Date: Fri, 03 Aug 2007 12:33:16 -0400 From: Chuck Swiger <cswiger@mac.com> To: Wojciech Puchar <wojtek@wojtek.tensor.gdynia.pl> Cc: freebsd-questions@freebsd.org Subject: Re: UFS and spreading data Message-ID: <46B358CC.8010804@mac.com> In-Reply-To: <20070803145528.B16127@wojtek.tensor.gdynia.pl> References: <20070803145528.B16127@wojtek.tensor.gdynia.pl>
next in thread | previous in thread | raw e-mail | index | archive | help
Wojciech Puchar wrote: > AFAIK UFS try to spread data quite evenly on disk to different cylinder > group - for large files, so small files can get it's space near inodes > etc.. Yes, UFS leaves some free space in each cylinder group if it can so that it can grow (especially small) files locally; big files will get spread across cylinder groups as a result. > but i would like to clear things up: > > i will set up say 3 disks with gconcat and make one partition for all > data on it. then i will populate it with all things and use it. > > will the data be quite spread on disks, so accesses to different things > could be done in parallel to 3 disks, or will it rather use space on one > disk first, then on second then on third. > > i'm asking about it as i prefer gconcat over gstripe as i can add more > disks to gconcat and do growfs then making system EASILY expandable. Using a stripe is going to give reliably-balanced I/O load to the underlying physical disks. If the concat is mostly empty, then no, I/O won't be evenly balanced. If you mostly fill it up and are doing multithreaded I/O to lots of files scattered all around, than concat should be OK. -- -Chuck
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?46B358CC.8010804>