Slashdot is powered by your submissions, so send in your scoop

 



Forgot your password?
typodupeerror
Linux Business

Journal: Looking at Linux Filesystems, Reiser-FS discussion

Journal by jets42
I've been thinking for a few YEARS now about changes that should be made to unix/linux file systems. I used to work on VAX/VMS computers, and still miss some of the things liek file versions... I like several of the things I've read about Reiser-FS http://www.namesys.com/
and will read more of the discussion here: http://interviews.slashdot.org/comments.pl?sid=66948&threshold=1&mode=thread&commentsort=0&op=Change

It looks like ReiserFS has many of the software hooks thtat would allow me to add things I've wanted under unix for a long time...

Some of my (random) thoughts:
File System Modifications:
  • Keep MANY file attributes, not just version
  • Hidden (grocked .dot-hide notation to be obsoleted)
  • Backup (hooks to support optimum in/out)
  • Comments + notes (think MAC forks or metadata)
  • Version control (CVS like even, built under the hood)
  • Access logs (triggers possible)
  • File Version Support:
    - Make it automatic, with file system control Settings
    Version frequency (can preset parameters in attrib)
    Each change, open + close
    can disable on file granularity, for DBMA & big files
    can open() func easily reach older versions?
    VER++ conditionals:
    - Daily, or past N-Age
    - Never
    - Past certain percent
    - Group version control
    Storage type
    - Delta SCCS like
    - Complete duplication
    Retention
    - Forever
    - N-Age
    - N-Version copies
    - Rollbacks...
    Access
    - All see same
    - Rollback views
    - See all individual files
    - See one file, others hidden
    - Forks- can different people branch..?
    - See what other people have done
    - VAX/VMS

Can you keep list of ASSOCIATED FILES..? and category, and type of relationship, and importance.


  • - Associate executables that can/have open file
    - Which exe originally wrote file
    - What project is this part of?
    - what other dependant files? (make like)
    - what files depend on you?
    - categories? (sys,doc, config, user pref, tracking, accounting, etc)
    - keywords? importance? (context)
    - How much of this can be automatically maintained? ... setup... initialized
    - we REALLY need metadata, associated with files
    - How to you handle metadata with older unix commands? TAR, serialize, cat, grep, cpio, backup/restore

security concerns Activity logging in file system (machine form & people form) weeding out misc from important? to answer: What has been happening on this machine in last week? (with 100 users)
- or what was I working on yesterday?
- or what jobs did run (outcome, data changed)
- or what jobs WILL be run? (expect load)
- or network activity + flavors + levels + who + which programs +
- oh can you link net activity to people, programs, ports, files, ???
- and what about databases: big nebulous hole, who changed what, with what program, queries, time

Associate MANY files with
- What other files are used in conjunction with this one, or for similar purposes..? For instance: hereer are all of the ifles that have to do with networking, or a project, or a database, or a dat-set NNN, or backups, or X-Windows configuration... yadda yadda...
- You will need a separate place in metadata database to keep the association
- Visibility might be system side, or particular user, or group, perhaps context specific (by user, role, etc)
- How about relevance factor- or strength of association???
- General category - as opposed to link to group??
- Files can/will be in MANY groups, not just one- with strengths diff of association to each group
- You will want to search/list files by this information - including many categories
- It sure seems like it becomes an SQL database, imbedded into filesystem, or associated like seen project
- system performance impacts?

Torque is cheap.

Working...