Announcement

Collapse
No announcement yet.

Why FreeBSD Doesn't Aim For OpenMP Support Out-Of-The-Box

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

  • #41
    Originally posted by nslay View Post

    You're full of crap.
    Go read http://www.akira.ruc.dk/~keld/teachi...ormance.4x.pdf . Look up the performance counters and the profiling models. Does this look like something the guy writing a file-manager should be messing around with? And why should the GNU or CLANG people waste a second targeting what is clearly HPC applications? If Intel \ IBM wants this, they can spare the man-power and money on it.
    Every once in a while you get an academic writing something like https://www.computer.org/csdl/procee...0/8717a063.pdf that illustrates the trickiness of implementing a general purpose approach to OpenMP. Read on a few of those and you'll learn why no one in a rush to run this on grandma's new i3.

    Comment


    • #42
      Originally posted by c117152 View Post

      Go read http://www.akira.ruc.dk/~keld/teachi...ormance.4x.pdf . Look up the performance counters and the profiling models. Does this look like something the guy writing a file-manager should be messing around with? And why should the GNU or CLANG people waste a second targeting what is clearly HPC applications? If Intel \ IBM wants this, they can spare the man-power and money on it.
      Every once in a while you get an academic writing something like https://www.computer.org/csdl/procee...0/8717a063.pdf that illustrates the trickiness of implementing a general purpose approach to OpenMP. Read on a few of those and you'll learn why no one in a rush to run this on grandma's new i3.
      You're completely misinformed if you think OpenMP is exclusively used by HPC. It's found out in the wild in products too. Nobody has to worry about windows threads or pthreads or the capabilities of a processor or writing assembly to do tricky things because OpenMP hides all of that from you. It also tries to protect you from a lot of parallel programming bugs that are extremely hard to find and debug. OpenMP does a swell job at performance assuming you have any idea what you're doing and the costs of using (or not using) the various directives. That really goes for anything though... pthreads included (for your low level implementation non-sense).

      You really are seriously full of crap.

      Comment


      • #43
        Originally posted by nslay View Post

        You're completely misinformed if you think OpenMP is exclusively used by HPC. It's found out in the wild in products too. Nobody has to worry about windows threads or pthreads or the capabilities of a processor or writing assembly to do tricky things because OpenMP hides all of that from you. It also tries to protect you from a lot of parallel programming bugs that are extremely hard to find and debug. OpenMP does a swell job at performance assuming you have any idea what you're doing and the costs of using (or not using) the various directives. That really goes for anything though... pthreads included (for your low level implementation non-sense).

        You really are seriously full of crap.
        The profiling it takes isn't worth the trouble for most things. I don't know why we're even having this discussion in 2016. Every OpenMP 2.5+ (circa mid 2000s) publication repeats the same conclusions saying the overheads must be tightly profiled otherwise you're gaining nothing and just wasting your time.

        Just because someone is using a feature doesn't mean it's done wisely or productively.

        Link a paper saying you can just toss OpenMP on general compute problems and get away without taking the hardware into consideration through careful profiling and code fracturing and then I'll pick up this discussion thread. Otherwise, I'm tired of hearing beat up sales pitches and tired anecdotes.

        Comment


        • #44
          Originally posted by Pawlerson View Post
          I'd love to see examples. If it's true I wonder why BSD is nearly non existent in server market share, mission critical computing and so on.
          Pawlerson, Server guys generally DO NOT CARE about network latency and performance. IF THEY DID, every server on the internet would run on MCoreRT ( http://www.mcorelab.com/product-rt.html ) as it is proven, that it is faster that Linux' kernel network stack.

          These Server guys would stick to Linux, even if there would emerge an OS that is an order of magnitude faster. PERIOD.

          Comment


          • #45
            Originally posted by Pawlerson View Post
            I'd love to see examples. If it's true I wonder why BSD is nearly non existent in server market share, mission critical computing and so on.
            Server guys generally don't care about network latency and performance. If they did, they would use the MCoreRT* network stack, which can be deployed on windows and linux in userspace. Far superior to OS's build-in network stacks. PERIOD.

            Not to forget, that the Internet Routers (core routers**, not SOHO routers) are not powered by Linux. There running on something real.

            * http://www.mcorelab.com/product-rt.html

            ** https://en.wikipedia.org/wiki/Core_router

            Comment


            • #46
              Why are my comments not working? I now posted 2 times.

              Comment


              • #47
                An infamous Linux troll said: "If it's true I wonder why BSD is nearly non existent in server market share, mission critical computing and so on."

                Here is the answer:

                Server guys generally don't care about network latency and performance. If they did, they would use the MCoreRT* network stack, which can be deployed on windows and linux in userspace. Far superior to OS's build-in network stacks. PERIOD.

                Not to forget, that the Internet Routers (core routers**, not SOHO routers) are not powered by Linux. There running on something real.

                * http://www.mcorelab.com/product-rt.html

                ** https://en.wikipedia.org/wiki/Core_router

                Comment


                • #48
                  Originally posted by Pawlerson View Post
                  I'd love to see examples. If it's true I wonder why BSD is nearly non existent in server market share, mission critical computing and so on.

                  There's better option than iptables: nftables.
                  I posted article for few times which proves it's not. Generic Linux distributions like Ubuntu and others that come with MAC are more secure.
                  -Juniper core routers (and most of their other, smaller stuff) are running on JunOS, which is based on FreeBSD (FBSD10 on these days). You asked for an example.
                  -"Better option" which is not even ready..
                  -Bucket of bs. By paraphrasing Murphy's Law, everything more complicated is also more vulnerable..
                  http://www.computerweekly.com/news/450301748/Linux-botnets-on-the-rise-says-Kaspersky-DDoS-report

                  “Linux servers often contain common vulnerabilities, but not protection from a reliable security solution, making them prone to bot infections,” said Oleg Kupreev, lead malware analyst at Kaspersky Lab.

                  “These factors make them a convenient tool for botnet owners. While attacks carried out by Linux-based bots are simple, they are effective and can last for weeks without the owner of the server knowing it is the source of an attack,” he said.

                  Comment

                  Working...
                  X