Messages in this thread | | | Date | Tue, 23 Mar 1999 15:56:54 -0800 (PST) | From | Drew Roselli <> | Subject | Re: disk head scheduling |
| |
Matthew Jacob writes: > At any rate, my notes are pretty incomplete on this talk- it *was* a quick > work-in-progress talk (< 7 minutes presentation+questions), and I can't > find the online version of the abstract, so maybe the author (I *believe* > that is 'drew@cs.berkeley.edu'- that's what my printed abstract says- but > it might have been Jeanna Neefe (neefe@cs.berkeley.edu)- I was quite > spaced out by that point) could comment or make a more complete version of > the paper available for perusal? > > > -matt > Right, I gave this talk. My presentation suggested enriching the file system-storage system interface so that storage systems (which can be quite complex today) have information pertinent to making intelligent storage decisions.
I've put the talk slides online and the WIP abstract is included below. http://http.cs.berkeley.edu/~drew/Papers/papers.html
There currently isn't a complete version of the paper -- research on the topic is still in progress.
--Drew
-------------------------------------------------------------------------
Enriching the File System-Storage System Interface
Many file systems are designed to carefully place data on disk to provide good performance. In order to do this, the file system must know specific characteristics of the storage device. However, today's high-end storage system products are complex. They typically contain many disks, persistent memory, and one or more redundancy schemes. These storage systems use their own layout scheme, which may undermine the layout proposed by the file system. The file system-storage system schism leads to the following dilemma in layout decisions: the file system has more information about the likely access patterns of files but the storage system has more information about the performance of the storage media.
I propose that the file system provide abstract rather than absolute file positions to the storage system. In addition, the file system can predict how long a given block is likely to live and how frequently it will be accessed. This information can then be sent as a hint to the storage system along with the block to be stored.
- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.rutgers.edu Please read the FAQ at http://www.tux.org/lkml/
| |