THE LAFS KERNEL FILE SYSTEM

OVER THE PAST 15 YEARS IMPROVEMENTS IN THE PRICE AND CAPACITY OF DISK DRIVES HAS ROUGHLY MATCHED THE DRAMATIC GAINS MADE IN PROCESSOR SPEEDS. HOWEVER, RAW I/O AND SEEK RATES OF DISKS HAVE IMPROVED FAR LESS, AND THE BASIC ARCHITECTURE OF KERNAL FILE SYSTEMS (THE LEVEL CONCERNED WITH THE STRUCTURE AND ORGANIZATION OF THE DISK BLOCKS MAKING UP EACH FILE) HAS REMAINED ESSENTIALLY UNCHANGED. THE RESULT IS THAT TODAY'S HIGH PERFORMANCE SYSTEMS ARE OFTEN SEVERELY LIMITED BY FILE SYSTEM THROUGHPUT. THIS RESEARCH AIMS TO INVESTIGATE A NEW KERNEL FILE SYSTEM ARCHITECTURE CALLED LAFS, WHICH THROUGH THE USE OF A NEW ADDRESSING PARADIGM, OFFERS HIGHER AND SCALABLE PERFORMANCE, HIGHER RELIABILITY, AND GREATER EASE OF USE THAN OFFERED IN CURRENT DESIGNS. IMPROVEMENTS ARE AVAILABLE FOR BOTH MULTI THREADED DIVERSE LOADS (E.G. FILE SERVERS, TIMESHARING, DBMS) AND FOR SINGLE STREAM APPLICATIONS (E.G. COMPUTATION INTENSIVE ANALYSIS, GRAPHICS). PHASE I WILL RESULT IN A FUNCTIONAL EMULATION OF THIS SYSTEM OPERATING ON A SINGLE DRIVE HOSTED IN A CONVENTIONAL WORKSTATION. PERFORMANCE UNDER VARIOUS LOADS WILL BE COMPARED TO THE STANDARD SYSTEM. THESE RESULTS AND THE HOST SYSTEM SW AND INTERFACES DEVELOPED WILL BE DIRECTLYUSED IN PHASE II/III FOR A FULL SCALE PROTOTYPE.