value if necessary. */
That wasn't the deal-breaker. Permission was also conditional on prior approval of every patch Debian applied to its package. This would mean we could be prevented from selectively fixing bugs in a stable release. One might hope that Debian could trust Mozilla not to withhold approval for patches unreasonably, but the mutual trust had kind of broken down after the apparent change of policy by Mozilla and a claim that the existing patches were poor quality code. (The same patches were approved in the Ubuntu package, so the code wasn't really the issue.)
Slashdot Mirror
Right. Some old toolchains had the convention of using exactly 32 bits for each symbol name, with 3 characters in each 16 bits using RADIX50 encoding (where the 50 is octal, meaning 40 decimal). This makes symbol comparison very simple, but of course means symbols must be truncated to 6 characters and case-folded. The C90 standard specifies undefined behaviour for programs using identifiers that would collide under this mapping.
Actually it's possible to transfer 10 gigabits over cat 5e. It requires fairly heavy DSP though, and cat 6a for a 100 metre run. That's still unshielded though.
What made you think your competitors' ASICs were less buggy?
I'm sure they're interested in selling music from the most popular artists. However, their business model involves high volume, low price and no DR, which isn't to the liking of the major labels which most of those artists are signed to. I certainly wouldn't want them to compromise on that. The majors will probably have to deal on their terms eventually.
eMusic has music by many well-known artists. Not so much of the current hits, admittedly. See top rock/pop artists for this month.
UK wasn't a valid ISO 3166 country code until recently. I believe it's now accepted as an alias for GB, but only because of the existence of the TLD.
Here's the commercial referred to there.
Windows comes with the "cscript" and "wscript" front-ends to VBScript and Javascript, and there are many scriptable components in the OS.
One of the conditions for MS OEM pricing is not selling "naked PCs", so no.
Some manufacturers still run DOS-based manufacturing tests, perhaps because DOS boots so quickly. The major reason for HP and Dell to offer FreeDOS is that their contracts with MS forbid them to sell "naked PCs". As for why FreeDOS and not Linux... at a guess, it avoids the question of which Linux, and those who immediately replace it with Linux can't claim technical support for either OS.
It's certainly required for suppliers of server components.
IP and TCP have checksums in their respective packet headers. NICs that perform hardware checksum generation need to buffer each whole packet before transmission. (Ethernet has its checksum at the end.) The NIC also needs to buffer on both the receive and send sides to allow for DMA latency.
The next major source of latency is the need for system calls for I/O and to copy received data from kernel buffers to user-space buffers. There are various systems that allow processes to do network I/O directly, though most of them are proprietary (rDMA protocols) or insecure (TCP/IP and network driver in user-space with a standard NIC).
Finally, signal processing for physical layer signalling can take significant time. That depends on what the physical layer is. Short-range shielded links don't need so much signal processing as twisted-pair.
Out of these, using some sort of PCI-PCI bridge in place of Ethernet could save you a bit of buffering and a bit of signal processing. But it's the kernel calls that are the real killer. The company I work for has done work on user-level TCP/IP networking with hardware support (now called OpenOnload). Verari reported on our first NIC that "[t]he streaming latency is sub-3 microseconds, which is competitive with specialized high-speed networks, such as InfiniBand and Myrinet".
That's stretching matters a bit. Trespass is not normally a criminal offence, but it is a tort - that is, the property owner can be awarded damages against a trespasser. Also, a trespasser in a house might be considered to threaten the occupant (that could be assault) or to be preparing for burglary (breaking and entering). (IANAL.)
It seems to me that Windows 2000 was a bigger change than Vista. The reason Vista took so damn long was that development was "reset" (i.e. the development branch was abandoned) half way along.
I've not seen this, but I always used XP Professional as a domain member. Maybe that's the difference?
The SSH file transfer protocol supports retrieving byte ranges, so it is possible to implement reget. I made a patch to add it to OpenSSH some time ago, though it has not been merged.
No, the GPL says this option "...is allowed only for noncommercial distribution and only if you received the program in object code or executable form with such an offer, in accord with Subsection b above."
ZFS is licensed under CDDL (and probably GPLv3 in future). Neither of these is compatible with GPLv2, so an in-kernel implementation would be of dubious legality. Nothing Apple does with it is going to change that.
Actually there's a "Web Edition" of Windows 2003 with some server capabilities removed and a lower licence fee; they quote a standard price of $399.
DVD+R discs have more precise positioning information which allows a burner to continue burning after a buffer under-run and still make a valid disc. They can also be marked as DVD-ROM. These both help to improve compatibility. Still, I agree the capacity difference here is more obvious and significant.
There's no requirement to distribute source at the same time. As an alternative, you can offer to distribute source using common media formats or protocols, which Debian does. Some DVD build configurations do include source though.
Will make you many friends among those who read diffs!
/* The C standard says both 0 and EXIT_SUCCESS are allowed as indication of success. It's up to the implementation to convert the return value if necessary. */
That wasn't the deal-breaker. Permission was also conditional on prior approval of every patch Debian applied to its package. This would mean we could be prevented from selectively fixing bugs in a stable release. One might hope that Debian could trust Mozilla not to withhold approval for patches unreasonably, but the mutual trust had kind of broken down after the apparent change of policy by Mozilla and a claim that the existing patches were poor quality code. (The same patches were approved in the Ubuntu package, so the code wasn't really the issue.)
BP also cuts corners on maintenance, leading to spillage and fatal explosions.