View this PageEdit this PageUploads to this PageVersions of this PageHomeRecent ChangesSearchHelp Guide

cache del informe sobre sistemas distribuidos

IBM afs v 3.6 || openAfs || Arla Project


Andrew FileSytem initially developped at Carnegie Mellow University ( CMU ), then by TransArc, then IBM.
IBM forked the developpement tree in 2000 to give birth to OpenAfs, open-source, free and widely used (IPL licence).
Recently, the ArlA project implements Afs Protocol under GPL licence.
At this time the client-side is supported, the server-side is undergoing heavy developpement.

  • AFS supports a maximum file size of 2 GB.
  • RPC based.
  • AFS supports a maximum volume size of 8 GB. In AFS version 3.5 and earlier, the limit is 2 GB. There is no limit on partition size other than the one imposed by the operating system.
  • less CPU usage than NFS 2-4 times less. definetly faster. great scalability.
  • RedHat 6.x & kernel 2.2.x, likely 2.4.x as well .
  • kerberos-like or kerberos-based security.
  • client caching , server replication
  • 'encrypted' file system & authentification...
  • a fileserver can be an authserver..
  • used in many Universities ... designed to support 5,000-10,000 clients.

IBM:
  • NFS/AFS translator to keep volumes accessible by nono-afs client.
  • underliying rcp , rsh ?
http://oss.software.ibm.com/developerworks/opensource/afs/docs/html(...)
  • rsh , inetd rcp, etc... remplacement...
  • tons of documentation

  • Arla is a free implementation . client-side, working on server-side.

  • Andrew II fs.

  • Solaris client&server
  • Linux Client&server


Coda


Coda is a project also managed in CMU. it's aim is to be "better" than Afs in disconnected and weakly connected mode.

  • from Afs2.
  • RPC based.
  • no compagny support ..
  • security -> weak build-in or kerberos patch. weak == XOR scrambling, but the challenge protocol is secure.
  • namespace, volume-sharing, not directory.
  • Coda is a forked of version of AFS that support disconnected and weakly connected mode better then AFS.

client point of view :
  • disconnected operations , caching ...
  • open() means copy() - based on file replication/caching.
  • small 10-200 Mo caching recommended. It seems that coda's users have larger cache area.
  • file to cache selectionnable ( hoarding ).
  • kernel support needed . very little patch, user-land application ( Venus ).
  • on update conflicts, user might need to resolve it manually..

server pov:
  • replication
  • consistency & transactions


  • DFS sysadm seems to prefers Coda to Afs.
  • Coda seems not to be scalable :
The documentation says the RVM metadata needs to be 4% of the total shared size and it needs to be backed by virtual memory for that same quantity.

  • Linux client&server
  • Solaris client , no server


Ficus, Rumor & ROAR


  • designed at UCLA for large scale DFS.
  • peer-to-peer structure.
  • optimistic consistency approach.
  • intelligent resolvation of update conflicts.
  • Rumor user-space implementation is beta stage.
  • intelligent hoarding.
  • Really accurate for mobile and disconnected DFS.

Sistina GFS 5.1


  • used for cluster's filesystem on fast LAN.
  • not useable on WAN networks.

DAFS ( http://www.dafscollaborative.org )


  • file sharing on SAN . Not for WAN .

Sprite fs, part of SpriteOS


  • no caching on write-shared files.
  • spritely NFS ?
  • less CPU-eater than NFS.
  • superseeded by afs.


xFS - A Wide Area Mass Storage File System


  • wwwos93.ps
  • multi hierarchical & caching
  • -> IBM XFS ? no network issue and LAN-applicated.

Amoeba


  • Amoeba is a DOS, so there is a DFS in it..
  • micro-kernel and stuff..

Odyssey


  • application-aware.
  • need the application to be compliant.


SFS @ sfs.net


  • secure, encrypted...
  • basicaly , it's a encrypted NFS.
  • relies on nfsv3.
  • easy to install.
  • a lack of pam modules ?

SFS @ http://www.cs.auckland.ac.nz/~pgut001/sfs/ (...)


  • secure, encrypted...
  • windows & Dos

Lustre


  • Another LAN-cluster dfs.


LBFS - A Low-bandwidth Network File System


  • Whole file Caching
  • NFS-like protocol. Over-NFS implementation.
  • Chunk SHA-1 calculation and comparaison.
  • Very _LOW_ bandwith utilisation.
  • SFS encryption support
  • Not for disconnected operations.
  • High disk load.
  • Need a Berkeley DB.
  • Probably high CPU load .
  • Less bandwith use than NFS,AFS...
  • Faster execution time.
  • Designed for Wan access.
  • 2 implementations
    • a) with the xfs-arla client - caching, chunking but no authentification
    • b) with the SFS client - authentification but no chunking,
  • small- freebsd nfs kernel patch, no linux patch .

InterMezzo


  • pretty much like Coda.
  • AFS -> COda -> Intermezzo
  • Aiming disconnected ops.
  • GPL
  • Module-Compilated in Redhat 7.3
  • More efficient than Coda is.

Link to this Page