Date: Tue, 20 Apr 2004 10:30:05 -0700 (PDT) From: Tom <tom@sdf.com> To: Kris Kennaway <kris@obsecurity.org> Cc: stable@freebsd.org Subject: Re: HTT and SMP question Message-ID: <20040420102137.Q97493@light.sdf.com> In-Reply-To: <20040420081334.GA56291@xor.obsecurity.org> References: <20040419085841.GB64662@freenix.no> <20040419090049.GA51659@chihiro.leafy.idv.tw> <20040419093523.GA34419@xor.obsecurity.org> <20040420081334.GA56291@xor.obsecurity.org>
next in thread | previous in thread | raw e-mail | index | archive | help
On Tue, 20 Apr 2004, Kris Kennaway wrote: ... > My parallel package builds are slower on build machines with HTT. > It's not a magic bullet. It is more complicated than that. There isn't just one kind of HTT. Intel has been tweaking the HTT feature since it was first released. First generation HTT was known to have some performance irregularities. HTT is designed to utilize more instruction units on the CPU at the same time by running two threads of execution. So at worst, it shouldn't cause any slow down versus non-HTT. However, first generation HTT definitely did slow down on some work loads. Intel says they have fixed this in the newer generation processors. There are also FreeBSD factors: FreeBSD 4 compiled with SMP, avoids a lot of locking code. FreeBSD 4 with SMP has a lot of extra locking code, and that code has cost associated with it. The cost of SMP locking in FreeBSD 4, may out way the benefit of HTT. With FreeBSD 5, the SMP locking is done quite differently, and it is done all the time anyhow (since FreeBSD 5 is SMP by default). So HTT on FreeBSD 5 is more likely to be beneficial than HTT on FreeBSD 4 (with a non SMP kernel). > Kris Tom
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20040420102137.Q97493>