Abstract Attribute-Based Prediction of File Properties(16)

We present evidence that attributes that are known to the file system when a file is created, such as its name, permission mode, and owner, are often strongly related to future properties of the file such as its ultimate size, lifespan, and access pattern.

6.1Benchmarking Attribute-Based Systems

One of the dif?culties of measuring the utility of attribute-based hints in the context of real?le systems is ?nding a suitable benchmark.Synthetic workload gener-ators typically create?les in a predictable and unrealistic manner–they make little or no attempt to use realis-tic?le names or mimic the diverse behaviors of differ-ent users.If we train our models on data gathered when these benchmarks are running then our predictions will probably be unrealistically accurate,but if we train on a workload that does not include the benchmarks,then our predictions for the?les created by the benchmark will be uncharacteristically bad.

Our solution to this problem is to construct a benchmark directly from traces of the target work-load,thereby ensuring that the associations between?le names,modes,and uids during the trace will resemble those present in the actual workload.This leads imme-diately to a new problem–in order to replay the traces, we need a real?le system on which to play them.The usual solution to this problem is to recreate the traced ?le system from a snapshot of its metadata taken at a known time,and then begin replaying from that time [28].This method works well when snapshots are avail-able,and when a suitable device is available on which to reconstruct.Unfortunately we have neither–there are no publicly-available snapshots of the systems from which the traces were taken,and even if there were,reconstruct-ing them would require at least500GB of disk space and many hours of set-up time per test.

搜索“diyifanwen.net”或“第一范文网”即可找到本站免费阅读全部范文。收藏本站方便下次阅读，第一范文网，提供最新工程科技Abstract Attribute-Based Prediction of File Properties(16)全文阅读和word下载服务。