From: Greg Ercolano <erco@(email surpressed)> Subject: [Q+A] What are the benefits/drawbacks of AFP/NFS/SMB mounting on Date: Mon, 17 Oct 2005 16:50:21 -0700 |
Msg# 1062 View Complete Thread (1 article) | All Threads Last Next |
> What are the goods and bads of the different network file system mounts > under OSX? I know AFP, NFS, and SMB are available.. which is best? For Rush, it's best if you use a file system that remains mounted regardless of interactive user activity, ie. if the interactive user logs out, you don't want mounts to disappear, as this will affect renders running in the background with Rush. NFS seems the best choice for that, and for handling permissions. Unfortunately NFS doesn't handle resource and data forks well under OSX.. *** UPDATE JANUARY 2011 ************************************************************ *** Since this article was posted, Apple nfs has since been stabilized greatly. *** In Tiger (10.4.x) and up, resource/data forks are now handled properly. *** Apple has chosen to go the NFS route, since the other network file protocols *** (AFP and SMB) are simply not multiuser file systems. ************************************************************************************ ..it leaves behind little "._" files which work fine until the user tries to move the original file to a different directory, without also moving the "._" file, causing the data/resource fork association to be lost. I advise configuring mounts to automatically be mounted on boot in OSX. There are some articles in the rush newsgroup on this. *** UPDATE JANUARY 2011 ******************************************************************** *** Articles I would recommend on configuring mount-on-boot: *** *** [OSX/ADMIN] Boot script to mount multiple file servers *** http://seriss.com/cgi-bin/rush/newsgroup-threaded.cgi?-view+1847+1847 *** *** [OSX/Admin] How to create a custom boot script to mount a file server *** http://seriss.com/cgi-bin/rush/newsgroup-threaded.cgi?-viewthread+1710+1731+1783 *** *** Subject: [OSX/SYSADMIN] Example NFS setup for Rush on a small 'pure OSX' network *** http://seriss.com/cgi-bin/rush/newsgroup-threaded.cgi?-viewthread+1295+1297+1298+1839 *** ******************************************************************************************** Also, which network file system protocol you choose can make a difference in speed of rendering; under OSX you have many choices: AFP, SMB, NFS. Each has its own benefits and drawbacks. In short: AFP - BENEFITS: native to OSX, handles resource/data forks easily. DRAWBACKS: some servers don't implement AFP very well, and are suseptible to load. AFP does not handle multiuser very well; similar to SMB, all file I/O through the AFP mount will inherit the permissions of the user who the mount was created as. Also, mounts may automatically disappear if the interactive user logs out. And finally, as of Tiger (10.4) AFP became a single-user filesystem, in that only the owner of the mount point can access the server.. preventing multi-user access. For more see: http://seriss.com/cgi-bin/rush/newsgroup-threaded.cgi?-viewthread+1013+1014+1015+1019+1020+1021+1024 NFS - BENEFITS: can be mounted on boot, and mounts will stay in place even if user logs out (so as not to affect background rendering). Handles multiuser access correctly; if root makes the mount, any user can access the mount, and permissions will work normally; files will be owned and permed by the user accessing the file, and not by the user the mount was created as. DRAWBACKS: OSX uses ._XXXX files to handle resource and data forks, which can cause trouble when users move files around at the unix prompt with 'mv', but forget to move the ._XXXX files too. NFS may need tuning to work efficiently with some servers. NFS uses UDP instead of TCP, which means under high load, it may not handle throttling well. (TCP implementations of NFS, if available, may work better for bandwidth control) SMB - BENEFITS: Well, if you have Windows file servers, this is often the only way to mount them. In some cases people have reported faster file i/o with samba than with NFS. Depends on tuning. DRAWBACKS: not a multiuser file system; like AFP, all file I/O through the smb mount will inherit the permissions of the user who the mount was created as. And finally, as of Tiger (10.4) SMB became a single-user filesystem, in that only the owner of the mount point can access the server.. preventing multi-user access. See: http://seriss.com/cgi-bin/rush/newsgroup-threaded.cgi?-viewthread+1013+1014+1015+1019+1020+1021+1024 There may be other reasons; those are ones I've run into. [EDIT: added links re: single user access limitations of SMB and AFP -- erco 10/20/08] |