Date: Wed, 08 Jan 2003 17:35:12 -0800 From: Terry Lambert <tlambert2@mindspring.com> To: Michal Pasternak <michal@pasternak.w.lub.pl> Cc: paul beard <paulbeard@mac.com>, advocacy@freebsd.org, sellis@telus.net Subject: Re: [Fwd: web write-up] Message-ID: <3E1CD1D0.7D8914B8@mindspring.com> References: <3E1C8287.1020506@mac.com> <3E1C9249.551D6932@mindspring.com> <20030108220533.GB18151@pasternak.w.lub.pl>
next in thread | previous in thread | raw e-mail | index | archive | help
Micha=B3 Pasternak wrote: > We all know, that 99% of benchmarks and statistics are untrue, > but does any of you know about a recent surveys comparing > speed of (network|filesystem|vm|anything else) on FreeBSD > and Linux? I don't know of any recent statistics in this area. You would be better off asking the question after the end of Februrary, so someone could point you to the most recent Usenix proceedings, which will happen about then. > I'm especially concerned about softupdates + dirhash compared > to ext3fs or reiser. What specifically would you like to see compared, that's actually representative of expected real-world performance? For example, creating a boatload of files, all in the same directory, is never likely to happen with real applications, as long as they are written by people who know what they are doing, and want their performance to be protable to UNIX systems, in general (one of the problems with the mail server that started this thread, two yearts ago). Reiser is very hard for me to comment on; I believe that, even after they removed preserve lists, it still substantially infringes US Patent #5666532 (USL/Novell). Personally, I also dislike "dirhash", since in my opinion, it does not scale; it's author admits as much, and, for my money, it is addressing a symptom rather than a real problem. Comparing Extent based file systems or log structured file systems against block clustering-with-hashed-selection filesystems is, I think, comparing apples and oranges. They solve different problems, and the ones I think are most important are the ones that occur in the acknowledge the old dictium: "The steady state of disks is 'full'" -- Kirk McKusick =2E..meaning that you'd expect more of the system time, percentage-wise, to be taken up by the operation of a cleaner, on FS's whose designs did not include intrinsic protection against fragmentation. If, on the other hand, you have some ideas, then you should pursue them, and publish your papers at one of the academic conferences. > NVidia's drivers performance should be > also easy to perform. Notification of completion isn't really a feature of this hardware; I think the best you will be able to measure is something not very useful, like "filled triangles permitted by the driver to be queued to the device, per second, with no knowledge of the real render-rate", and I don't know how useful that would be, anyway. Even if I would personally disagree, most people seem to believe the propaganda that FreeBSD is not an OS for the desktop, or for games on the desktop. Taking that into account, though, since the company that made the cards made the drivers for both the OS's you are imputing a comparison between, without coming right out and saying it, I expect that they are as fast as the hardware vendor could driver their hardware. Any differences would be in what the programmers learned from previous implementations (which would favor the drivers implemented later over those implemented earlier). In any case, if you think it's possible to compare something that's really tied to presentation to a human, and do it so it's not a subjective comparison, you should do it. You still have time to submit your paper for the Freenix track for the Summer Usenix. -- Terry To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-advocacy" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?3E1CD1D0.7D8914B8>