A problem with this occurs when you have a multiprocessor system. If you have a line in an L1 cache that you want to write to (from the other processor, to make things interesting), you need to hunt down the line in not just an L2 or L3 cache, but also L1 and pull it into your own L1. This'll increase your latency because of the time it takes to find that line. You can see how this could be a pretty vicious thing if you have a heavily-contended lock in a particular cache line (and it's almost guaranteed to make a 1-by system perform better than a 2-by or greater system).
You can have directory mechanisms (that could add up to gigabytes of additional memory on larger systems, not used for storage) to quickly look up locations in a centralized place. You won't find this sort of thing on a desktop PC (usually a 1-by or 2-by, anyway), but on the larger machines (e.g. 8 processors or more, 16GB of RAM or more) it's definitely an issue.
I recently issued a purchase order for some translation services to get one of our products translated into Danish. The Purchasing Department made a mistake and accidentally purchased Denmark. What am I to do now?
I might buy the "within an hour" statement if the disclosure report fully described the problem and a fix.
Three days is short for a large organization. Somebody in charge needs to be convinced of the problem (managers do have many responsibilities and despite what we might hope to believe, fixing vulnerabilities cannot always be worked into a tight schedule). Then a competent developer needs to be allocated to do the work.
Most organizations will have to do some sort of quality assurance/testing on all changes to the software. It's irresponsible to not do this. That's another group, another manager, another several days. If you're close to an already-scheduled release date, the fix will probably be bundled in there. It costs A LOT of money to create new installation media, to mail the fixes to customers or to even go through (yet another) group to provide fixes on an external FTP server. In all, 45 days could be quite short!
Opinion [flamebait] section - it seems that in an open source environment, the testing sometimes isn't quite as disciplined (offset by more frequent releases). An acceptable fix may already exist in the disclosure report because somebody other than the programmer who applies the main source tree fix has been able to think about the problem. This also reduces the time to provide an official fix for this sort of project.
Being an IRCop on one network and friends with several IRCops on other networks, I've seen my share of political battles and obnoxious users. We're on a small network, so it isn't a big deal to link in another server every once in a while.
When turbulence happens, a branch of the network sometimes gets shaken out. We had a network of servers in the United States and a slew of machines in the Czech Republic. There were a few problems with timezones (the only time I could consistently talk to the opers over there was around 8am EST). Over the summer, a few US servers dropped (some IRCops left because they no longer had as much free time, families, etc) and the Czech network became its own autonomous network.
When IRC is fun, it's a lot of fun. Unfortunately, there are always a few snotty users who think it's their devine right to pester the IRCops for weeks on end or packet a server. At some point, the IRCops have had a bad day and things like *!*@*.home.com get banned. If there were some way we could uniquely lock out a user by retinal scan or a Bad Breath breathalizer test when they connect (anyone up for re-writing identd's to do this?) we'd love to have it. However, we're stuck with broad bans in order to keep ourselves sane. It's not nice to the users, but there isn't a better solution yet.
Particularly in the cases where an IRCer is good at social engineering, we have problems. Some of those users have managed (through various subtle methods) to get O: lines on our servers and our network then goes to pieces until we figure out what happened. To be honest, I don't understand what causes people to feel the need to do that every three months, but it happens.
I used to spend a lot of time on irc.cs.cmu.edu (EFNet) and irc.cis.pitt.edu before that (they allowed bots!), before they were packeted so many times that our upstream cut them off. To me, that was when EFNet suddenly lost its appeal, because it became a chore to find a server where I could keep a stable client connection. I believe that EFNet will continue to exist as increasingly smaller numbers of large servers, as IRCops get tired of the problems and the fun (or power trips) become less rewarding.
Some instructions will cause fewer changes within a processor than others. When you change the value coming out of a transistor, you have a change in the current consumption for that transistor. By reducing the type of operations performed, you can easily change the heat dissipation of your CPU.
For instance, if you hit heavily over a large area of cache memory (in particular, caches get hot) and change data, you will see greater heat dissipation than if you run a tight loop that (for instance) assigns a variable to its current value. If you don't believe me, try sticking your thumb on an idle 486 versus one running seti at home. Better yet, read the LM75/78/whatever temperature sensor on your brand spanking new machine. You bet it changes with load.
Just a note - you can use a low-pass filter to remove those high frequencies that you talk about. Even so, the sampling frequency would have to be at least twice the encoded high-frequency signal (Nyquest rate) in order to have a chance of being sampled accurately. Ideally, you would bump that up to 3 - 5x (give or take) the highest frequency that you want to reproduce.
I do not doubt the existance of watermarks that can survive D->A->D for several generations, however.
As far as user interfaces go, I wouldn't want to write them in SML, either!:)
I wouldn't call grep or wget simple (see the documentation if that's at all unclear!). I don't believe that those two in particular would be terribly bad, though. Performance would probably be really crummy, but I could imagine that some of the more mundane uses of wget (recusively mirroring a website) would come naturally to a functional language.
On the other hand, I think the user interface or presentation of a program should be well-separated from it's data structures or, more towards the dollars sign, its business logic. Why can't the application tier of an e-commerice site be SML-based? With the appropriate connectors (CORBA-based or otherwise) to databases and either a slew of HTML-generating code or some other connector to JSPs, I could see it happening. Would it catch on? I don't know. Java has positioned itself to work well like this. An event-driven program would probably crawl in SML, there are graphical toolkits for it, but I have never used them. If SML is used for the innards, however, it could be much more manageable.
The flipside is that if you're willing to put more time into abstracting and kneeding (as I like to call it) the functionality of your project, the time actually spent coding and debugging can be lowered significantly (in my experience, the total time from start to finish is about the same, but I am generally more confident and sure that the project is finished if I spent time thinking about it). This applies to any language, not just functional languages (where code without forethought probably won't even compile in SML)
Also, I found that when I was writing SML (only in a college course), I really didn't need a debugger. When I was really stuck, sometimes a little string printing function was useful, but beyond that it was much easier (and more revealing to stand back for a few minutes and look at what was really going on). And of course, SML has the wonderful property that if it compiles without warnings, it will practically never crash (though it may never stop recursing, either:) ).
In my mind, programs written in functional languages have a natural modularity to them. You don't find a lot of speghetti code or unneeded work. I still find myself using C++ and Java in the real world, because a 50MB runtime library is rather heafty. I'd also have trouble finding coworkers who could tolarate the language.:)
The issue with allowing the database to warm up is that you're attempting to get the database into a steady-state condition, e.g. your disk cache is not stale or any database cache that would be beneficial to the performance of the database is not stale.
A database that has been open and running for 2 minutes is not the same as a database that has been up for 30 days (the latter has been warmed up). It is how the server will actually be operating for a majority of its use.
The warm up is not, however, something that you would do when you start up the server processes. It's simply to get the system in a state like it would be in when you have it running for a long time. We're talking about two different things.:)
Try the TPC-C reports here. These are not specifically comparisons between vendors, but they are all reports of systems running a common benchmark. The reported benchmarks all went through approved TPC audits before being released. (On an aside, Microsoft just withdrew their results on SQL Server 2000 because they failed the audit, it's on ZDNet somewhere, I don't have the link).
The article was listed as a column, not a full-disclosure benchmark report. I'll let the author slide on those points, but here are some things to think about when running a benchmark:
He gave a very vague description of the layout of his database and the nature of the queries. Without that, it is impossible to reproduce his results.
He did not describe his client/driver configuration, for all we know, they could have been programs running on the same machine as the DB server
For database benchmarks such as TPC-C (you can see the stringent reporting specifications at www.tpc.org) it is common to allow the driver program to warm up the database for a long amount of time (45 - 60 minutes seems to be common now, depending on the size of the database) then allow the database to reach steady state after at least 20 minutes. He does not mention whether he gave time for the database to stablize. Transient numbers are not a good indication of actual system performance.
His reported statistics are lacking. Saying that PostgreSQL is 2 - 3 times faster than MySQL leaves a 50% uncertainty in his measurements. At times, he reports 2 - 3, and others he reports simply 3. In fact, I don't see any solid measured numbers for queries/second, transactions a second or some other metric.
Granted, most websites running either of these software packages are probably not going to see very heavy loads, but if you're deciding between two vendors, it's good to know exactly *what* you are comparing.
There's nothing fundamentally wrong with using both the current system and alternate. For instance, if you're looking up a.COM,.NET,.ORG or.EDU you'd have to end up looking at the old root servers and the authoritive DNS servers for that domain, anyway.
Alternic tried this back around 1996. Here's a link to a boardwatch article that discusses their system. It transparently handled regular Internic (now Network Solutions) requests as well as their own names/TLDs. They mentioned that you could get your own TLD for $98/year. How cool is that?:) I never actually tried changing my DNS servers over to theirs, just because it didn't seem to be catching on at all.
There are also large high-speed laser printers for this sort of thing. A company that I worked at a few years ago had several 300ppm laser printers for printing medical bills. It was truly impressive to see them spit out a pile of paper (quite noisy, however).
Similarly, DECstation 5000's have three fans in the case, but they run extremely quietly. They are positioned to blow air across the entire motherboard. Granted, these are diskless boxes and the 10-year-old full-height SCSI drives make quite a racket, but the machine itself is pretty quiet.
Storing things in the other room, particularly with a keyboard/mouse/video extender is an expensive (for decent components) solution, but it works well.
Yep, I remember it! Unfortunately disk 2 had a few bad sectors, so I couldn't get too far in the actual game. However the little robot design modules worked like a charm. I spent hours just playing with little robots. I guess both disks have gone bad/are lost by now... too bad:(
As far as the bandwidth issue goes, you can set up relays servers on higher-bandwidth pipes. So, a main server only has to transmit a few streams (perhaps only to specified relays) and if your relays can be the real bandwidth hogs.
This also means that there is more than one URL for your stream, unless you set up a DNS server to let clients have an equal chance of getting each available server. There are, of course, many other ways to handle this as well.
If you download the shoutcast server, make sure you get the 1.6.something beta release. The "stable" 1.5.0 release has a rather annoying bug where relays will silently stop serving data (even though the console looks perfectly normal) when the uplink connection is lost -- and it makes no attempt to reconnect.
This is fixed in the latest beta, which seems to be stable. The problem appears in both Linux and Win32 servers in my experience, I haven't verified it for other platforms.
Yes, you can do this. The mode called 'ad-hoc' mode is available on Lucent cards (it's not an 802.11-compliant mode) and allows many machines to talk to each other directly, without the need for an access point/base station. I have personally tested it with up to 6 cards and it works like a charm. The setting is available simply as a little option in the driver (or checkbox in Windows). I don't think any other brands support 'ad-hoc networking.'
Slashdot Mirror
You can have directory mechanisms (that could add up to gigabytes of additional memory on larger systems, not used for storage) to quickly look up locations in a centralized place. You won't find this sort of thing on a desktop PC (usually a 1-by or 2-by, anyway), but on the larger machines (e.g. 8 processors or more, 16GB of RAM or more) it's definitely an issue.
I recently issued a purchase order for some translation services to get one of our products translated into Danish. The Purchasing Department made a mistake and accidentally purchased Denmark. What am I to do now?
-- puzzled@microsoft.com From Redmond, WA
Who's limiting you to the visible spectrum? :-)
Three days is short for a large organization. Somebody in charge needs to be convinced of the problem (managers do have many responsibilities and despite what we might hope to believe, fixing vulnerabilities cannot always be worked into a tight schedule). Then a competent developer needs to be allocated to do the work.
Most organizations will have to do some sort of quality assurance/testing on all changes to the software. It's irresponsible to not do this. That's another group, another manager, another several days. If you're close to an already-scheduled release date, the fix will probably be bundled in there. It costs A LOT of money to create new installation media, to mail the fixes to customers or to even go through (yet another) group to provide fixes on an external FTP server. In all, 45 days could be quite short!
Opinion [flamebait] section - it seems that in an open source environment, the testing sometimes isn't quite as disciplined (offset by more frequent releases). An acceptable fix may already exist in the disclosure report because somebody other than the programmer who applies the main source tree fix has been able to think about the problem. This also reduces the time to provide an official fix for this sort of project.
[Scene: MIR]
Ooops! Boris just opened up the airlock agai --- SCHOOOOOOMMMMPPPPPPPPP.
[Begin unsettling silence]
When turbulence happens, a branch of the network sometimes gets shaken out. We had a network of servers in the United States and a slew of machines in the Czech Republic. There were a few problems with timezones (the only time I could consistently talk to the opers over there was around 8am EST). Over the summer, a few US servers dropped (some IRCops left because they no longer had as much free time, families, etc) and the Czech network became its own autonomous network.
When IRC is fun, it's a lot of fun. Unfortunately, there are always a few snotty users who think it's their devine right to pester the IRCops for weeks on end or packet a server. At some point, the IRCops have had a bad day and things like *!*@*.home.com get banned. If there were some way we could uniquely lock out a user by retinal scan or a Bad Breath breathalizer test when they connect (anyone up for re-writing identd's to do this?) we'd love to have it. However, we're stuck with broad bans in order to keep ourselves sane. It's not nice to the users, but there isn't a better solution yet.
Particularly in the cases where an IRCer is good at social engineering, we have problems. Some of those users have managed (through various subtle methods) to get O: lines on our servers and our network then goes to pieces until we figure out what happened. To be honest, I don't understand what causes people to feel the need to do that every three months, but it happens.
I used to spend a lot of time on irc.cs.cmu.edu (EFNet) and irc.cis.pitt.edu before that (they allowed bots!), before they were packeted so many times that our upstream cut them off. To me, that was when EFNet suddenly lost its appeal, because it became a chore to find a server where I could keep a stable client connection. I believe that EFNet will continue to exist as increasingly smaller numbers of large servers, as IRCops get tired of the problems and the fun (or power trips) become less rewarding.
For instance, if you hit heavily over a large area of cache memory (in particular, caches get hot) and change data, you will see greater heat dissipation than if you run a tight loop that (for instance) assigns a variable to its current value. If you don't believe me, try sticking your thumb on an idle 486 versus one running seti at home. Better yet, read the LM75/78/whatever temperature sensor on your brand spanking new machine. You bet it changes with load.
Some people really connect with 80-pins and integrated power. Don't dis SCA chicks!
I do not doubt the existance of watermarks that can survive D->A->D for several generations, however.
Quoted from bash-2.04/doc/bashref.info
This is Edition 2.4, last updated 14 March 2000, of `The GNU Bash Reference Manual', for `Bash', Version 2.04.
Copyright (C) 1991-1999 Free Software Foundation, Inc.
So, before you start shelling out bad arguments, please remember that it's Posix 1003.2-compliant code that you're bashing.
Parse error on line 37:
Mis-matched parentheses. Bailing out! Error near:
tend to watch less TV news all the time (The rise of Net news and related
For some reason, the mental image of UnixWare crawling around in an aquarium seems quite fitting.
I wouldn't call grep or wget simple (see the documentation if that's at all unclear!). I don't believe that those two in particular would be terribly bad, though. Performance would probably be really crummy, but I could imagine that some of the more mundane uses of wget (recusively mirroring a website) would come naturally to a functional language.
On the other hand, I think the user interface or presentation of a program should be well-separated from it's data structures or, more towards the dollars sign, its business logic. Why can't the application tier of an e-commerice site be SML-based? With the appropriate connectors (CORBA-based or otherwise) to databases and either a slew of HTML-generating code or some other connector to JSPs, I could see it happening. Would it catch on? I don't know. Java has positioned itself to work well like this. An event-driven program would probably crawl in SML, there are graphical toolkits for it, but I have never used them. If SML is used for the innards, however, it could be much more manageable.
Also, I found that when I was writing SML (only in a college course), I really didn't need a debugger. When I was really stuck, sometimes a little string printing function was useful, but beyond that it was much easier (and more revealing to stand back for a few minutes and look at what was really going on). And of course, SML has the wonderful property that if it compiles without warnings, it will practically never crash (though it may never stop recursing, either :) ).
In my mind, programs written in functional languages have a natural modularity to them. You don't find a lot of speghetti code or unneeded work. I still find myself using C++ and Java in the real world, because a 50MB runtime library is rather heafty. I'd also have trouble finding coworkers who could tolarate the language. :)
A database that has been open and running for 2 minutes is not the same as a database that has been up for 30 days (the latter has been warmed up). It is how the server will actually be operating for a majority of its use.
The warm up is not, however, something that you would do when you start up the server processes. It's simply to get the system in a state like it would be in when you have it running for a long time. We're talking about two different things. :)
Try the TPC-C reports here. These are not specifically comparisons between vendors, but they are all reports of systems running a common benchmark. The reported benchmarks all went through approved TPC audits before being released. (On an aside, Microsoft just withdrew their results on SQL Server 2000 because they failed the audit, it's on ZDNet somewhere, I don't have the link).
- He gave a very vague description of the layout of his database and the nature of the queries. Without that, it is impossible to reproduce his results.
- He did not describe his client/driver configuration, for all we know, they could have been programs running on the same machine as the DB server
- For database benchmarks such as TPC-C (you can see the stringent reporting specifications at www.tpc.org) it is common to allow the driver program to warm up the database for a long amount of time (45 - 60 minutes seems to be common now, depending on the size of the database) then allow the database to reach steady state after at least 20 minutes. He does not mention whether he gave time for the database to stablize. Transient numbers are not a good indication of actual system performance.
- His reported statistics are lacking. Saying that PostgreSQL is 2 - 3 times faster than MySQL leaves a 50% uncertainty in his measurements. At times, he reports 2 - 3, and others he reports simply 3. In fact, I don't see any solid measured numbers for queries/second, transactions a second or some other metric.
Granted, most websites running either of these software packages are probably not going to see very heavy loads, but if you're deciding between two vendors, it's good to know exactly *what* you are comparing.Alternic tried this back around 1996. Here's a link to a boardwatch article that discusses their system. It transparently handled regular Internic (now Network Solutions) requests as well as their own names/TLDs. They mentioned that you could get your own TLD for $98/year. How cool is that? :) I never actually tried changing my DNS servers over to theirs, just because it didn't seem to be catching on at all.
You're fooling yourself. You'd just get called twice as much by the same people.
There are also large high-speed laser printers for this sort of thing. A company that I worked at a few years ago had several 300ppm laser printers for printing medical bills. It was truly impressive to see them spit out a pile of paper (quite noisy, however).
Storing things in the other room, particularly with a keyboard/mouse/video extender is an expensive (for decent components) solution, but it works well.
Yep, I remember it! Unfortunately disk 2 had a few bad sectors, so I couldn't get too far in the actual game. However the little robot design modules worked like a charm. I spent hours just playing with little robots. I guess both disks have gone bad/are lost by now... too bad :(
This also means that there is more than one URL for your stream, unless you set up a DNS server to let clients have an equal chance of getting each available server. There are, of course, many other ways to handle this as well.
This is fixed in the latest beta, which seems to be stable. The problem appears in both Linux and Win32 servers in my experience, I haven't verified it for other platforms.
Yes, you can do this. The mode called 'ad-hoc' mode is available on Lucent cards (it's not an 802.11-compliant mode) and allows many machines to talk to each other directly, without the need for an access point/base station. I have personally tested it with up to 6 cards and it works like a charm. The setting is available simply as a little option in the driver (or checkbox in Windows). I don't think any other brands support 'ad-hoc networking.'