On Mon, 2005-12-05 at 20:17 +0200, Muli Ben-Yehuda wrote: > On Mon, Dec 05, 2005 at 07:43:28PM +0200, Gilboa Davara wrote: > > On Mon, 2005-12-05 at 17:49 +0200, Muli Ben-Yehuda wrote: > > > On Mon, Dec 05, 2005 at 05:05:04PM +0200, Gilboa Davara wrote: > > > > > > > The application we tested on was a kernel based network system that did > > > > a lot of I/O (with network cards and FS) and a huge memory hog. > > > > We never really bothered to find out why HT performed so bad (we already > > > > decided to dump the Xeons and use the AMD Opteron instead), though the > > > > cache trashing does sound reasonable. > > > > > > What was the test? HT compared to what? two physical CPUs? single > > > physical CPU? > > > > IBM e345 Dual Xeon with two 2.8/533 Xeons. > > SuperMicro single socket Xeon with one 2.4/533 Xeon. > > Thanks for the specs, but that doesn't answer my question. Unless you > mean that the first configuraton (e345) was 20% slower than the > second?!
Oh sorry. IBM e345: Configuration 1/1: RHEL3, 2.4.21, HT off, 2 CPUs. Configuration 1/2: RHEL3, 2.4.21, HT on, 4 CPUs. (2 CPU * 2 Logical cores.) SuperMicro: Configuration 2/1: Embedded Linux, 2.6.10, HT off, 1 CPU. Configuration 2/2: Embedded Linux, 2.6.10, HT on, 2 CPUs. (1 CPU * 2 Logical cores.) > > Cheers, > Muli ================================================================= To unsubscribe, send mail to [EMAIL PROTECTED] with the word "unsubscribe" in the message body, e.g., run the command echo unsubscribe | mail [EMAIL PROTECTED]