|
|
The future exists first in the imagination, then in the will, then in reality. | |
|
WHITEBOARDS links into revise activity how to use SITENAV meetings what's new? whiteboards post article frontpage downloads ORIENTATION legalisms history glossary participants BACK-ENDS udanax-green udanax-gold ALGORITHMS coordspaces enfilade ent OLD MANUALS XIA HELPING puzzles needs funding site-traffic admin
|
Taken from a forum message by Eric Hill on Dec 31, 1988. The Unusually Reliable Disk Interface developed by Michael Mc Clary gives us a degree of incremental recovery from system failures that we have long been wanting. As with any heavily cached system, Xanadu runs the risk of having an inconsistent data structure on the disk after crashes and power failures. Buffering in most operating systems and high-end disk drives only aggravate this problem. The URDI allows a consistent, although not necessarily current disk data structure to be achievable after when restarting the system after some, but not all failure modes. Some the modes that we are not protected against are spiral writes and head crashes. The metaphorical model that Michael has used is Fibber Mc Gee's closet. The closet represents the main storage area on the disk where most reads take place, and a consistent data structure is more or less guaranteed during execution of user requests. Writes, however, go through a staging area on their way to the closet, where groups of snarfs are placed in temporal order. At the end of each group is a marker indicating that if all of the staging area up to that point has been put in the closet, the disk will be consistent. After a failure, this area will be processed before acknowledging any requests. While this will result in some delay at system restart, it is immensely faster than reevaluating the entire transaction log. --JohnDougan? |
|
| All trademarks and copyrights on this page are owned by their respective companies. Comments are owned by the Poster. The Rest ©2001 Jeff Rush. |