Date: Tue, 10 Oct 2006 10:59:48 +1000 From: "Arthur Hartwig" <a_hartwig@fastmail.fm> To: "Charles Ulrich" <charles@idealso.com> Cc: freebsd-smp@freebsd.org Subject: Re: FreeBSD 6.1 Instability Message-ID: <1160441988.23832.272935291@webmail.messagingengine.com>
next in thread | raw e-mail | index | archive | help
I have had good experience with SMP in FreeBSD 6.0. I have a system with two CPU packages with hyperthreading (4 CPUs in all). It runs the SMP kernel, mostly supporting software development (file serving, compiling, etc). It runs for weeks without stalling or hanging. I have noticed though that the SMP performance in networking, particularly forwarding, is not nearly as good under SMP as on a single CPU system. I wonder if you are getting sustained heavy network traffic. The LAN driver polling option works well on SMP systems and has a measure of protection against heavy network loads consuming all the available CPU time (the drivers effectively stop processing incoming packets if the system idle time gets too low). # man polling gives more details on polling. The kernel needs to be built with option DEVICE_POLLING, polling needs to enabled on individual interfaces (can be done in /etc/rc.conf) and, depending on your network load, you might want to tweak sysctl variables such as kern.polling.user_frac Arthur ext Charles Ulrich wrote: > Greetings, > > We have been running FreeBSD on our mail servers for about as long as I can > remember. Recently, we decided to go SMP to handle increased mail load. After > assembling the hardware, installing the OS and software, and restoring all of > our data, we noticed in testing that our first machine began hanging > semi-regularly when it began processing lots of mail. Disabling SMP > eliminated the hangs completely. We tried it all again on completely > different hardware with exactly the same result. Our conclusion: somethings's > buggy in SMP. > > Here are the symptoms. The machine hangs, and becomes completely > unresponsive. It looks like a deadlock. It will sometimes respond to the > power button and shut down (without being able to first sync and unmount > filesystems), and sometimes the power button event gets caught in the > deadlock. Sinceit's not actually a crash, there is no core dump or other > debugging information. In the most recent situation, it hung at different > points every time I tried to compile ezm3, after successfully compiling other > packages. > > We're system administrators, not kernel hackers, so this is a plea for help. I > wouldn't know where to start, but I'm hoping someone can point me in the > right direction. We're also willing to give a (trustworthy) FreeBSD developer > root access to the test machine since it's just sitting idle right now. If > you need to crash it, that's fine. We'll have people during normal business > hours who know how to push a reset button. > > Thanks for your time. -- Arthur Hartwig a_hartwig@fastmail.fm -- http://www.fastmail.fm - A no graphics, no pop-ups email service
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?1160441988.23832.272935291>