CONSTRUCTING A FLEXIBLE INTERNET-SCALE TIME-SHARING SYSTEM USING DETERMINISTIC CHECKPOINTING

摘要

Distributed systems and clustering have both grown intornmainstream phenomena, now used throughout academiarnand industry. Despite advances in a wide variety ofrninterconnects, including home Internet connections atrnspeeds and latencies only available in a data center severalrnyears ago, the scheduling and use of these systems is stillrnfirmly rooted in the era of batch processing. Batchrnprocessing provides a basic scheduler and simplernprogramming model for many types of computation, butrnlacks the flexibility and efficient resource utilization ofrneven the most rudimentary of time-sharing systems. Withrna large number of use cases now constrained by thesernlimitations, time-sharing concepts must once again comernto the rescue. Based on experience from therndistributed.net, Folding@home, and Storage@homernsystems, it can be demonstrated that the Internet hasrnadvanced and can now meet the higher requirements forrntime-sharing. This paper will explore those requirementsrnand the potential benefits of moving beyond batchrnprocessing and into time-sharing for Internet-scalerncomputations, and lay out a method for deterministicrncheckpointing required to implement such a system.