Coss Under standing:

            This is what my under standing about the coss file system through your document COSS basics.


Here I am giving my pictorial representation of the COSS working.


Consider the memory stripe .


A              B            C               D





Assuming above memory stripe contains 4 blocks ABCD. With the predefined object size of 13KB each.


Only the object size 13KB and less will be stored. Once it gets filled it will be transferred to a hard disk. If the mem-hit request comes, it will be served from the stripe in memory.


If there is a cache-hit request then the object will be fetched from the disk and will be stored in to head of the current memory stripe.


Once it is filled it will be transferred in to the disk.


Here what I think as a main disadvantage. If my understanding is correct


Redundancy of the object in the cache.  More write in the disk, which should be avoided.


This continues until the disk gets filled. Once disk is full the tail gets removed from the disk (LRU).


Adrian what will be amount performances, we are going to get once it is implemented successfully.


For COSS file system is there a limit that we should have only the cache size of 2GB.


I discussed with Joe he told as


 COSS will /not/ work right now.  It is broken under Linux.


What is the other possibilities of improving the Disk I/O ?  If you have any ideas share with me I am really interested init?