It's not that the app is written crappily. Understand that when you mark a message as junk mail, you have to update the database which stores the word occurence frequencies for spam and ham. If a message contains 1000 unique tokens (including server names, non-standard headers), then you're going to need to potentially update the database file in 1000 different places.
Presumably this database file is memory mapped, so really it's up to the OS to cache and batch up these frequent writes.
What's really needed is a way to mark a file with a special attribute that says: this file gets updated a lot, don't flush the buffer cache so damn often.
A way to do that could solve a lot of problems that many different applications have when doing stuff to a file on flash media.
In either case, the device would have to be aware of the data contained within to know whether or not to accept/deny a request.
Equipped with said smarts, a USB-based device could "correctly" respond to anything asked of it by the host controller, but only returning 0s or garbage if it wants to not share some specific data.
The type of connection between the host and device is immaterial.
The only way the device itself transparently could encrypt/decrypt contents would be with some kind of password/key interface on the device itself. This would make the device somewhat biggish to support the PIN/password entry area. Also you'd have to add a few more chips to the device to support the cryptographic functions which could definitely change the footprint and power requirements... unless some chip vendor thinks it'd be neat to add those features to the next set of flashUSB single-chip solutions.
But if the protection is software based than you don't need to worry about tampering (since the data on the fob is useless otherwise). Existing chipsets would suffice.
Windows EFS is the answer, really. It uses 3DES or AES and it's easy to manage. You get it for free with 2000 and XP Pro... so... why not use it?
CSRSS.EXE is the equivalent to X. Only in this case X is using all kernel drivers (framebuffer driver, GPM/HID-style input). The framebuffer driver if you will is a combination of WIN32.SYS and whatever components are provided by your video card.
Each windows process with an application window makes (what I understand) to be a kernel call directly into WIN32.sys to get a handle so it can talk to CSRSS.exe and register itself to receive events from there, and to post things. It uses some kind of IPC that is abstracted by the linked-in portions of the Win32 API in user space.
Not only does CSRSS handle drawables and events however. It also supports the creation of user threads, spawning processes and syncronozation objects and some other things I can't remember.
Anyway the important thing is that it's not magic. It works just like X11 does just that the API doesn't have the option of being network transparent. The whole winsock thing was added "on the side" and has nothing to do with these APIs.
if you have a CPU-intensive thread with a large working set that you don't want moved in and out of a CPU's cache, then it makes sense to "dedicate" a CPU to that task.
OTH it'd be nice if the operating system scheduler could figure stuff like that out automatically by looking a thread's resource usage history, but I think that most multitasking OSs can do this anyway.
Are there any exploits in the wild that detect and get around Deep Freeze? If not then it is probably a better solution than most. (Sort of like Windows File Protection... )
You can get this kind of seperation if you go to a three tier architecture... one where apache is a user that has large read-only and resource management capabilites... and then a "data" user runs behind the scenes in the internal (or external) application layer running J2EE or Zope or something. Here you provide an application-level disconnect. You control the code that turns public access requests as one user over an internal network into operations running as a user with database trampling privledges.
Ultimately you have to define a "policy" which validates the external requests and proxies access... through some sort of abstract permission based hierarchy or in this case with front-end and middle-tier validation/business logic.
As I recall from the AI classes I took one of the big limiting factors to the success or failure or some net-based AIs is determining the kinds of input and output layer nodes... what to feed in and what default node weights to set.
More concise and domain specific inputs helps to make the net more flexible and less artificially constrained by input vectors (too many extra degrees of freedom = excellent but unrealistic "fit")
Well then someone will eventually figure out how to trick Adobe Acrobat into saving the decrypted form of the document somewhere. At that point the document is untrackable. All it takes is once.
OSX gives you enough rope to hang yourself with. And just enough extra to moor a yacht.
They were just nice enough to give you a default environment that doesn't need tinkering and you'd never have to pop the hood. This is why you pay for it.
Who the hell thought that using a DVD array would be better than a disk array/SAN? Just the thought of a mechanical anything moving constantly to keep up with random access makes my head hurt. If you don't have at least 16 DVD readers it's pointless. You'd need at least that just to saturate SCSI bandwidth. And all those moving parts. GAASAAAH.
And $10000+ when 1TB RAID would cost the same at the time of purchase... double GAAAH.
*Cough cough* *wheeze*
I sure hope they fired the guy who thought of that brilliant idea.;-)
This is probably overkill, but it is a really cool piece of equipment, and it doesn't rely on shitty windows software to do it's job. Unfortunately it costs $10000 fully loaded with 4 DVD-RW drives.
I just got this buzzword bingo vibe. You see it a lot when anything relating to apple comes up. You know, the speculation runs rampant whenever any tiny thing at the apple store changes... sigh.
What the hell are you talking about... link it into search engines? Christ almighty.... oh and nice bit about the calculators, that always gets you modded up a few times by HP calc die-hards around here.
And office integration. Some nice change tracking/merging features (office specific). That's basically it. It makes sense Microsoft wants the company, it's perfect (especially so they can ditch LiveMeeting and Sharepoint)
...provided you make a very small patch. Basically you need to a) make sure all visibility and collision routines don't take shortcuts to determine what sector you're in vs. what is visible. That is, I don't do the LOS or height collision test unless my sector (according to my sector attribute) is visible from your sector (according to your sector attribute). And that sector attribute updates ONLY occur when crossing two-sided linedefs. The code may already be this way... I haven't checked.
The only thing you need to do is to make any necessary changes to the BSP tree builder (for your level editor) to convince it that overlapping sectors are okay, and to always make them NOT visible to each other. I think it'd even be okay if they auto-partition them into subsectors... so long as the checks operate at that level.
Slashdot Mirror
It's not that the app is written crappily.
Understand that when you mark a message as junk mail, you have to update the database which stores the word occurence frequencies for spam and ham. If a message contains 1000 unique tokens (including server names, non-standard headers), then you're going to need to potentially update the database file in 1000 different places.
Presumably this database file is memory mapped, so really it's up to the OS to cache and batch up these frequent writes.
What's really needed is a way to mark a file with a special attribute that says: this file gets updated a lot, don't flush the buffer cache so damn often.
A way to do that could solve a lot of problems that many different applications have when doing stuff to a file on flash media.
In either case, the device would have to be aware of the data contained within to know whether or not to accept/deny a request.
Equipped with said smarts, a USB-based device could "correctly" respond to anything asked of it by the host controller, but only returning 0s or garbage if it wants to not share some specific data.
The type of connection between the host and device is immaterial.
The only way the device itself transparently could encrypt/decrypt contents would be with some kind of password/key interface on the device itself. This would make the device somewhat biggish to support the PIN/password entry area. Also you'd have to add a few more chips to the device to support the cryptographic functions which could definitely change the footprint and power requirements... unless some chip vendor thinks it'd be neat to add those features to the next set of flashUSB single-chip solutions.
But if the protection is software based than you don't need to worry about tampering (since the data on the fob is useless otherwise). Existing chipsets would suffice.
Windows EFS is the answer, really. It uses 3DES or AES and it's easy to manage. You get it for free with 2000 and XP Pro... so... why not use it?
Someone correct me if I'm wrong here...
CSRSS.EXE is the equivalent to X.
Only in this case X is using all kernel drivers (framebuffer driver, GPM/HID-style input).
The framebuffer driver if you will is a combination of WIN32.SYS and whatever components are provided by your video card.
Each windows process with an application window makes (what I understand) to be a kernel call directly into WIN32.sys to get a handle so it can talk to CSRSS.exe and register itself to receive events from there, and to post things. It uses some kind of IPC that is abstracted by the linked-in portions of the Win32 API in user space.
Not only does CSRSS handle drawables and events however. It also supports the creation of user threads, spawning processes and syncronozation objects and some other things I can't remember.
Anyway the important thing is that it's not magic. It works just like X11 does just that the API doesn't have the option of being network transparent.
The whole winsock thing was added "on the side" and has nothing to do with these APIs.
if you have a CPU-intensive thread with a large working set that you don't want moved in and out of a CPU's cache, then it makes sense to "dedicate" a CPU to that task.
OTH it'd be nice if the operating system scheduler could figure stuff like that out automatically by looking a thread's resource usage history, but I think that most multitasking OSs can do this anyway.
Right?
Are there any exploits in the wild that detect and get around Deep Freeze? If not then it is probably a better solution than most. (Sort of like Windows File Protection... )
You can get this kind of seperation if you go to a three tier architecture... one where apache is a user that has large read-only and resource management capabilites... and then a "data" user runs behind the scenes in the internal (or external) application layer running J2EE or Zope or something.
Here you provide an application-level disconnect. You control the code that turns public access requests as one user over an internal network into operations running as a user with database trampling privledges.
Ultimately you have to define a "policy" which validates the external requests and proxies access... through some sort of abstract permission based hierarchy or in this case with front-end and middle-tier validation/business logic.
As I recall from the AI classes I took one of the big limiting factors to the success or failure or some net-based AIs is determining the kinds of input and output layer nodes... what to feed in and what default node weights to set.
More concise and domain specific inputs helps to make the net more flexible and less artificially constrained by input vectors (too many extra degrees of freedom = excellent but unrealistic "fit")
Do tell.
Well then someone will eventually figure out how to trick Adobe Acrobat into saving the decrypted form of the document somewhere.
At that point the document is untrackable. All it takes is once.
You're comparing Fluxbox, which is just a WM with a bare environment, to GNOME (which is a WM + "gnome-session").
Metacity probably couldn't run on it's own without gnome-session or a large amount of configuration.
But you could definitely run fluxbox _instead_ of metacity inside a GNOME session. And you'd still have the same memory footprint, I imagine.
So really it's that a minimal WM + a decent starting configuration is a better fit for you than the full blown gnome-powered nautilis desktop.
OSX gives you enough rope to hang yourself with.
And just enough extra to moor a yacht.
They were just nice enough to give you a default environment that doesn't need tinkering and you'd never have to pop the hood. This is why you pay for it.
And the worst kind... he's looking for an Apple-fanboy dicksucking.
I undid moderations to post this, btw.
I guess you can call it an art colony. Isn't it right by SUNY/Dutchess? That's the reason why it's changed, if anything.
Make no mistake.
No system can be completely protected purely in software running on untrusted hardware.
The real website is here . Don't visit any imposter site with "weird" characters in the URL.
Who the hell thought that using a DVD array would be better than a disk array/SAN? Just the thought of a mechanical anything moving constantly to keep up with random access makes my head hurt.
;-)
If you don't have at least 16 DVD readers it's pointless. You'd need at least that just to saturate SCSI bandwidth. And all those moving parts. GAASAAAH.
And $10000+ when 1TB RAID would cost the same at the time of purchase... double GAAAH.
*Cough cough* *wheeze*
I sure hope they fired the guy who thought of that brilliant idea.
Product Brochure
This is probably overkill, but it is a really cool piece of equipment, and it doesn't rely on shitty windows software to do it's job. Unfortunately it costs $10000 fully loaded with 4 DVD-RW drives.
I just got this buzzword bingo vibe. You see it a lot when anything relating to apple comes up. You know, the speculation runs rampant whenever any tiny thing at the apple store changes... sigh.
What the hell are you talking about... link it into search engines? Christ almighty.... oh and nice bit about the calculators, that always gets you modded up a few times by HP calc die-hards around here.
And office integration.
Some nice change tracking/merging features (office specific). That's basically it. It makes sense Microsoft wants the company, it's perfect (especially so they can ditch LiveMeeting and Sharepoint)
So I don't know how people can confuse those two words. Please learn the difference. It's _very_ annoying, even more so than loose/lose
...provided you make a very small patch. Basically you need to a) make sure all visibility and collision routines don't take shortcuts to determine what sector you're in vs. what is visible. That is, I don't do the LOS or height collision test unless my sector (according to my sector attribute) is visible from your sector (according to your sector attribute). And that sector attribute updates ONLY occur when crossing two-sided linedefs. The code may already be this way... I haven't checked.
The only thing you need to do is to make any necessary changes to the BSP tree builder (for your level editor) to convince it that overlapping sectors are okay, and to always make them NOT visible to each other. I think it'd even be okay if they auto-partition them into subsectors... so long as the checks operate at that level.