Date: 20 Apr 1999 16:26:28 +0300 From: Ville-Pertti Keinonen <will@iki.fi> To: cyouse@cybersites.com (Chuck Youse) Cc: hackers@freebsd.org Subject: Re: NT4 server 2.5 times faster than Linux Message-ID: <86iuar5sor.fsf@not.demophon.com> In-Reply-To: cyouse@cybersites.com's message of "19 Apr 1999 16:55:06 %2B0300" References: <199904152225.SAA20324@etinc.com> <99041909500702.38298@ns1.cybersites.com>
next in thread | previous in thread | raw e-mail | index | archive | help
cyouse@cybersites.com (Chuck Youse) writes: > The ability of NT to take better advantage of multiple processors is a function > of design, rather than "working with Intel". NT follows the microkernel design > model, with message passing, which is much easier to spread over multiple > processors than a typical monolithic kernel design. A monolithic kernel design does not restrict scalability, as long as locking is sufficiently fine-grained. In Linux, it isn't (even though some people claim that it scales just as well as truly fine-grained kernels). Current Linux versions *do* have slightly finer-grained locking than before (it's finer-grained than FreeBSD) but most filesystem and network system calls hold the giant lock during everything, so things like accessing cached data can't be expected to scale. Basically, the scaling of Linux/SMP (and FreeBSD/SMP) is currently pretty bad for kernel-intensive applications, NT doesn't need to be exceptionally good to be better. Message passing can help parallelism for asynchronous operations, but this is not typical in a general-purpose system. To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-hackers" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?86iuar5sor.fsf>