Re:what a predicament ...
on
Linux Turns 10
·
· Score: 1
> I'm not the kernel, but if I were, the top of my wishlist would be a VM that actually works.
I'm not sure why, exactly, your post was moderated up (perhaps because it was lengthy?) -- you've not given any reasoning for saying the VM subsystem doesn't work, you just give an ad hominem (which, judging by the moderation, was effective in itself). Care to explain what, exactly, is wrong with the VM subsystem?
I'll field this one.
I write this as a kernel compiles in the background. 2.2.19. I was testing Web-1 with a number of 2.4 kernels, but it has freaked, oopsed, and panicked its way to 3 crashes in the last month, so I'm packing it away for awhile. 2.4 does okay on my desktop and the squid servers, but lions' dens we call web servers just eat it alive.
The oopsen all came out of the disk/io cache and most are posted to lkml, but it's like talking to a wall.:(
I do think it will work out in the end, though. We had fun with the 2.2.x Intel and 3Com NIC drivers up until around 2.2.12 or so. That's the curse of a high-traffic network -- you get to be the unwilling guinea pig for all of those 'close, but not quite' kernels.
Has anyone done a cost-efficiency comparison of dual-cpu performance vs. a simple cpu when considering the costs involved (special SMP boards, etc.) In otherwords is it more economical to buy two web servers or one smp server with tons of ram?
As others here mentioned, the choice between SMP and multiple boxes depends a lot on your current bottlenecks and the amount of communication between tasks.
For instance, if your web server is memory speed or I/O bound, and updates are managed by a central authority (like an SQL server), go for a pair of servers. That doubles your system bus and disk subsystem capacity over a single SMP system.
On the other hand, if your server is CPU bound and/or requires fast communication between tasks (like updates to on-disk data), go for the SMP. Sure, the board is more expensive, but you only need one (same goes for drives, cases, network cards, etc).
Web servers generally fall into the I/O bound category, so two servers is probably your better bet unless you handle quite a bit of dynamic content.
The Pentium 4 is MUCH faster than Athlon on FP. [clip] SPEC benchmarks are considered to be considerably more scientific and reliable than the toy-type unsophisticated benchmarks you see on peecee hardware review sites.
They are still benchmarks, though. If you look at a broad range of benchmarks, the pattern is exactly what you would expect, considering the technology.
The P4 excels in predictable, tight-looped applications, especially en/decoding tasks. In broad-looped, unpredictable, or multi-tasking situations, that deep pipeline and RDRAM prove lethal as the processor burns off an insane number of CPU cycles getting its act back together after each misprediction.
I'm sure the P4 kicks butt in SPEC faceoffs, but SPEC benchmarks don't help me with games and work.
We do not grant freedom to those people who would use or have used it to take others freedom away.
Sure we do. The next time a Klan rally comes to your town, just try to stop it. People try to every time, and I love the irony of them protesting (speaking freely) to ban the Klan's free speech. I may not like what they say, but there are plenty of idiots in the world; I'm not about to start the censorship snowball over it.
Linus actually mentioned this on LKML when the Crusoe was first announced. Linux on the bare metal was faster, but not dramatically. The translation caching absorbed most of the overhead.
The speed increase isn't worth the loss in flexibility. The primary advantage Transmeta has is the freedom to completely redesign the chip and instruction set to reduce heat and power consumption. Crusoe has enough speed for its target market, so running Linux on the bare metal wouldn't improve much.
There are real PCI modems around, but you have to look carefully. I picked up the USR 56K Performance Pro Modem: http://www.usr.com/products/home/home-product.asp? sku=3CP5610A
I'm sure you can finder cheaper ones if you try. It works just like an ISA modem.
BTW: is sending a threating C&D letter the same a 'suing'?
No, a C&D is the first step at resolution. It only threatens to sue if the alleged abuse continues.
The purpose of a C&D is two-fold. Obviously, it's much cheaper than a lawsuit and usually effective. In addition, if you need to take the abuser to court at a later time, the company can submit the C&D as proof that has made efforts to protect its trademark. It also proves that the abuser knew about the violation and continued to abuse it anyway (for punitive damage).
Come again? The late 2.3 and early 2.4-tests have been a pain in the ass. You must've been oblivious to the entire ext2fs corruption and GINORMOUS memory leaks. Even with SMP, they were unsuitable for production useage.
It depends on your needs. Our squid servers (abused by ~1.5 million hits/day) have used test5 since August.
I admit, I haven't noticed fs corruption on them. I have noticed ps not accounting for 30-50MB of memory, but it's an acceptable loss for the ~50% performance boost.
"Can't you see that Open Source/AMD is better, fast, and cheaper?"
"Yeah, but change is bad..."
Oh please. I've used AMD in my home system since the early K6's, but I still won't put them in a server. Why?
No ECC functionality - Kind of useful in a server.
No MicroATX boards with video and a good 100BaseT chip - Pretty much required for a 1U server.
No multi-processor chipsets - Yeah, yeah, it's "just over the horizon." That doesn't help me right now.
This is an improvement over the K6/K6-2 chipsets, which could not be set to power on when plugged in. You power cycled the box and it stayed off. That really sucked.
I have no problem with AMD CPUs, but nobody seems interested in targetting the server market with them.
Well, tough snuff. That's like saying we must ban automobiles, since that will put horse and carriage makers out of business.
You talk about that as if they didn't try. To the contrary, many communities passed laws and regulations to outright ban or ridiculously restrict automobiles. The most common example was pushed by the Farmer's Anti-Automobile Society (FAAS) in Pennsylvania where:
"Automobiles travelling on country roads at night must send up a rocket every mile, then wait ten minutes for the road to clear."
"If a driver sees a team of horses, he is to pull to one side of the road and cover his machine with a tarpaulin or dust cover that has been painted to blend into the scenery."
"In the event that a horse refuses to pass a car on the road, the owner must take his car apart and conceal the parts in the bushes."
If I'm running a query that is updating a number of financial records to place orders and proccess information, I don't want that process to bomb out half finished
Fair enough, but don't buy into the myth that the majority of dynamic pages on the web involve critical information.
The majority of dynamic web pages are a dizzying array of counters, polls, news, comments, and other trinkets. Given how rarely a box goes down unexpectedly, losing a vote or comment is trivial.
Even in an e-commerce situation, not all data needs transactions. Take product info, for example. It's almost all read-only and staff-updated.
Mysql works wonderfully for dodging concurrency problems in a load-balanced cluster and offers faster, easier ways to sort data than a file system. BDB or no, I wouldn't trust it with financial transactions, but it's great for the splash of interaction most sites need.
> Why is it that a *nix box getting compromised = 'Excellent, now we can patch the hole'
I think that's the precise difference, right there. Judiging by preliminary reports, it looks like one-click execution of attachments aided (to put it lightly) this breach. That same design flaw spawned Melissa and the Love virus.
At this point, I think most would agree that the potential damage from that "feature" is far greater than the potential convenience and time savings. Yet here we are, many months and two huge outbreaks later, and that same feature is still enabled by default.
> but an NT machine = their security "sucks"?
It sucks, but not simply because it's NT, or closed source, or because it has security bugs and design flaws. It sucks because they don't fix it and you can't.
Unix used to be like that, actually. The Morris Worm was their sign to clean up their act and patch some holes. Microsoft ignored the closest Windows equivalent, Melissa. They ignored the Love virus, too. Maybe this will hit home enough to actually spur some action.
Anyway, legally binding contracts must be an exchange between the parties. IOW, a contract saying I'll give you money for no reason at all is not legally binding. This occasionally comes up in charity law -- an organization can't sue you if you promise to donate money but don't.
In this situation, mp3.com won't get back any of the money they've paid (including the settlement over past infringements), but the contract ends as soon as this bill is signed.
Sort of like announcing that a new root-kit has been released, or that a new weakness was found in bind or sendmail.
OH! Wait, thise are security sites read by admins who want to PREVENT break-ins!
(yeah, I know it's a troll, but I'll bite anyway)
No, it is NOT like annoucing a root kit. Like bomb making instructions, root kits are often involved in crime, but it is perfectly legal to distribute either one. Illegally distributing ISOs is (obviously) illegal.
ISONews basically takes reports about crimes in progress and announces them to a gang neighborhood instead of law enforcement.
The line can be a fine one, I admit. For instance, you can legally post a root kit. However, if you posted a list of vulnerable IPs next to it and some or all of those IPs were later cracked by a root kit downloaded from your site, you are probably an accessory before the fact.
(as an aside, I agree that Sony, Sega, etc should seriously consider using these sites as a sting operation instead of busting them up)
While IANAL, the 5th only protects you from having to admit to a crime.
Encryption keys protect digital evidence just as doors and safes protect physical evidence. The 4th requires them to have a search warrant but, once they do, refusing to turn over the key is obstruction of justice.
After some poor experiences with servers in the K6-2/3 range, I'm a little gun shy of VIA and FIC.
With 8 128MB ECC DIMMs, the systems only recognized one trio as the full 384MB. All other combinations gave us either 128 or 256. Any pair would give us the full 256 and the Intel boards recognized any trio as 384.
The one using IDE also had a hard lockup. Not even the power switch worked -- had to pull the plug.
K6/233 on Intel worked like a charm, so I don't blame AMD. Now that AMD makes a chip set, I may give them another chance.
IANAL, but trademarks must be defended to allow a way for them to expire.
Patents and copyrights have a limited lifespan (too long in some cases, but that's a topic for another time). Once they expire, people no longer need to worry about accidentally violating one.
Trademarks don't expire as long as they're defended. If they lasted forever, we would have a lawsuit mine field as people drag ancient, obscure trademarks out of their family history and sue.
Slashdot Mirror
I'm not sure why, exactly, your post was moderated up (perhaps because it was lengthy?) -- you've not given any reasoning for saying the VM subsystem doesn't work, you just give an ad hominem (which, judging by the moderation, was effective in itself). Care to explain what, exactly, is wrong with the VM subsystem?
I'll field this one.
I write this as a kernel compiles in the background. 2.2.19. I was testing Web-1 with a number of 2.4 kernels, but it has freaked, oopsed, and panicked its way to 3 crashes in the last month, so I'm packing it away for awhile. 2.4 does okay on my desktop and the squid servers, but lions' dens we call web servers just eat it alive.
The oopsen all came out of the disk/io cache and most are posted to lkml, but it's like talking to a wall. :(
I do think it will work out in the end, though. We had fun with the 2.2.x Intel and 3Com NIC drivers up until around 2.2.12 or so. That's the curse of a high-traffic network -- you get to be the unwilling guinea pig for all of those 'close, but not quite' kernels.
As others here mentioned, the choice between SMP and multiple boxes depends a lot on your current bottlenecks and the amount of communication between tasks.
For instance, if your web server is memory speed or I/O bound, and updates are managed by a central authority (like an SQL server), go for a pair of servers. That doubles your system bus and disk subsystem capacity over a single SMP system.
On the other hand, if your server is CPU bound and/or requires fast communication between tasks (like updates to on-disk data), go for the SMP. Sure, the board is more expensive, but you only need one (same goes for drives, cases, network cards, etc).
Web servers generally fall into the I/O bound category, so two servers is probably your better bet unless you handle quite a bit of dynamic content.
They are still benchmarks, though. If you look at a broad range of benchmarks, the pattern is exactly what you would expect, considering the technology.
The P4 excels in predictable, tight-looped applications, especially en/decoding tasks. In broad-looped, unpredictable, or multi-tasking situations, that deep pipeline and RDRAM prove lethal as the processor burns off an insane number of CPU cycles getting its act back together after each misprediction.
I'm sure the P4 kicks butt in SPEC faceoffs, but SPEC benchmarks don't help me with games and work.
Sure we do. The next time a Klan rally comes to your town, just try to stop it. People try to every time, and I love the irony of them protesting (speaking freely) to ban the Klan's free speech. I may not like what they say, but there are plenty of idiots in the world; I'm not about to start the censorship snowball over it.
Linus actually mentioned this on LKML when the Crusoe was first announced. Linux on the bare metal was faster, but not dramatically. The translation caching absorbed most of the overhead.
The speed increase isn't worth the loss in flexibility. The primary advantage Transmeta has is the freedom to completely redesign the chip and instruction set to reduce heat and power consumption. Crusoe has enough speed for its target market, so running Linux on the bare metal wouldn't improve much.
There are real PCI modems around, but you have to look carefully. I picked up the USR 56K Performance Pro Modem: http://www.usr.com/products/home/home-product.asp? sku=3CP5610A
I'm sure you can finder cheaper ones if you try. It works just like an ISA modem.
- Pop-ups - they can all go die
- Masked links - I like to see where I'm going
- Scrolling text in the toolbar - Like above, they block the destination of links
- Cross-site scripting (onload=...)
- and so much more
I'll grant you, Javascript has it's uses, but designers usually just use it for useless annoying crap.Java is generally used for more productive purposes, but it likes to crash Netscape and it's a recurring security problem on IE.
No, a C&D is the first step at resolution. It only threatens to sue if the alleged abuse continues.
The purpose of a C&D is two-fold. Obviously, it's much cheaper than a lawsuit and usually effective. In addition, if you need to take the abuser to court at a later time, the company can submit the C&D as proof that has made efforts to protect its trademark. It also proves that the abuser knew about the violation and continued to abuse it anyway (for punitive damage).
If you find sorting through your porn collection 'tedious,' you need better porn.
It depends on your needs. Our squid servers (abused by ~1.5 million hits/day) have used test5 since August.
I admit, I haven't noticed fs corruption on them. I have noticed ps not accounting for 30-50MB of memory, but it's an acceptable loss for the ~50% performance boost.
"Yeah, but change is bad..."
Oh please. I've used AMD in my home system since the early K6's, but I still won't put them in a server. Why?
- No ECC functionality - Kind of useful in a server.
- No MicroATX boards with video and a good 100BaseT chip - Pretty much required for a 1U server.
- No multi-processor chipsets - Yeah, yeah, it's "just over the horizon." That doesn't help me right now.
This is an improvement over the K6/K6-2 chipsets, which could not be set to power on when plugged in. You power cycled the box and it stayed off. That really sucked.I have no problem with AMD CPUs, but nobody seems interested in targetting the server market with them.
You talk about that as if they didn't try. To the contrary, many communities passed laws and regulations to outright ban or ridiculously restrict automobiles. The most common example was pushed by the Farmer's Anti-Automobile Society (FAAS) in Pennsylvania where:
Fair enough, but don't buy into the myth that the majority of dynamic pages on the web involve critical information.
The majority of dynamic web pages are a dizzying array of counters, polls, news, comments, and other trinkets. Given how rarely a box goes down unexpectedly, losing a vote or comment is trivial.
Even in an e-commerce situation, not all data needs transactions. Take product info, for example. It's almost all read-only and staff-updated.
Mysql works wonderfully for dodging concurrency problems in a load-balanced cluster and offers faster, easier ways to sort data than a file system. BDB or no, I wouldn't trust it with financial transactions, but it's great for the splash of interaction most sites need.
Naturally she'll use a tether to turn that excess speed into energy.
It's really not that hard. Simply put, if an offer must clarify that "this is not a scam," that's exactly what it is.
> Why is it that a *nix box getting compromised = 'Excellent, now we can patch the hole'
I think that's the precise difference, right there. Judiging by preliminary reports, it looks like one-click execution of attachments aided (to put it lightly) this breach. That same design flaw spawned Melissa and the Love virus.
At this point, I think most would agree that the potential damage from that "feature" is far greater than the potential convenience and time savings. Yet here we are, many months and two huge outbreaks later, and that same feature is still enabled by default.
> but an NT machine = their security "sucks"?
It sucks, but not simply because it's NT, or closed source, or because it has security bugs and design flaws. It sucks because they don't fix it and you can't.
Unix used to be like that, actually. The Morris Worm was their sign to clean up their act and patch some holes. Microsoft ignored the closest Windows equivalent, Melissa. They ignored the Love virus, too. Maybe this will hit home enough to actually spur some action.
blah blah NAL blah...
Anyway, legally binding contracts must be an exchange between the parties. IOW, a contract saying I'll give you money for no reason at all is not legally binding. This occasionally comes up in charity law -- an organization can't sue you if you promise to donate money but don't.
In this situation, mp3.com won't get back any of the money they've paid (including the settlement over past infringements), but the contract ends as soon as this bill is signed.
OH! Wait, thise are security sites read by admins who want to PREVENT break-ins!
(yeah, I know it's a troll, but I'll bite anyway) No, it is NOT like annoucing a root kit. Like bomb making instructions, root kits are often involved in crime, but it is perfectly legal to distribute either one. Illegally distributing ISOs is (obviously) illegal.
ISONews basically takes reports about crimes in progress and announces them to a gang neighborhood instead of law enforcement.
The line can be a fine one, I admit. For instance, you can legally post a root kit. However, if you posted a list of vulnerable IPs next to it and some or all of those IPs were later cracked by a root kit downloaded from your site, you are probably an accessory before the fact.
(as an aside, I agree that Sony, Sega, etc should seriously consider using these sites as a sting operation instead of busting them up)
This network has been mentioned on Slashdot four out of the last five days. We're still alive and well, so move along already.
Time to give some other sites their beating.
While IANAL, the 5th only protects you from having to admit to a crime.
Encryption keys protect digital evidence just as doors and safes protect physical evidence. The 4th requires them to have a search warrant but, once they do, refusing to turn over the key is obstruction of justice.
That works really well until you try to order something or do anything else which requires your real identity.
After some poor experiences with servers in the K6-2/3 range, I'm a little gun shy of VIA and FIC.
With 8 128MB ECC DIMMs, the systems only recognized one trio as the full 384MB. All other combinations gave us either 128 or 256. Any pair would give us the full 256 and the Intel boards recognized any trio as 384.
The one using IDE also had a hard lockup. Not even the power switch worked -- had to pull the plug.
K6/233 on Intel worked like a charm, so I don't blame AMD. Now that AMD makes a chip set, I may give them another chance.
IANAL, but trademarks must be defended to allow a way for them to expire.
Patents and copyrights have a limited lifespan (too long in some cases, but that's a topic for another time). Once they expire, people no longer need to worry about accidentally violating one.
Trademarks don't expire as long as they're defended. If they lasted forever, we would have a lawsuit mine field as people drag ancient, obscure trademarks out of their family history and sue.