[PREV - HANS_REISER]    [TOP]

FILE_SYSTEMS_REISING


                                                     February 19, 2017     
                                                                           
Hans Reiser wrote software: he                                            
developed one of the first               I've never gotten a good feel for
serious file systems for Linux,          why Reiser4 was never adopted into
"Reiser3" and was working on a           linux core: as I remember it, Linus
follow-on project "Reiser4" that         Torvalds was complaining that
was never really accepted by the         Reiser wouldn't modify his code to
linux developers.                        comply with linux standards... it
                                         sounded to me like some sort of
  I've always liked Resier3, and         social breakdown.  Hans Reiser
  regret that it seems to be             could certainly be pretty crazy,
  becoming a second-class citizen        and Linux Torvalds can be pretty
  in the linux world: there are a        cantankerous himself...
  number of file systems available      
  now that are designed to scale        
  upwards to deal with very large              LINUX_REISING
  files, but Reiser FS (and Reiser      
  FS alone, as far as I know) was       
  designed to scale down to cover       
  the case of large numbers of          
  small files.                          
                                        
     That fits in well with the    
     way some classic unix              
     software works e.g.  email         
     servers, or (a personal            
     favorite) the "mh"-style                  
     mail client software.                  There are a number of odd hacks    
                                            that are suprisingly common    
     Most of us are somewhat aware of       where a system administrator    
     the limitations of our filesystems,    will decide to alphabetize a    
     and have gotten used to dancing        large set of files, and move    
     around them: if you tried to           them into sub-directories named    
     architect software to use lots of      for each letter.  That's       
     small files these days you'd find      evidently a performance win for  
     someone intoning "don't treat the      many systems... but why would    
     file system like a database!".         anyone put up with having to do    
                                            this kind of work manually?    
     Hans Reiser (and again, Hans                                      
     Reiser alone as far as I               (On the old Stanford TOPS-20
     know) decided that the file            systems, everyone's usernames had
     system should *be* a                   an odd structure to them like
     database, and went after               "J.JBRENNER", i.e.  "JBRENNER" in
     those restrictions.                    the subdirectory "J".)
                                        
       He published a paper on                  Consider the "mbox" file
       the subject of "namespace                format (popularized by
       unification", which tells                the Berkeley "mail" program)
       you a lot about the thinking             which stores a large
       that underlied Reiser4...                stash of many mail messages
                                                glued together into one
           Note: he named his                   file.  That's no doubt more
           company "NameSys".                   "efficient" in many respects,
                                                but why should a programmer
                                                be forced to work that way?
                                                Isn't it reasonably natural
                                                to think in terms of one
                                                file per message (and one
                                                directory per mail folder,
                                                and there you have the old
                                                "mh" approach).



    From Wired magazine, 2006.
    http://www.wired.com/wired/issue/15-07
    Joshua Davies:

    "Reiser4 is built on this principle. It contains
    a single registry-- known as a balanced tree--        The use of the term
    to organize every piece of data in the operating      "balanced tree" here
    system. All programs can employ the same              is just enough off
    nomenclature to access information. In the            to make it clear that
    traditional Windows or Unix file system, each         Davies isn't a
    application uses different terminology to track       programmer.
    data. As a result, programs don't communicate
    efficiently with one another, which creates           The registry isn't
    walls between data. In Reiser's idealized             a balanced tree...
    vision, a simple search for the word 'Nina,' for      we might say it's
    instance, will turn up emails in Outlook, images      keys are arranged
    stored in MyPictures, and credit card charges in      in a balanced tree,
    Quicken. Google's desktop search and Apple's          or that the index
    Spotlight feature can do this, but they're            is a balanced tree,
    afterthoughts: The ability of Reiser4 to scour        or something like
    every document is embedded in the operating           that.
    system itself. It lays the foundation for a
    digital universe where there is no                       But then, just
    discrimination-- nothing is walled off. In this          calling it a
    unified space, all things can be known."                 "registry" is
                                                             a little tone
      Obvious question:                                      deaf: it reminds
      "Security?"                                            of the windows
                                                             registry--
      One would hope                                         widely regarded
      they thought                                           as a disaster by
      about it.                                              unix people
                               This use of language is       everywhere, who
                               essentially a matter of       prefer files
                               social flags, however         of "plain text",
                               much we might want to         (whatever that's
                               pretend it's a matter of      supposed to be).
                               technical understanding.   
                                                              
                                    For all I know, Davies may        
                                    understand the *idea* of          
                                    balanced trees very well.         
                                                                                
                                       (Perhaps better than I do: my            
                                       understanding of such topics is vague    
                                       unless I feel a need to dive into        
                                       them, and typically I don't: I don't     
                                       claim to have a CS background, and       
                                       make no apologies for not thinking       
                                       like someone who does.)                  
                                                                                
                                                                                
                                                                               
                                                                         

--------
[NEXT - REISERS_GUILT]