From: Michał B. <mic...@ge...> - 2010-01-28 10:04:04
|
> -O mfscachefiles > preserve file data in cache (default is not to use data cache) Now this option should be used: -o mfscachemode=X where X can be: YES - all files are stored in a local cache NO - no files are stored in a local cache AUTO - storing in a local cache depends on 'allowdatacache' flag which can be set individually for every file MooseFS is a network file system and generally speaking files should not be kept in cache because from the moment of putting a file in cache another computer may change it. So cache should be used only there where you are sure about this. > -O mfsattrcacheto=SEC > set attributes cache timeout in seconds (default: 1.0) This is time of keeping file attributes in cache which are used by kernel by simple operations like "rm", "mv", etc. The system tests many times attributes of the same i-nodes and that's why we should allow for using cache here. The time should be relevantly short but not shorter than time needed for an individual operation like "rm" or "mv". We suggest leaving the default value. > -O mfsentrycacheto=SEC > set file entry cache timeout in seconds (default: 0.0, I.e. no > cache) > -O mfsdirentrycacheto=SEC > set directory entry cache timeout in seconds (default: 1.0) Both options affect keeping entry in a folder ("lookup" operation - mapping file or folder name to its i-node). This situation is similar to the above. Every operation in which we refer to the full path causes that the system every time scans file system in order to locate the given object. Most of these operations concerns folders and folder's attributes do not change often. So we decided to store in cache subfolders entries. That's why there are two options: "mfsdirentrycacheto" for subfolders and "mfsentrycacheto" for other objects. We also suggest leaving the default values here. > -O mfswritecachesize=N > specify write cache size in MiB (in range: 16..2048 - default: > 250) In order to increase performance writing in MooseFS is doen by "mfsmount" in background. That's why some part of RAM has to be used as a write buffer. This option allows to change this buffer. Lower value means lower memory usage but also potentially slower writing speed. Greater value means higher memory usage and in some cases can increase write operations. If there is no need to keep this value smaller, we recommend not to change it. In pratice increasing this buffer value does not influence a lot. So the most important is possibility to control cache of file content (which is important for the real system performance) with these options: mfsseteattr -f allowdatacache FILE_1 [ FILE_2 ... ] mfsdeleattr -f allowdatacache FILE_1 [ FILE_2 ... ] Kind regards Michał Borychowski MooseFS Support Manager _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ Gemius S.A. ul. Wołoska 7, 02-672 Warszawa Budynek MARS, klatka D Tel.: +4822 874-41-00 Fax : +4822 874-41-01 > -----Original Message----- > From: Steve [mailto:st...@bo...] > Sent: Monday, January 25, 2010 3:17 PM > To: Michał Borychowski > Subject: RE: [Moosefs-users] mfs mount generates high trafic from > chunks > > I understand the outline of what a cache does but under what situation > would > it be useful ? What is the downside ? > > Any guidance on choosing parameters or is a case of trying values and > measuring results ? > > > > > > > > -------Original Message------- > > > > From: Michał Borychowski > > Date: 25/01/2010 10:48:36 > > To: 'Steve' > > Subject: RE: [Moosefs-users] mfs mount generates high traffic from > chunks > > > > Steve, let me know more precisely what did you mean by your question > below?? > > > > > > > Regards > > Michał > > > > -----Original Message----- > > From: Steve [mailto:st...@bo...] > > Sent: Thursday, January 21, 2010 3:27 PM > > To: moo...@li... > > Subject: Re: [Moosefs-users] mfs mount generates high traffic from > chunks > > > > Hi, > > > > > > > > Is there any extra docs on the cache pro's and con's, guidelines for > usage ? > > > more explicit than the man page. |