Accurate and Efficient Rplaying of File System Traces准确、高效的文件系统的轨迹回放

1、accurate and efficient replaying of file system tracesnikolai joukov, timothywong, and erez zadokstony brook university(fast 2005) usenix conference on file and storage technologiespresented by hsu hao chenoutlinenintroductionndesignnarchitecturenreproduce original timing problemnreplayfs tracenthre

2、ads and their schedulingnzero copying of datanfile system cachesnimplementationnevaluationnconclusionsintroductionntrace replaying is useful for file system benchmarking, stress-testing, debugging, and forensics.nfile system traces can be captured and replayed at different logical levels:nsystem cal

3、lsnvirtual file system (vfs)nnetwork level for network file systemsndevice driverdesignnarchitecture(1/2)ntracefs: replays traces captured using stackable file systemdesignnarchitecture(2/2)nreplayfs: vfs-level replayerdesignnreproduce original timing problemif the treplayer tuser then timeing and i

4、/o rate could not be reproduced correctlydesignnsystem-call replayers problem:nuser modenredundant data copying between user and kernel buffersnpage eviction is not completely controlled from the user levelnreplaying processes can be preempted by other tasksnsome kernel are not preemptive and have l

5、ong execution pathdesignnreplayfs trace(1/4)designnreplayfs trace(2/4)ntracefsna trace captured by a tracer is often portable, descriptive, and verbose to offer as much informationntrace compilernuser mode program for conversion and optimization of the traces raw tracesnsplits the raw tracefs trace

6、into three components:ncommandnresource allocation table (rat)nbufferdesignnreplayfs trace(3/4)designnreplayfs trace(4/4)memory buffers are accessed for reading only because the information read from the disk is discarded.designnthreads and their schedulingnreplayfs issues requests to the lower file

7、 system on behalf of different threadsnresource contention (disk head repositioning, locks, etc)nreplayfs reuses threads if possiblenpre-spinnincrease event precision (standard event timers 1ms)nclock threadncpu cycle countersdesignnzero copying of datanthere is no easy way a user-mode program can r

8、ead data but avoid copying it to user space.nuse kernel-mode benefitna data page that belongs to the trace file can be simply moved to the target le by just changing several pointersdesignnfile system cachesnreplaysfs supports three replaying modes for dealing with read operationsncurrent cache stat

9、enreplayfs calls all the captured buffer read operationsnoriginal cache statenreads are invoked on the page level only for the pages that were not found in the cache during tracing.nreads are not replayed at allimplementationnlinux kernel and now both tracefs and replayfs can be used on either 2.4 o

10、r 2.6 linux kernels. kernel moduleapplication programkernel moduleevaluationntest environmentn1.7ghz pentium 4 machine with 1gb of ramnsystem disk was a 30gb 7200 rpm ide formatted with ext3nthe machine had two maxtor atlas 15,000 rpm 18.4gb ultra320 scsi disks formatted with ext2nstoring the traces

11、 and the replayfs tracesevaluationnevaluation tools and workloadsnam-utils buildnbuilding am-utils is a cpu-intensive benchmarknpostmarknsimulates the operation of electronic mail serversnpreadnevaluate replayfss cpu time consumption.nit spawns two threads that concurrently read 1kb buffers of cache

12、d data using the pread system call.npread performed 100 million read operations.evaluationnmemory overheads56%70%45%evaluationntiming precision of replaying(1/2)time (seconds)time (seconds)number of operationsevaluationntiming precision of replaying(2/2)time (seconds)time (seconds)number of operatio

13、nsevaluationncpu time consumption32%61%user-level replayers cannot replay traces like pread at the same rate as the originalconclusionsntrace replaying offers a number of advantages for file system benchmarking, debugging, and forensicsnreplaying has three distinct benefits:ncapture and replay all file syste