Date: Wed, 18 May 2011 12:24:25 +0200 From: Bernt Hansson <bernt@bah.homeip.net> To: Devin Teske <dteske@vicor.com> Cc: FreeBSD Questions <freebsd-questions@freebsd.org> Subject: Re: Buildworld Benchmarks Message-ID: <4DD39E59.60606@bah.homeip.net> In-Reply-To: <022d01cc14f5$a37ef010$ea7cd030$@vicor.com> References: <022d01cc14f5$a37ef010$ea7cd030$@vicor.com>
next in thread | previous in thread | raw e-mail | index | archive | help
2011-05-18 02:50, Devin Teske: > Hi List, > > What's the fastest anyone has every completed buildworld on a single machine? No idea. > The reason I ask is because we just got some new hardware in and decided to > benchmark it using buildworld. Nice! > Just as a quick test, we decided to perform "make -j 48 buildworld". We finished > in approximately 9 minutes. For me on the desktopsystem 33 min with make -j 24 buildworld Half an hour is half the lunchbreak. FreeBSD 8.2-STABLE #0 CPU: AMD Athlon(tm) II X2 250 Processor (3013.63-MHz K8-class CPU) Origin = "AuthenticAMD" Id = 0x100f62 Family = 10 Model = 6 Stepping = 2 real memory = 17179869184 (16384 MB) avail memory = 16530083840 (15764 MB) FreeBSD/SMP: Multiprocessor System Detected: 2 CPUs FreeBSD/SMP: 1 package(s) x 2 core(s) > I think that we can improve upon that, but am having a bit of difficulty. > > Can anyone offer any pointers in how to achieve the fastest buildworld possible? > No particular reason... we're just trying to push the boundaries of what's > possible. Check the handbook, single user comes to mind. > For reference the machine we're compiling on is a dual-socket Nehalem Xeon > (six-core per proc; HTT enabled; 24 total CPUs presented by APIC) with 48GB of > RAM, an LSI MegaSAS RAID controller, and an LSI 2Gbps Fibre Channel HBA going to > an 8TB NEC D-4 array. > > ASIDE: Doing the same buildworld on a 4-disk ZFS raidz yielded approximately > 11-minutes. Performing the buildworld on the NEC D-4 over the 2Gbps FC HBA > yielded approximately 12 minutes. And for some unknown reason, performing > buildworld on tmpfs yielded 13 minutes. > > We thought going tmpfs would make things faster, but that resulted in over 13 > minutes (huh? you'd think a RAM disk would be smoking compared to even the SSDs > that we used to achieve ~9 min; do note that we did make sure to nullfs mount a > tmpfs-based directory onto /usr/obj -- though the performance of that nullfs > mount might have hurt the test, not sure).
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?4DD39E59.60606>