Erich,

and suddenly server3 literally explodes, like, there's no way I can get
anything on the disk back, is my entire file system screwed?  Or just
some files?

You will lose,
- Any file whose stripe is stored on that node is screwed.
- You won't lose the entire file system unless the node that crashed
was the MD server and you configured the system with exactly 1 MD
server.
- Access to any files whose metadata was stored on the node that crashed.

You can still access parts of the file provided the server that went
down did not store the metadata required to access the file.

 For example, if I have a 4MB file, is that single file
spread out over the 4 nodes or is it striped in a linear fashion all on
nodes, i.e. does node 1 fill up first then go to node 2, etc.?

It is striped in a linear fashion across all the nodes in a
round-robin fashion (by default). There are other distributions for a
given file.
Spose file stripe is 4 MB, you will have 0-4MB on node 0, 4-8 MB on
node1, and so on. The starting node number is randomized on file
creation so that we don't fill up node 0's disk space unfairly and
cause needless hotspots.

I'm
assuming it's not exactly set up as a RAID5 type redundant system, such
that if I lose a node and then replace it, it will "rebuild" the file
system data on that node?

Nope; there is no RAID-5. It is a classic RAID-0 style distribution.


I need to be clear on this before putting a lot of time into it, but it
sounds like this might be a good solution for our firm, as we have a 200
node cluster each with one 500GB disk, 400GB of which can be leveraged
to a massive parallel file system (400GB x 200 nodes = one big ~80TB
distributed file system).  But that assumes that there is no redundancy,
other wise that 80TB would be more like 50-60TB max or something because
there would be some redundancy in there... ?

There is no in-built redundancy yet. Unless you setup something like drbd.
Hope you find this information useful.
Do keep us posted.
thanks,
Murali

Thanks for any clarification!

-Erich
_______________________________________________
Pvfs2-users mailing list
[email protected]
http://www.beowulf-underground.org/mailman/listinfo/pvfs2-users

_______________________________________________
Pvfs2-users mailing list
[email protected]
http://www.beowulf-underground.org/mailman/listinfo/pvfs2-users

Reply via email to