Slashdot Mirror
Solaris Systems Programming
Several such guides have popped up over the years, such as The UNIX Programming Environment (Kernighan & Pike 84), Advanced UNIX Programming (Rochkind 85), The Magic Garden Explained (Goodheart & Cox 94), Advanced UNIX Programming (Gay 00) (that's not a typo, there really are two books with the same name), UNIX Systems Programming (Robbins & Robbins 03), UNIX Systems Programming for SVR4 (Curry 96), and the undisputed heavyweight champ, Advanced Programming in the UNIX Environment (Stevens 92).
Each of these books is distinctive, yet they share a number of topics. Essential topics include low- and high-level IO, signal handling, processes, IPC, and basic file system mechanics. In the more modern books, we see the inclusion of popular topics such as threading. Discussion directed toward broader topics of UNIX vary widely, namely due to the OS agnostic nature of such guides, despite the fact that until recently many books tended to slant toward SunOS/Solaris. Regardless of how many systems programming texts have appear, however, most programmers will agree that Stevens' guide is the only truly definitive choice. Since its release, there has been little challenge to its prominence, despite the emergence of Linux as a major UNIX implementation, despite several newer systems programming books, and even the 2nd edition of Rochkind's guide. But all of this now changes thanks to the release of Rich Teer's Solaris Systems Programming.
At a whopping 1248 pages, this volume dwarfs just about every systems programming book available by over 500 pages. It avoids the distractions of OS ambivalence by being tailored to Solaris, but is applicable to any UNIX platform available including Linux. Its layout is similar to that of Stevens' or Curry's books but builds significantly on each topic.
New systems programmers will immediately appreciate Teer's completeness, both in topic coverage and in his example code. Almost every code example is complete and runnable, unlike many of the other guides that demonstrate a topic only in an abstract function rather than complete program. Essential topics for completeness which have remained surprisingly absent from nearly every guide available (such as memory, code security and 64-bit topics) are thoroughly covered. A striking example is coverage of memory topics. When I pulled volume after volume off the shelf of my local bookstore and looked up "memory" in the index of each, I found surprisingly few even cover the topic beyond explaining the difference between stack and heap. In fact, many don't even include the malloc() function. Solaris Systems Programming is the only book I've ever found so complete in its memory discussion that it not only covered stack and heap, all the available memory management functions, but even discusses such important topics as memory alignment!
A complete chapter on secure C programming is provided, thoroughly discussing such important topics as buffer overflows, chroot jails, and program environment. A good number of tips are provided to help you immediately incorporate better security into your app whether it's a real concern (for now) or not. Combine this with a complete chapter on resource control and limits, including discussions on system information, the /proc file system, and some Solaris-specific resource control facilities, you can write more intelligent, less obtrusive, and better-tuned applications.
The coverage of advanced IO topics (including STREAMS) and file system coverage are superior to that in any other text I've seen. System admins will appreciate the in-depth coverage of file system topics that have only seen this sort of detail in books such as Solaris Internals (Mauro & McDougall, 00). This level of discussion allows not only a better understanding of file system and IO techniques, but also the clarity to immediately start building your own tools that allow you to interact with file system with far greater precision than ever before. Other topics, such as memory mapped IO, have rarely seen such detailed coverage.
A full treatment of IPC topics are handled, but like Stevens', these techniques are discussed using conventional concurrency techniques such as fork(). A discussion of POSIX threading is absent and regarded as too large a topic to address properly in a systems programming book and the reader is urged to consult a complete guide to the topic such as Programming with POSIX Threads (Butenhof, 97). While some readers might be put off by this, you'll appreciate how this keeps IPC discussions unencumbered. POSIX threading is mentioned where applicable, so it's not at all ignored, but readers of Rochkind's 2nd Edition or Robbins' books will notice that introduction of a PThreads overview can quickly overwhelm the rest of the text. Unique to any other text with which I am familiar is the inclusion of a section on Solaris Doors (also applicable to the Linux implementation), which is the fastest IPC method in Solaris, introduced with Solaris 2.6.
Something that both new and seasoned programmers will appreciate is the inclusion of a chapter on utility functions, and another on localization. The utility chapter provides great a discussion of (and reference to) the often-used functions that many other books ignore, such as string handling and manipulation functions, memory management, byte arrays, temporary files, error reporting, command-line argument parsing, character classes and more. While it's true that these aren't strictly systems programming topics, they are inevitably going to be topics of interest to most programmers. It is the inclusion of such topics that allows you to take your pile of reference books and replace it with this single volume.
A major topic to systems programmers today is 64-bit programming. Naturally, Solaris is a robust 64-bit environment, and is well handled in this book. programmers new to 64-bit environments, whether on Linux, Solaris, or other UNIX platforms, will greatly appreciate the gentle introduction to 64-bit coding, as well as best-practice techniques and sprinkled 64-bit wisdom throughout the text.
Like it or not, Solaris is the dominant commercial UNIX platform in the market today and will be for the foreseeable future. This guide doesn't pull any punches in giving you the best information available to exploit that environment to its full potential. If you're a programmer, this book gives you a single reference to consolidate your library and give you a new appreciation for familiar topics and entry point to things that you might have never leveraged before (Doors, 64-bit optimization, etc). If you're a system admin, you'll find a whole new appreciation for Solaris and UNIX in general with unparalleled understanding of how they really work under the covers, especially if you've already read Solaris Internals. Everyone will love the detail and completeness, combined with with the hundreds of tips (not to mention nifty Solaris trivia) scattered throughout the book. Rich's style is compelling and relaxed, very readable in front of your keyboard or with a cup of coffee on the porch. And readers will enjoy his sense of humor, which is admittedly subtle; experienced programmers and system admins, though, will enjoy the book's wit.
Finally, given the impending release of Solaris 10, yet another aspect of this book needs to be considered: it's an essential companion for DTrace users! Rich couldn't have possibly foreseen this need when he started writing the book, but it is extremely important today. Solaris 10 provides more visibility and debugging tools than any other UNIX system in existence today, the most popular of which is DTrace. But all of these tools expect the user to have a certain level of understanding of the system itself. This book should be standard issue for any sysadmin that ever plays with Solaris 10. When doing system root-cause analysis with DTrace, this book becomes an essential reference, especially if you are allergic to system headers. If you have been using DTrace and getting lost, or feel that you just don't know Solaris the way you need to, buy this book and you'll find the confidence and skills to get you back on track.
You can learn more about Solaris Systems Programming on Rich Teer's home page for the book. On that page the full contents and index are available, including a sample chapter (Ch 8 "System Information and Resource Limits," 62 pages!). You can also visit Teer's personal home page to learn more about him and his work. You can purchase Solaris Systems Programming from bn.com. Slashdot welcomes readers' book reviews -- to see your own review here, read the book review guidelines, then visit the submission page.
Am I the only one that can't read the word holistic without thinking about electric monks on horses?
Certainly, there are plenty of UNIX users and admins who understand semaphores but have never written a threaded application, and C programmers who have never left the Windows world,
Newsflash: Windows implements IPC mechanisms. You know, like all modern multitasking OSes?
Other newsflash: C programmers can be found under Windows, Unix, MacOS10, BeOS, GEM, ThingamabobOS... In fact, real professional programmers can program in anything under any environment, they just happen to be a bit more proficient under certain kinds of environments.
"A door is what a dog is perpetually on the wrong side of" - Ogden Nash
Amzon has a great deal on a new book released by O'Reilly. Programming the 8080 in assembler.
Comment removed based on user account deletion
Big Gay Al
Do not try to read the dupe, thats impossible. Instead, only try to realize the truth
What truth?
There is no dupe
...is like the relationship between Petrol and a Car.
You can put a monkey in a car and they might dent it, pee on it, scratch it... but it'll carry on (mostly) working. Let the monkey lose with Petrol and the whole damn thing is going to blow and some poor sod will lose their eyebrows.
Beep beep.
Who is this guy really, Robert Jordan?
Dogma - "let's just say we'd like to avoid any empirical entanglements."
The title is systems programming, not device drivers.
7 November 2006: The day Americans realized corruption and incompetence weren't addressing 11 September 2001
It's one of the most generalized C/UNIX programming books I've been able to find; it doesn't pidgeon-hole itself into a particular *nix. After all, C in one Unix should ideally be portable to another Unix.
Q: What do you think about American Culture?
A: I think it's a good idea.
(adapted from Gandhi)
and there is a philosophical connection.
basically C/C++ try to do things the machines way. if the machine can do it, C/C++ can do it. Unix benefits from that thinking... except when it comes to ejecting the fucking CD!
just a little joke with my humor.
-pyrrho
I was in a discussion recently and someone asserted that the source for solaris 8 was open, and freely available. Is there any truth to this, or were they thinking of the rumour that solaris 10 is supposedly going to be OSS?
/cynicism on
/cynicism off
Do they have SCOs approval to print this?
Ok this joke has been beat to death.
Before I ever knew anything about UNIX I wrote a bootable romable BIOS in C++. I would say that C definitely supercedes UNIX. But could you make an OS without C? Must be because people used to do it. Unix needs C. C would exist without Unix.
By the author's description, I'd say that the topics braches in this book cover two university level courses, Computer Architecture which entails registers, alignment, buses (Structured Computer Organization Tanenbaum) as well as Operating systems internals (APUE, Stevens or Modern Operating Systems, Tanenbaum)
The problem being that both instructors need to agree on the book to get the benefit over the diverging information.
In an academic standpoint, the book's too large to serve as a workable text, and too specific to be used for multi-course uses. Of course I've never actually seen the book, so its all speculation based on the review.
For personal uses, I'm sure the added insights would be nice for those who haven't been beaten over the head with alginment and register offsets from schooling.. (*arg*).
Bye!
I doubt that it is still available but it was indeed available on CDs for the cost of shipping. I ordered it for x86 in 2001 and was going to run it on a pII 350...never got that bored ;)
that it's not even considered as something worth mentioning in programming job ads? I mean it's automatically assumed that you know it. Or maybe it's not needed. When was the last time you were asked by a recruiter whether you knew unix IPC?
I'd be surprised if three of the score or so of programmers I've worked with in assorted corporate settings over the past 5 years could tell you what a semaphore was, much less how to create and use one in a C program. In fact, in 15 years of programming in the "real world" I could count on one hand the number of times I've ever seen another programmer use "fork" in their code, even when forking off another process and establishing communication between the two processes would have made the job much easier. And don't get me started on the IPC blunders I've seen over the years from "professional" programmers. And as bad as those are, they pale in comparason to the code I've seen some Java programmers squat and shit. Er... you get the idea...
I'm trying to teach myself to set people on fire with my mind... Is it hot in here?
I prefer this URL:
/ richteer-20/
:-)
http://www.amazon.com/exec/obidos/ASIN/0201750392
At least that way, I get some credit from Amazon.com for the referral.
Faster inter-thread communication, and only needing to mmap files once.
``fork() is a cheap operation on unix so what is advantage of threading?''
Guess what? Typical UNIX software doesn't use threading. Forking is much easier, was there first, and is usually not significantly less efficient.
``why did apache 2.0 use threads?''
Probably because it runs on systems that don't have cheap forking (like Windows). Besides, IIRC threading is only one of the mechanisms that apache2 can use, and you can use forking if that suits you better.
Please correct me if I got my facts wrong.
Not quite. COW: copy on write. Basicly you use the same physical page in both processes, until one or the other makes a write to that page. Then and only then do you make a copy.
I still have more fans than freaks. WTF is wrong with you people?
You can purchase Solaris Systems Programming at the Tattered Cover. This is not an affiliate link. I post this beacuse the Tattered Cover works to protect First Amendment Rights. Thank you.
Asserting the First Amendment rights of its customers, the Tattered Cover Bookstore challenged a search warrant obtained by police that sought information about all books purchased by a customer in a 30-day period. The ACLU of Colorado filed an amicus brief arguing that the state constitutional right of free expression requires special procedural protections when the government seeks information about who is reading which particular books. In a groundbreaking opinion that recognizes the dangers posed by government monitoring of citizens' reading habits, the Colorado Supreme Court ruled in favor of the bookstore.
Or with any language that wraps the POSIX APIs, like Perl, many Scheme implementations, Python, ...
Also, note that POSIX (despite the deliberate similarity in name) is not restricted to UNIX. IBM makes a few very ununixlike OSes that are yet POSIX compliant.
Please correct me if I got my facts wrong.
Easy access to shared variables. If you fork, each process gets their own copy of a variable. An update by process A does not propogate to process B. This is a two sides issue, as it also makes race conditions possible if you don't protect data structures correctly. Of course, you can use shared memory, but thats more of a hassle to set up.
The other advantage is task switching speed. When switchign between applications (such as forked processes), you need to do a full task switch- registers, stack, memory, cache invalidate, etc. Very expensive. When switching between threads you only need to swap out the registers, stack, and program counter. Very cheap.
I still have more fans than freaks. WTF is wrong with you people?
Security should always be important whether you're providing a network server, a setuid application, or neither of these things.
In many cases security issues arise from having malicious input cause an exploit, even in non-security-sensitive applications if you're not careful unexpected input can cause a crash which might be just as painful from a user point of view.
Too many people forget that security is a process, and not an addon.
Many good tips on secure programming can be found in David Wheeler's Secure Programming For Linux and Unix HOWto.
Read it, even if you dont think security is important for you yet. It's only a matter of time until it will be.
``Anyways who write modules for solaris ?''
Quite a few people, actually. And with the plans to release the source, the number will likely grow.
Please correct me if I got my facts wrong.
ive read it like: "Build your own solar system". wouldnt that be great?
/troll
IMHO, Schwartz and McNealy are already "communicating" from outside our solar system, so why the hell not?
Soko
"Depression is merely anger without enthusiasm." - Anonymous
> I mean, who has "Gay" in the name?
There is a law firm in town named "Gay and Associates".
I almost pissed myself laughing one day when I had to drop some papers off there.
Do daemons dream of electric sleep()?
In an ideal (simple) software implementation fork() is cheap. However, try running a few thousand simultaneous, long-running connections to an Apache 1.x server (and mod_xxx extensions) and see it grind on a moderate machine - if it even gets that far (say, 15MB per process with a memory-tuned mod_perl). A fully-functional, threaded java-based server will get there with ease, and probably Apache 2.x too. See the real-time, multipart mime Chart on our website for an example.
Galmarley - Free research on economic hi
Got my copy at SD Tech books, so you won't get the Amazon credit.:-(
You and your colleagues are way out of touch. Solaris has been smashig Linux performance on several new benchmarks. See Sun's web site...
Just a quick comment. With linux, the clone() syscall lets you create a new child with the ability to specify at a very fine-grained level what you want to share. File descriptors, memory map, signal handlers, etc. can all be specified separately.
Then, in the scheduler it is smart enough to consider how much is actually being shared between the old task and the new task, and switch only the bits that actually need it. Plus, since processes in linux are very light weight, there isn't actually that much difference switching between threads in the same process and threads in different processes.
If you read "The Art of UNIX Programming", threads are considered the last resort if no other IPC mechanism is fast enough. This is an extremely rare occurrence. Separate processes with specific areas of shared memory is almost always sufficient, and far less likely to cause problems.
I think it is a culture thing. For a long time, Unix thread code didn't port.
Some flavors didn't even have threads.
So if you wanted to write a product that worked on multiple Unix variants, you used forking and IPC instead of threading.
It took a long time for pthreads to catch on.
Windows programmers generally don't worry about porting so they took to threads more quickly.
"We can't solve problems by using the same kind of thinking we used when we created them." -- Albert Einstein
I'm not sure what you mean by "memory" in the full task switch list. I suspect you mean "address space"; swapping that could be expensive if, for example, you have to invalidate a TLB and re-populate it.
If by "cache invalidate" you mean invalidate the caches that sit between the CPU and main memory, that might not be necessary if, for example, the cache is physically indexed and tagged - and possibly not even if it's virtually indexed if, for example, it's physically tagged.
So, while there is a cost to process switching if it involves address-space switching, it's not necessarily as large as you might think.
Yes, by memory I mean address space. Which generally means invalidating TLBs and switching page tables (and on some systems like x86, possibly paging in the page tables).
Yes, by cache invalidate I meant the cache between the CPU and memory. I'm at work, so I can't really check sources, but IIRC there's a lot of cases where cache and its associated buffers need to be invalidated. Possibly not all, I won't pretend to be on the cutting edge of that stuff.
At any rate, the basics is- process switching between 2 threads is very cheap, switching between two processes is more expensive. The exact ratio between the two depending on a lot of factors. It is still an advantage for threaded apps, wether you need that advantage or not depends on the application.
I still have more fans than freaks. WTF is wrong with you people?
Perhaps. But it's also far less error prone than threads sharing variables when one thread forgets to get its locks right. That in turn leads to more reliable and stable software, and is generally considered a better model. With mmap/MAP_SHARED, you have to explicitly reference the data you wish to share, so the programmer tends to think more about what they're doing. With threads, everything is shared by default, and it's much easier to overlook the fact that another thread might be using that data.
"The invisible and the non-existent look very much alike." -- Delos B. McKown
So, coming from the perspective of someone who knows how to move around in UNIX but not much else, would this book be good for me? I have no problem with any of the topics mentioned, I am just worried that my lack of experience combined with my lack of Solaris will make this book too hard to understand. BTW, you can get it a bit cheaper here.
well for one threads have shared memory so communication is simply reading/writing a globally protected variable.
not to mention that most (not linux, until very recently) have a kernel level threads, which can map n-to-m to user-space threads. that means when some app creates some number of threads n, that they may only be backed by m real context executions within the kernel. this is a major advantage over forking. if you need a thread for just I/O where an execution context isn't required, it's much lighter weight than creating an entirely new process.
if you don't know why threads are better than creating processes, then you should read a good intro to threads. I'd suggest one that talks about solaris threads or the linx ntpl threads.
yeah but in fork you still have to create and copy the page table. In thread you dont. Also context switching is cheaper for threads than process.
The dicothomy netween processes and threads is false/religious/historical/etc, just use the right tool for the job, that's all.
What's in a sig?
Windows programmers generally don't worry about porting so they took to threads more quickly.
Whaddaya mean Windows programmers don't worry about porting? I've seen *lots* of Windows programs that boast about their portability.
They run on Windows 95, Windows 98, Windows 98SE, Windows ME, Window NT 3.5, Windows NT 3.51, Windows NT 4.0, Windows 2000, Windows 2000 SP1, Windows 2000 SP2, Windows XP...
Note to ACs: I usually delete AC replies without reading them. If you want to talk to me, log in.
yeah but in fork you still have to create and copy the page table.
Not true for clone() in Linux. clone() is really an elegant solution to the whole problem; it's just fork() (which is very easy to work with) with fine-grained control over how much the current and new processes share (you can't call them parent and child, because clone() allows a process to create siblings, not just children).
But, of course, if you need portability and don't want to suffer the (small but non-trivial) overhead of fork(), then you need to use pthreads. They're also not that tough to work with, but are less elegant and less flexible than clone().
Note to ACs: I usually delete AC replies without reading them. If you want to talk to me, log in.
Despite this thread being conducted with typical /., unrelated comments, I actually thought this was a great book review. I know this book has been a big help for my husband with his work and I'm glad to see there is a review of it for the /. crowd, so maybe it can be of similar assistance for others.
Since I dont know about the clone(), I cant comment but all unix os have a vfork which does not create a new page table. The idea is that user will immediately do the exec after the vfork.
Is clone() the clone of vfork in linux ? (Pun intended)
OK Linux troll, if you're not going to be programming on Solaris/Sparc systems, why are you bothering to even post in this story?
"People who do stupid things with hazardous materials often die." -- Jim Davidson on alt.folklore.urban
IIRC Solaris has switched to an 1-1 threading model by default in its latest version. And also IIRC, linux has had a 1-1 model all this time because Linus thought it was good enough and way simpler than a n-m model.
Is clone() the clone of vfork in linux ? (Pun intended)
:-)
Actually vfork(), as defined by POSIX at least, doesn't create a child with a shared memory space, exactly. Well, maybe it does, but you're not supposed to use the memory. In a POSIX program you're not supposed to modify any memory in a vforked child (behavior is undefined if you do). The only purpose of vfork() is to avoid wasting time copying all of the page tables immediately before calling exec(), which will replace the whole current process anyway.
I wouldn't be surprised if there are some implementations of vfork that do act like Linux clone() with the CLONE_VM flag, but you can't portably depend on that behavior (of course you can't portably depend on clone() at all).
On Linux, vfork() is a special case of clone().
Note to ACs: I usually delete AC replies without reading them. If you want to talk to me, log in.
IIRC, linux has had a 1-1 model all this time because Linus thought it was good enough and way simpler than a n-m model.
And because of that, Ingo Molnar developed the O(1) scheduler which make the 1-1 model perfectly efficient. Linux also has the clone system call which lets you share whatever you want to share, including process memory.
With these improvements, PThreads are now inferior for anything except porting code. Oh, and PThreads provides its own mutexes, which are pretty much obsoleted by Linux's new futexes that have the advantage of not being specific to PThreads.
When all you have is a hammer, every problem starts to look like a thumb.
Advanced Unix Programming And to my surprise it was recently updated from the 1992 version!!!
Way to go!
Solaris 8 Kernel Module Programming
THIS THING CAN TURN ON A DIME, MACROSSZERO STYLE ALSO FUCK BETA, ~NYORON
In the future it would be nice if you had a a disclaimer that you're trying to make money.
It depends what you call "typical".
A web server generally doesn't need shared data. Apache 1.x only uses shared data for a small amount of statistics gathering/monitoring, which is a very small amount of work compared with the job of actually serving HTTP, so the minor overhead in managing the shared data doesn't hurt very muvh at all.
The same applies to other kinds of server where little or no synchronisation is needed (e.g. telnet server, ftp server, ssh server).
At the other end of the spectrum is a program which does need synchronisation, but each request/response is quick. An X server is a good example. In this case, a single-task tight event loop architecture is appropriate.
And in the middle, you have a lot of jobs which aren't so simple. A typical DBMS is a good example of this. You have multiple concurrent tasks, which may take an unbounded amount of time, and which may require a nontrivial amount of synchronisation. The jobs aren't effectively sequential, meaning that SMP scalability is possible, but they're not completely isolated either. This is the kind of task where multi-threading is almost always the right answer.
Multi-threading is also commonly found in interactive GUI applications, where there may be long background tasks in conjunction with low-latency requirements.
So while it's true to say that typical Unix software, in the sense of most software that comes packaged on the CD from your vendor, is not multi-threaded, that's because most of that software doesn't have difficult synchronisation or scalability problems to solve. On the other hand, a lot of the software that you might buy or obtain after that is indeed multi-threaded.
sub f{($f)=@_;print"$f(q{$f});";}f(q{sub f{($f)=@_;print"$f(q{$f});";}f});
If you read that book, you will find that the author is heavily biassed against threads. You get the impression that he's never had to solve a difficult synchronisation/scalability problem in his life.
Not that there's anything wrong with that. Threads are inappropriate for a lot of tasks, and if your programming career isn't very varied (from a concurrency point of view!) you may never come across a problem for which threads are the best answer. On the other hand, there are some problems for which they're the only practical answer. Being able to see which is which before you start coding is part of being a good developer.
Threads get a bad rap for two main reasons:
The last point is the most unfair. The concurrency problems in writing a typical DMBS or a typical GUI web browser would be just as tricky to solve with fork() than with threads (possibly even trickier, which is, no doubt, why fork() isn't used for these jobs).
Just saying "threads are bad, m'kay" cheats you and your users out of a good solution to many problems which turn up in practice.
sub f{($f)=@_;print"$f(q{$f});";}f(q{sub f{($f)=@_;print"$f(q{$f});";}f});
threading is considered to be more lightweight, using less memory, which is a good thing for servers with heavy loads. In reality not all threading implementations are really lightweight though(ex perl ithreads)
the nicest part about programming in threads is the ability to share data without having to do crazy ipc stuff. This is used alot in servers, one client connects, changes the state of the server slightly, and the other clients and admin want to know what happened.
Selling software wont make you money, selling a service will.
In the future it would be nice if you had a a disclaimer that you're trying to make money.
Who cares?
Heaven forbid someone take a moment to try to save us some money by providing a link to a cheaper store and then, **GASP** expect compensation.
Compensation which, by the way, does not at all effect the price of the book for us.
Keep your eyes out, Solaris is being open sourced. The pilot has been underway now for a few weeks. I'm sure that there will be a big announcement and press release when it happens, and you can bet that Jonathon will certainly mention it.
Before the license flames start, there is a commitment that the license will be OSI compliant. There would be little point to the exercise otherwise.
Tp.
So creating a thread that immediately exits is faster than creating a process that immediately exits. Woo-hoo.
A more interesting case is if they keep running together. In the thread, everything is shared: there must be a locks on the heap to synchronize access, so you pay a price on every malloc and free[*]. (It should not be necessary to point out that these are far more common than creating threads.) There are similar issues for per-process in-kernel structures such as the file table. Because more data is shared, there is more likely to be contention between threads from the same process on different CPUs.
Some of tridge's benchmarks show that on $vendor Unix, using threads can be twice as slow for a server-side tasks as using processes. On Linux, threads are only slightly slower.
Shared-everything is often a slower model, not faster.
[*] OK, it doesn't have to be a lock, but the heap does have to be thread-safe, which makes it slower or more complex or both.
Trying to introduce threads has been an enormous disruption to the development model they had around 2000, and relatively few new features have gone in since then.
See for example this Interview with Rasmus Lerdorf last year:
Maybe getting a Windows port was worth all that suffering, but for Apache Unix users threads have been a catastrophe.
I agree.
The guy above saying that threads are faster to create because of page table copying is completely missing both the good and bad points of threads.
If you read "The Art of UNIX Programming", threads are considered the last resort if no other IPC mechanism is fast enough.
That's only because the book's author never have to deal with scenarios where threads are the logical solution, ie graphical interfaces.
Don't take me wrong the book is great and after too many years of experience, I still subcribe almost 90% of the views expressed on the book, but the 'unix only' background/bias of the author is evident.
As for threads or processes, just use the right tool for the job, there are no magic rules.
What's in a sig?
Yes, I know it was a joke. But...
All of those versions listed have threading.
There: Something at a specific location.
Their: Owned by someone.
Please make sure your english compiles.
This got a funny mod, but those OS (or OS versions if you prefer) represent a huge percentage of the market out there.
You can save even more money if you buy it secondhand or borrow it from the university library and use a scanner to copy the pages, or just nick it from the desk of the geeky kid who has just gone out of the lab to buy a snack from the vending machine.
Vintage computer adverts: http://www.vintageadbrowser.com/computers-and-software-ads
Grow a penis and install Unix.
Or it was the other way around?
What's in a sig?
This reminds me of a joke (which may only be accessable to English and Australian people).
Q. What's the naval equivalent of wine, women and song?
A. Rum, bum and gramophone records.
Boom, boom!
What a long, strange trip it's been.
When I was at university, we were specifcally told _not_ to write anything like:
main() { while (1) fork(); }
It would result in suspension of the offender's account for about 3 years (the exact same 3 years you were planning on spending at university to complete your CS degree).
What a long, strange trip it's been.
This got a funny mod, but those OS (or OS versions if you prefer) represent a huge percentage of the market out there.
So? Calling such applications "portable" is still funny.
Note to ACs: I usually delete AC replies without reading them. If you want to talk to me, log in.
I used threading in PL/1 (processes) and C (threads), and the amount of reinvent-the-wheel I had to do was really excessive. I eventually learned threading via Brinch Hansen's Concurrent Pascal, where classes and monitors (the model for Java's "protected classes") were first-class objects in the language and runtime.
Net result? I don't want to do threads in C ever again. Much like I never want to do typsetters in assembler again. So I'll use fork.
--dave
davecb@spamcop.net
This is arguably a programming language issue more than anything else.
C has a very limited abstraction/protection model, which encourages manual lock management, so it's much easier to forget to lock something correctly than in, say, C++ (to pick but one). C also has issues with unlocking resources on exceptional conditions, which makes it difficult to write robust code in a multi-threaded environment.sub f{($f)=@_;print"$f(q{$f});";}f(q{sub f{($f)=@_;print"$f(q{$f});";}f});
Just being anal... clone() doesn't quite solve the whole problem. To pick one example: with clone(), tasks still have parent/child relationships. Unix semantics are that only a parent can wait() on a child, but pthreads semantics state that any thread can pthread_join() another (in the same process). There are a number of other problems like this (and they really do cause problems in practice) which make clone() less than useful in some kinds of application.
sub f{($f)=@_;print"$f(q{$f});";}f(q{sub f{($f)=@_;print"$f(q{$f});";}f});
UNIX, in all its many forms, was developed by developers for developers
It starts badly and goes downhill from there. Unix was developed by bored research scientists who were frustrated by the bureaucracy of their MULTICS system (even the name is a play on it) and its first real use was for typesetting. They used C because it was the best compromise at the time between portability and fine-grained control of the machine.
Then he goes on to talk about semaphores and threads and a bunch of stuff he thinks makes him sound smart.
I wish newbies would grow out of the "unix mystique". There's nothing magical about it. It's a tool for getting a job done, nothing more, nothing less.
Just being anal... clone() doesn't quite solve the whole problem.
Good point. If course you can still code synchronization between tasks explicitly, and use shared memory to tranfer return values, so the inability for siblings to wait() on one another isn't a debilitating limitation.
But I'm sure there are situations in which using pthread_wait() is more convenient.
Note to ACs: I usually delete AC replies without reading them. If you want to talk to me, log in.
No, this is not a troll, despite what you may have initially assumed; hear me out.
The problem is that I *DO* program SunOS/Sparc systems, by virtue of having a portable project across Linux/x86, AIX & OSX/PPC, SunOS/Solaris, FreeBSD/x86, and Cygwin & Win32/X86. Of those, Sparc is the biggest PITA to support: rebuilding takes FOREVER, so the test/debug/test cycle is just really painful.
Really, the price/performance ratio is just not there, but we support it. Look, for the amount of money we spent on what I consider close to top-of-the-line Sparc servers to use for development, we could've bought a nice car. It does not perform nearly as well as much cheaper but much faster x86 and PowerPC machines! We're talking compile/link times in the span of hours which take minutes on x86 and PowerPC.
Care to tell me why anyone should be programming on Solaris/Sparc? Maybe Solaris/x86, but I don't think that's such a large marketshare to warrant porting serious projects to it.
Seriously, what do people do these days with Solaris programming-wise that they cannot do easier, cheaper, and faster with other architectures/machines/OSes?