Date: Wed, 3 Sep 1997 17:45:27 -0700 (PDT) From: lamaster@george.arc.nasa.gov To: current@FreeBSD.ORG Subject: Re: IDE vs SCSI was: flags 80ff works (like anybody doubted it) Message-ID: <199709040045.RAA27340@george.arc.nasa.gov>
next in thread | raw e-mail | index | archive | help
In re: dd if=/dev/rwd0 of=/dev/null count=1600 bs=64k "dd" is a nice benchmark for some purposes, but many people have frequent results from something a little closer to home. When building new kernels and doing "make world" and so on, I find that the system I use that is a 166 MHz Pentium (Triton I - no busmastering EIDE) EIDE + Seagate Medalist Pro drive (supposedly one of the fastest EIDE drives when I acquired it, recently) is about 2.5 - 3.0 X slower than the other system I use, a 200 MHz PPro (256K L1 cache) (ASUS Natoma) w/ Buslogic BT-958 and Quantum Atlas (I) (XP34300) [now almost two years old technology]. Now, I haven't checked to see how much of the difference is due to non-busmastering EIDE vs busmastering EIDE, (the Triton I doesn't do busmastering EIDE), how much is due to the slower seek/latency time of the Seagate, how much due to the slower CPU, but, one way or the other, the system with the fast SCSI drive sure is faster. Perhaps someone has studied these factors? I did a simple "make world" today (Wed Sep 3 17:17:18 PDT 1997), on the PPro200 SCSCI system, with today's 3.0-current, and the elapsed time was almost exactly 3 hours. It might be interesting to compare system components with faster/newer technology that can improve on the 3 hour time significantly (surely the Seagate Cheetah, IBM Ultrastar 2XP, and Quantum Atlas III are faster, perhaps, a PPro w/ 512 KB is significantly better, or possibly a Pentium II 266, although I suspect I/O is most of the time, not CPU compile time. How about a dual Natoma-based PPro 200 system using the new SMP kernel-possibly better I/O overlap, although, possibly it would thrash the disk and actually slow down the build. Maybe some SCSI controller is faster than the Buslogic, or maybe not?) So, any configurations out there significantly faster than 3 hours for a "make world" of 3.0-current? I doubt if any of those systems use EIDE disks, but, I have been wrong many times before. Results?
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199709040045.RAA27340>