Passive is not a single-connection transfer. It uses an outbound connection for the data (from the POV of the client), whereas non-passive uses an inbound connection. That's why you normally need pasv to go through a firewall.
Refcount almost certainly takes less time than other kinds of GC because it doesn't have to do any searching.
You may not have to search, but you have to take the time to increment and decrement the refcount each time. Depending on how things are done (for example, is the refcount incremented when a ref is pushed to the stack?), this can account for a whole lot of time. Throw in the time lost to locking if you're on a multithreaded system. Writes are slower than reads, too. You end up wasting a lot of CPU trying to maintain information about references, when you only need one bit of information: is it in use.
By contrast, modern generational garbage collectors use heuristics to minimize the searching done. Rather than having a lot of work to maintain information all the time that you're using it, you do a little work to GC it only in the GC.
It's true that refcout GC uses an extra integer per item of data, but how does that compare to the memory a regular GC must use to keep track of everything?
There isn't much "everything" to keep track of.
Refcounts need to track how many outstanding references exist, all the time. (Note: Deferred refcounting and other systems can help with that, but that's a time savings, not a space savings.) On the other hand, GC only cares about one thing: is X garbage? What's more, it only cares about it while GC is taking place.
Baby's First Mark-and-Sweep GC (written by undergrad students first learing about GC) use only one bit of information per object during the program's life, plus one word per object in the longest chain at GC time. This is much less than the one word per every object that most refcounting schemes use! In other collectors, auxilary storage may be only needed during the GC, and is very low. Knuth talks about a number of variations on keeping memory usage very low (and only during GC) but those books are at my office right now.
GC technology was a big discussion topic when memory was scarce, so there's been a number of algorithms written to keep memory usage low during GC. Modern GCs use more than the theoretical minimum, of course, but nowhere near one word per object. In many cases, a bit for marking garbage can be taken from the type tag in typed languages. The type tag may be a certain length-- eg, a byte-- for effeciency purposes, and therefore may have a bit to waste. That means that the GCs memory overhead is effectively O(1)... just enough for the generation table!
Do some research on GCs; "garbage collection FAQ" on Google would be a start. Typically, reference counting takes more memory and more time than more sophisticated GC techniques. Besides, Python doesn't refcount anymore.
I keep hearing this from different people. I do a lot of replacing stuff with alternatives. For example, I don't have Windows, I have FreeBSD w/ OpenOffice, etc. But as a TiVo user, I don't think that I could switch to a computer-based solution.
There's just something about sitting down at the TV with a remote that can't be replaced. Also, there's something about the TiVo user interface that I think would become a lot more difficult in the translation to a computer.
It's not just about recording shows. It's about changing the way you watch TV.
1. Booting diagnostic programs
2. The first disk of your downloaded FreeBSD install. (I think Debian does the same thing: you d/l a couple of floppies and the rest gets fetched during the install. Beats having to burn a CD!)
If you're a female, this could have the opposite effect intended.
Besides, I think there's laws against that and stuff. See the 1963 FCC Regulations Against Talking Dirty On The Phone To Somebody Who Doesn't Want To Hear It (RATDOTPTSWDWTHI).
but selling unauthorized modifications to someone else's work is clearly not fair use: this is no different from a third party selling DVDs of modified scenes from the original work,
One of these is offering somebody suggestions regarding when to fast forward, etc, for a profit. The other is copying copyrighted works and distributing them for profit. Even if I take the **AA party line, only one of these robs the distribution chain of their fair share of the $$.
The last guy I interviewed for a programming job was a Perl coder. I asked him if he knew Lisp. He replied that he didn't. I then took out a half-page of Lisp code, and told him to add a particular feature. This wasn't a test to see if he got the answer right; I wanted to see how he handles stressful situations, and how well he can adapt to unusual circumstances. (He was hired, by the way, and is doing great.)
Note that not all OOPs are the same. Generic function OOP uses a little bit different structure than methods-on-objects OOPs. Writing one of these in a language designed for the other makes for icky code.
I disagree. The most popular languages today all more or less follow this (mostly because they're all Algol descendants), but not all languages do.
Alan J. Perlis said, "A language that doesn't affect the way you think about programming is not worth knowing." I agree. Once you've learned C, then learning Pascal or Perl is nothing. But I've seen a lot of people who are sharp-on in Perl that couldn't wrap their heads around functional languages. Ditto for teaching people OO for the first time.
If you're just learning languages by thinking they're all the same, then you're not learning languages. Don't write Perl code in Lisp; learn Lisp.
When I was in college, my roommate loved playing Starcon 2. He bought an AdLib, so he could enjoy the sound. I expected only music, but to my surprise, the MOD-based soundtrack played perfectly, and all the sound effects were identical to what I heard on an SB-compatible!
I'm surprised that one of these labels hasn't already contact Steve Jobs and asked him to help them "get with it" technologically.
I don't know whether or not Apple can do that... they've got a non-compete agreement keeping them out of the music production industry. (This is one reason why one of their first sound samples was called "sosumi", aka "So sue me".)
I have been assigned to a project to figure out how to
make a diskless portable workstation (laptop) boot through a wireless
connection. The idea is to have a stateless client that stores no
local data (for security purposes).
What's the model here? Does somebody walk into a secure facility,
pick one up, use it to do some eyes-only investigation, and return it
when they leave? What are you trying to secure against? Tampering,
or somebody walking off with the data? The solution often depends on
the threat model.
If your threat model is to prevent against tampering, then you may
be better off exploring other options. For example, have you
considered read-only media? How about having the tablets re-ghosted
when they're returned, before they're re-issued? That can be done in
an automated fashion without a whole lot of hassle-- primarily through
the PXE extensions you've already investigated, combined with hardware
at the docking stations.
If you're trying to keep people from walking off with data, then
diskless isn't going to be the way to go. A lot of data gets left in
RAM after power is removed. (See Gutman, P., "Secure Deletion of Data
from Magnetic and Solid-State Memory," Proceedings of the Sixth USENIX
Security Symposium, July 1996, or do a Google search for "RAM
remanence".) You may have seen some computers-- notably the old
Macs-- that would power up with their last display still on the
screen!
Also, if you have a totally stateless box driven by a wireless LAN,
then some shmuck can easily sit in a van a half-mile away with his
laptop and find out everything you're wanting to keep private.
Stateless booting means that your encryption has to be bootstrapped!
A lot of naive ways of doing this exist, such as sending a root
filesystem with encryption keys already on it. Some of these open
themselves up to passive attacks. Even more sophisticated techniques,
such as DH, still are totally vulnerable to active attacks (like the
guy in the van pretending to be one of your tablets asking to be
bootstrapped).
I seriously suggest you rethink your security model. The Windows
compatibility is a big problem. It keeps some of the latest crypto
filesystems, etc. from becoming part of the solution. Something based
on VNC, Citrix, Windows Terminal Server, etc. may be helpful: make
sure the computer doesn't know more than it's telling the visitor.
Also, these small programs will tend to re-use the same part of memory
repeatedly, making RAM remanence slightly less of an issue.
I care about privacy, but when somebody else asks me about it, I don't know how to respond. How do you reply to people who say, "But I have nothing to hide!"
as of a little while ago (this may have changed, I'm not a TiVO subscriber) you were't allowed to opt out.
Huh? You call them, and tell them you want to opt out, and it's done. What are you talking about?
Re:Should there be a GNU-Google?
on
Google vs. Evil
·
· Score: 2
Actually, they couldn't. There's a lot of code in some of the FSF's software that, when it was contributed, the FSF signed papers to the contributor agreeing that it would always remain free.
So... where are the bits? I wanna run this. Did these changes get put into -CURRENT (yay, I gotta MFC), or does Kenneth keep them on his web page, or does he seriously want us all to email him, or what?
I'm not entirely sure why I'm doing this. I have Warcraft 3 running fine on WineX under the Linuxulator, with a stock -STABLE box. Well, I say fine... there's a lot of sluggishness in the mouse, but I don't see fixing syscalls being a help for that.
Slashdot Mirror
Let me get this straight. You went to search the web and got conflicting, likely ill-informed, and inconclusive reports. So you went to Slashdot?
Passive is not a single-connection transfer. It uses an outbound connection for the data (from the POV of the client), whereas non-passive uses an inbound connection. That's why you normally need pasv to go through a firewall.
You may not have to search, but you have to take the time to increment and decrement the refcount each time. Depending on how things are done (for example, is the refcount incremented when a ref is pushed to the stack?), this can account for a whole lot of time. Throw in the time lost to locking if you're on a multithreaded system. Writes are slower than reads, too. You end up wasting a lot of CPU trying to maintain information about references, when you only need one bit of information: is it in use.
By contrast, modern generational garbage collectors use heuristics to minimize the searching done. Rather than having a lot of work to maintain information all the time that you're using it, you do a little work to GC it only in the GC.
It's true that refcout GC uses an extra integer per item of data, but how does that compare to the memory a regular GC must use to keep track of everything?
There isn't much "everything" to keep track of. Refcounts need to track how many outstanding references exist, all the time. (Note: Deferred refcounting and other systems can help with that, but that's a time savings, not a space savings.) On the other hand, GC only cares about one thing: is X garbage? What's more, it only cares about it while GC is taking place.
Baby's First Mark-and-Sweep GC (written by undergrad students first learing about GC) use only one bit of information per object during the program's life, plus one word per object in the longest chain at GC time. This is much less than the one word per every object that most refcounting schemes use! In other collectors, auxilary storage may be only needed during the GC, and is very low. Knuth talks about a number of variations on keeping memory usage very low (and only during GC) but those books are at my office right now.
GC technology was a big discussion topic when memory was scarce, so there's been a number of algorithms written to keep memory usage low during GC. Modern GCs use more than the theoretical minimum, of course, but nowhere near one word per object. In many cases, a bit for marking garbage can be taken from the type tag in typed languages. The type tag may be a certain length-- eg, a byte-- for effeciency purposes, and therefore may have a bit to waste. That means that the GCs memory overhead is effectively O(1)... just enough for the generation table!
Do some research on GCs; "garbage collection FAQ" on Google would be a start. Typically, reference counting takes more memory and more time than more sophisticated GC techniques. Besides, Python doesn't refcount anymore.
I keep hearing this from different people. I do a lot of replacing stuff with alternatives. For example, I don't have Windows, I have FreeBSD w/ OpenOffice, etc. But as a TiVo user, I don't think that I could switch to a computer-based solution.
There's just something about sitting down at the TV with a remote that can't be replaced. Also, there's something about the TiVo user interface that I think would become a lot more difficult in the translation to a computer.
It's not just about recording shows. It's about changing the way you watch TV.
1. Booting diagnostic programs
2. The first disk of your downloaded FreeBSD install. (I think Debian does the same thing: you d/l a couple of floppies and the rest gets fetched during the install. Beats having to burn a CD!)
Is this the tune to something we may know?
If you're a female, this could have the opposite effect intended.
Besides, I think there's laws against that and stuff. See the 1963 FCC Regulations Against Talking Dirty On The Phone To Somebody Who Doesn't Want To Hear It (RATDOTPTSWDWTHI).
You gotta be kidding me!
but selling unauthorized modifications to someone else's work is clearly not fair use: this is no different from a third party selling DVDs of modified scenes from the original work,
One of these is offering somebody suggestions regarding when to fast forward, etc, for a profit. The other is copying copyrighted works and distributing them for profit. Even if I take the **AA party line, only one of these robs the distribution chain of their fair share of the $$.
I'll agree with you there.
The last guy I interviewed for a programming job was a Perl coder. I asked him if he knew Lisp. He replied that he didn't. I then took out a half-page of Lisp code, and told him to add a particular feature. This wasn't a test to see if he got the answer right; I wanted to see how he handles stressful situations, and how well he can adapt to unusual circumstances. (He was hired, by the way, and is doing great.)
Note that not all OOPs are the same. Generic function OOP uses a little bit different structure than methods-on-objects OOPs. Writing one of these in a language designed for the other makes for icky code.
I disagree. The most popular languages today all more or less follow this (mostly because they're all Algol descendants), but not all languages do.
Alan J. Perlis said, "A language that doesn't affect the way you think about programming is not worth knowing." I agree. Once you've learned C, then learning Pascal or Perl is nothing. But I've seen a lot of people who are sharp-on in Perl that couldn't wrap their heads around functional languages. Ditto for teaching people OO for the first time.
If you're just learning languages by thinking they're all the same, then you're not learning languages. Don't write Perl code in Lisp; learn Lisp.
When I was in college, my roommate loved playing Starcon 2. He bought an AdLib, so he could enjoy the sound. I expected only music, but to my surprise, the MOD-based soundtrack played perfectly, and all the sound effects were identical to what I heard on an SB-compatible!
Anybody know anything about this?
I don't know whether or not Apple can do that... they've got a non-compete agreement keeping them out of the music production industry. (This is one reason why one of their first sound samples was called "sosumi", aka "So sue me".)
What's the model here? Does somebody walk into a secure facility, pick one up, use it to do some eyes-only investigation, and return it when they leave? What are you trying to secure against? Tampering, or somebody walking off with the data? The solution often depends on the threat model.
If your threat model is to prevent against tampering, then you may be better off exploring other options. For example, have you considered read-only media? How about having the tablets re-ghosted when they're returned, before they're re-issued? That can be done in an automated fashion without a whole lot of hassle-- primarily through the PXE extensions you've already investigated, combined with hardware at the docking stations.
If you're trying to keep people from walking off with data, then diskless isn't going to be the way to go. A lot of data gets left in RAM after power is removed. (See Gutman, P., "Secure Deletion of Data from Magnetic and Solid-State Memory," Proceedings of the Sixth USENIX Security Symposium, July 1996, or do a Google search for "RAM remanence".) You may have seen some computers-- notably the old Macs-- that would power up with their last display still on the screen!
Also, if you have a totally stateless box driven by a wireless LAN, then some shmuck can easily sit in a van a half-mile away with his laptop and find out everything you're wanting to keep private. Stateless booting means that your encryption has to be bootstrapped! A lot of naive ways of doing this exist, such as sending a root filesystem with encryption keys already on it. Some of these open themselves up to passive attacks. Even more sophisticated techniques, such as DH, still are totally vulnerable to active attacks (like the guy in the van pretending to be one of your tablets asking to be bootstrapped).
I seriously suggest you rethink your security model. The Windows compatibility is a big problem. It keeps some of the latest crypto filesystems, etc. from becoming part of the solution. Something based on VNC, Citrix, Windows Terminal Server, etc. may be helpful: make sure the computer doesn't know more than it's telling the visitor. Also, these small programs will tend to re-use the same part of memory repeatedly, making RAM remanence slightly less of an issue.
Recognized over most of Europe, and used to be common in the States until, oh, I guess 1900 or something.
I care about privacy, but when somebody else asks me about it, I don't know how to respond. How do you reply to people who say, "But I have nothing to hide!"
Actually, they couldn't. There's a lot of code in some of the FSF's software that, when it was contributed, the FSF signed papers to the contributor agreeing that it would always remain free.
If I recall the VFW API right, it's very easy to write.
Terrific. Now go measure fifths of an inch and tell me again.
Has anybody heard the term "larrikin moon" (used in the article) before? This is a new one for me, and Google came up empty.
So... where are the bits? I wanna run this. Did these changes get put into -CURRENT (yay, I gotta MFC), or does Kenneth keep them on his web page, or does he seriously want us all to email him, or what? I'm not entirely sure why I'm doing this. I have Warcraft 3 running fine on WineX under the Linuxulator, with a stock -STABLE box. Well, I say fine... there's a lot of sluggishness in the mouse, but I don't see fixing syscalls being a help for that.
What makes you think he's on a crusade? Maybe he just wants to conveniently maintain his credit card account!