Hi, I've uploaded some of my fascinating Quake Live screenshot collection in this tiny imgur gallery: http://imgur.com/a/yjxRl
So the first problem went away after 1. upgrading kernel and 2. disabling hardware acceleration in Flash.
Now it's only like once a day that I have pink walls in game or everything is upside down, but game restart fixes it.
"Fixes" as in: there still are colorful stains on some walls but no problem, I can play at least, all praise nVidia's driver!
Before you say it's a hardware problem: I've had 3 different GPUs from 2 different manufacturers (all GTX 570) and 2 motherboards (same model, free upgrade to "B3" stepping Intel chipset last year). Shuffling hardware didn't fix anything, but kernel+driver update did a little.
Using VDPAU on certain input (either in MPlayer or enabling accel in Flash) makes whole system unstable, which basically is an exploitable DOS attack against the nvidia module.
If you say ATI/AMD cards have more issues, I really pity their users:(
A good exemple that sometimes the market is unable to find the most optimal solution and someone has to regulate.
Actually, if there were no patents, free market would produce a standard by this time. Smaller manufacturers would make their devices compatible with chargers that people already have, from big companies, most probably Nokia.
But in reality, noone is permitted to use the same connector because they'd have lawyers on their back. So the tiny players started to use standard mini-USB, because they aren't in business of selling chargers.
Take patents aside and see the market work. Until then, your mention of market being unable to do things is true. But reacting with regulation is hiding, not resolving the main problem.
Here in Poland it's already cheaper to buy a CPU fast enough to make decompression 4 times quicker then to pay for a link that can download a typical "business" (meaning tons of pictures - bells and whistles instead of real content) website in 10 seconds... The web server running such sites will benefit from using smaller files as well.
Besides, the algorithm will evolve and evetually decompression won't be so expensive anyhow.
Okay, it's 8 seconds to display 800 KiB file. But the same file without compression would be a 1142 KiB file (assuming the advertised 30% gain). It takes 85 seconds to download the additional 342 KiB of data on a standard modem (4 KiB/s). Compare 8 to 85.
My link is giving me 10 KiB/s (I pay $40 monthly for it), it's still 34 seconds compared to 8 seconds to decompress it. Think about it.
> It's not like you could have tunned this system to make it very secure
Well, just umount/dev/shm - you're done, the exploit won't work, tada.wav:) It's the kind of tuning anyone can do instantly and it will give you protection from script kiddies. Need I say more?
Well, the problem is that you need to recompile the program at all. Compilation for Itanium WILL give you performance boost no matter if your program was written with parallelism in mind or not. The compiler can store more variables in registers (meaning you can get tens of times faster execution of functions doing many operations on a small set of variables in a loop), can use faster instructions to fill memory with zeroes (e.g. for calloc), to copy structures or arrays, system libraries for sure have much faster array functions (bcopy, wmemset, etc.) and so on.
All of that doesn't require any real architecture change, just adding of few simple (for an Intel engineer;)) circuits, and (this is the tricky part) changing instruction set breaking compatibility with existant precompiled software. Also it's worth noting, some software requires porting (e.g. if you use int to store a void *, which simply won't work anymore).
Let's face facts - Itanium isn't specialised at all, it just isn't backwards-compatible, meaning users won't be able to play their old games on it. That's one of the two reasons (second is the price) for users NOT to switch to ia64, but switch to x86-64, which promises performance boost at little cost (that's for us penguins) and compatibility (for those installing 32-bit Windows on an Opteron) with performance boost in the future (when XP for x86-64 will come out, when there will be games for it and so on). x86-64 promises "fluent" switch, with a stage of 64-bit and good old 32-bit applications mixed in one system. In few years from now x86-64 will drop support for some obsolete instructions, work modes or whatever and we all will be in the age of the Itanium... but not with the Itanium.
What you're forgetting is that PaX stil uses software emulation of this NX-bit. They have paging-based and segmentation-based protection. First make my shell scripts run few times slower, second seems to be a little faster, but limits the address space for programs. Both are slowing down the whole system, one is limiting some programs' functionality. What newest Opterons and Pentium 4s do is giving you a hardware implementation. Set one bit and you're done, tada.wav. Noone is preventing you from randomizing addres bases, hiding addresses, restricting some privileges or capabilities, or even disabling access rights to/root to non-root users! They give you a great thing - non-executable pages, which you want to use and are using, with no performance cost (which you are accepting, but I'm not). Add few other protection mechanisms to this one and you've patched 95% of all exploitable bugs. Without hardware support, it would be a question of security vs. performance. Now this question is gone for non-executable pages. Next step would be to include ultra-fast random number generators in processors or even hardware support for processes/threads with different access privileges to memory, low-level IO, etc. etc. NX bit introduction is good for you, appreciate it and wait for more (maybe something like AS/400 features mentioned earlier?):)
I hardly doubt it that you will even notice a difference if Windows were to finally decise to include a software solution rather than using the hadware one provided by AMD
First thing, Windows XP SP2 DOES include a software solution which is enabled if you tell the system to protect programs and your processor doesn't have the bit in question.
Second thing, well, the protection is DISABLED by default for all programs except for a few network daemons (even on the newest AMD and Intel processors with non-executable page bit), so it's true you won't notice a difference - but enable it on your non-NX-processor and see for yourself (I haven't as I don't have Windows;)) - it should be slower, where NX-based protection would be as fast as no protection. As a Linux power-user I know Linux+grsecurity/Linux+openwall/GCC+IBM/any-other protection is slower than plain unpatched Linux/glibc/gcc on x86 without the NX wonder-bit. For some programs it's slower to a point where I can actually "see" it without measuring, that's just the way any software protection works. It can introduce little impact on performance, but there will be some impact anyhow.
What difference does the OS make? It's not hard to guess it's too much money for them to lose upon OS crash/hardware failure, thus they surely have backup hardware running in parallel. It's the custom software that is buggy and causing all this mess.
metux mpm is in a state similar to perchild. It conflicts with mod_ssl, it breaks on persistent connections, it's untested, unfinished, unmantained. I'm on their mailing list, but there haven't been any traffic since months and months now. Instead, people are using a patched Apache with a patched Linux kernel, but I haven't heard of any Linux distro supporting such crazy combo.
I'm using Apache 2.0 for various reasons, but I know one reason not to switch from 1.3: mod_bandwidth (I use "real" virtual hosts with separate IP addresses for sites that need bandwidth limits, but with Apache 1.3 this was so much simpler...).
It would be even nicer if they open source it:) I still play QuakeForge and I'm not alone at it, now UOForge is the new way to go! It could give the company many peoples appreciation, which combined with good rumours spread by these people can be a good advertising technique.
I guess it's rather related to some combination of tables, because Slashdot is rendered two times wider when I am logged in, and as far as I can tell, Slashdot is all left-to-right:)
I'm using Mozilla 1.7.3. I guess it's rather Gecko's problem, so it should be present at least in Firefox 0.9.
And to answer the main topic question - I'd switch if it were faster rendering lots of nested tables, not starting up (if FF really is starting up faster). I tried watching phpBB output in Mozilla and Firefox on Pentium 2 350 MHz. It took both programs 2 seconds to start up, but 10 seconds to display a long forum topic. Until Mozilla uses up-to-date (thus same of FF's) Gecko engine, I won't switch only to have 1/10th of it's features and the same slowness, sorry.
Here in Poland there was a proposal similar to the one described above. It contained a clause that all the information logged by providers have to be available to the police in unencrypted form, which basically means: if one of my clients uses SSH to connect to some foreign host in Iraq or Afghanistan, it's my responsibility to decrypt it and present raw transmission data to the authorities.
Not mentioning the amount of storage needed to keep everything transmitted over a relatively slow network (a megabit per second link can give you 3754 terabytes a year), imagine the processing power needed to decrypt few SSH sessions. Then imagine a price of link to an IAP having 100 clients, each of which have to pay for 1/100 of IAP-s supercomputer;)
I haven't RTFA, but I hope the EU won't try to push a project similar to the one which didn't work in Poland (for now, I'm sure there's new version in the works).
Slashdot Mirror
Hmm, is there a way to take over ownership of a comment? I posted as AC by mistake.
Hi, I've uploaded some of my fascinating Quake Live screenshot collection in this tiny imgur gallery: http://imgur.com/a/yjxRl
So the first problem went away after 1. upgrading kernel and 2. disabling hardware acceleration in Flash.
Now it's only like once a day that I have pink walls in game or everything is upside down, but game restart fixes it.
"Fixes" as in: there still are colorful stains on some walls but no problem, I can play at least, all praise nVidia's driver!
Before you say it's a hardware problem: I've had 3 different GPUs from 2 different manufacturers (all GTX 570) and 2 motherboards (same model, free upgrade to "B3" stepping Intel chipset last year). Shuffling hardware didn't fix anything, but kernel+driver update did a little.
Using VDPAU on certain input (either in MPlayer or enabling accel in Flash) makes whole system unstable, which basically is an exploitable DOS attack against the nvidia module.
If you say ATI/AMD cards have more issues, I really pity their users :(
A good exemple that sometimes the market is unable to find the most optimal solution and someone has to regulate.
Actually, if there were no patents, free market would produce a standard by this time. Smaller manufacturers would make their devices compatible with chargers that people already have, from big companies, most probably Nokia.
But in reality, noone is permitted to use the same connector because they'd have lawyers on their back. So the tiny players started to use standard mini-USB, because they aren't in business of selling chargers.
Take patents aside and see the market work. Until then, your mention of market being unable to do things is true. But reacting with regulation is hiding, not resolving the main problem.
What will the Ubuntu version look like when the year 3000 comes? Will it go from 999.10 to 1000.04?
If you want to make it more portable, replace "xfz" with "xzf" - f has to be just before file name in older/other implementations.
Here in Poland it's already cheaper to buy a CPU fast enough to make decompression 4 times quicker then to pay for a link that can download a typical "business" (meaning tons of pictures - bells and whistles instead of real content) website in 10 seconds... The web server running such sites will benefit from using smaller files as well.
Besides, the algorithm will evolve and evetually decompression won't be so expensive anyhow.
Okay, it's 8 seconds to display 800 KiB file. But the same file without compression would be a 1142 KiB file (assuming the advertised 30% gain). It takes 85 seconds to download the additional 342 KiB of data on a standard modem (4 KiB/s). Compare 8 to 85.
My link is giving me 10 KiB/s (I pay $40 monthly for it), it's still 34 seconds compared to 8 seconds to decompress it. Think about it.
One of them with over $10000 bid has >4000 rating, I guess he actually is willing to pay.
> It's not like you could have tunned this system to make it very secure
/dev/shm - you're done, the exploit won't work, tada.wav :) It's the kind of tuning anyone can do instantly and it will give you protection from script kiddies. Need I say more?
Well, just umount
Well, the problem is that you need to recompile the program at all. Compilation for Itanium WILL give you performance boost no matter if your program was written with parallelism in mind or not. The compiler can store more variables in registers (meaning you can get tens of times faster execution of functions doing many operations on a small set of variables in a loop), can use faster instructions to fill memory with zeroes (e.g. for calloc), to copy structures or arrays, system libraries for sure have much faster array functions (bcopy, wmemset, etc.) and so on.
;)) circuits, and (this is the tricky part) changing instruction set breaking compatibility with existant precompiled software. Also it's worth noting, some software requires porting (e.g. if you use int to store a void *, which simply won't work anymore).
All of that doesn't require any real architecture change, just adding of few simple (for an Intel engineer
Let's face facts - Itanium isn't specialised at all, it just isn't backwards-compatible, meaning users won't be able to play their old games on it. That's one of the two reasons (second is the price) for users NOT to switch to ia64, but switch to x86-64, which promises performance boost at little cost (that's for us penguins) and compatibility (for those installing 32-bit Windows on an Opteron) with performance boost in the future (when XP for x86-64 will come out, when there will be games for it and so on). x86-64 promises "fluent" switch, with a stage of 64-bit and good old 32-bit applications mixed in one system. In few years from now x86-64 will drop support for some obsolete instructions, work modes or whatever and we all will be in the age of the Itanium... but not with the Itanium.
I read it as "Titanic" :)
What you're forgetting is that PaX stil uses software emulation of this NX-bit. They have paging-based and segmentation-based protection. First make my shell scripts run few times slower, second seems to be a little faster, but limits the address space for programs. Both are slowing down the whole system, one is limiting some programs' functionality. /root to non-root users! They give you a great thing - non-executable pages, which you want to use and are using, with no performance cost (which you are accepting, but I'm not). Add few other protection mechanisms to this one and you've patched 95% of all exploitable bugs. :)
What newest Opterons and Pentium 4s do is giving you a hardware implementation. Set one bit and you're done, tada.wav. Noone is preventing you from randomizing addres bases, hiding addresses, restricting some privileges or capabilities, or even disabling access rights to
Without hardware support, it would be a question of security vs. performance. Now this question is gone for non-executable pages. Next step would be to include ultra-fast random number generators in processors or even hardware support for processes/threads with different access privileges to memory, low-level IO, etc. etc. NX bit introduction is good for you, appreciate it and wait for more (maybe something like AS/400 features mentioned earlier?)
I hardly doubt it that you will even notice a difference if Windows were to finally decise to include a software solution rather than using the hadware one provided by AMD
First thing, Windows XP SP2 DOES include a software solution which is enabled if you tell the system to protect programs and your processor doesn't have the bit in question.
Second thing, well, the protection is DISABLED by default for all programs except for a few network daemons (even on the newest AMD and Intel processors with non-executable page bit), so it's true you won't notice a difference - but enable it on your non-NX-processor and see for yourself (I haven't as I don't have Windows ;)) - it should be slower, where NX-based protection would be as fast as no protection. As a Linux power-user I know Linux+grsecurity/Linux+openwall/GCC+IBM/any-other protection is slower than plain unpatched Linux/glibc/gcc on x86 without the NX wonder-bit. For some programs it's slower to a point where I can actually "see" it without measuring, that's just the way any software protection works. It can introduce little impact on performance, but there will be some impact anyhow.
Moderators, you call parent Insightful? :)
The truth was: PEBKAC :)
What difference does the OS make? It's not hard to guess it's too much money for them to lose upon OS crash/hardware failure, thus they surely have backup hardware running in parallel. It's the custom software that is buggy and causing all this mess.
...and wanted to try out the speach of Opera 8-beta.
metux mpm is in a state similar to perchild. It conflicts with mod_ssl, it breaks on persistent connections, it's untested, unfinished, unmantained. I'm on their mailing list, but there haven't been any traffic since months and months now. Instead, people are using a patched Apache with a patched Linux kernel, but I haven't heard of any Linux distro supporting such crazy combo. I'm using Apache 2.0 for various reasons, but I know one reason not to switch from 1.3: mod_bandwidth (I use "real" virtual hosts with separate IP addresses for sites that need bandwidth limits, but with Apache 1.3 this was so much simpler...).
Wrong, in South Korea old people actually talk to their bikes instead of using phones. Germany is so yesterday.
Some of us are too shy to call vibrator-vagina-maker's tech-support... I admire you.
It would be even nicer if they open source it :) I still play QuakeForge and I'm not alone at it, now UOForge is the new way to go! It could give the company many peoples appreciation, which combined with good rumours spread by these people can be a good advertising technique.
I guess it's rather related to some combination of tables, because Slashdot is rendered two times wider when I am logged in, and as far as I can tell, Slashdot is all left-to-right :)
I'm using Mozilla 1.7.3. I guess it's rather Gecko's problem, so it should be present at least in Firefox 0.9.
And to answer the main topic question - I'd switch if it were faster rendering lots of nested tables, not starting up (if FF really is starting up faster). I tried watching phpBB output in Mozilla and Firefox on Pentium 2 350 MHz. It took both programs 2 seconds to start up, but 10 seconds to display a long forum topic. Until Mozilla uses up-to-date (thus same of FF's) Gecko engine, I won't switch only to have 1/10th of it's features and the same slowness, sorry.
Here in Poland there was a proposal similar to the one described above. It contained a clause that all the information logged by providers have to be available to the police in unencrypted form, which basically means: if one of my clients uses SSH to connect to some foreign host in Iraq or Afghanistan, it's my responsibility to decrypt it and present raw transmission data to the authorities.
;)
Not mentioning the amount of storage needed to keep everything transmitted over a relatively slow network (a megabit per second link can give you 3754 terabytes a year), imagine the processing power needed to decrypt few SSH sessions. Then imagine a price of link to an IAP having 100 clients, each of which have to pay for 1/100 of IAP-s supercomputer
I haven't RTFA, but I hope the EU won't try to push a project similar to the one which didn't work in Poland (for now, I'm sure there's new version in the works).