jms@dsl.cis.upenn.edu (Jonathan M. Smith) (01/17/90)
>Message-ID: <21156@pasteur.Berkeley.EDU> >Approved: comp-os-research@jupiter.ucsc.edu > >I am working on a file server project at HP-Lab's. We want to develop >a probabilistic model of the low-level I/O use (disk-block request >level) of existing file servers. A workload synthesiser will be based >on the model. The model will be verified against trace data taken >here at the lab's. > >Can anyone point me to past/current efforts to model low-level file >I/O? Send info to stata%cello@hplabs.hp.com. > The following paper might be useful. While it focuses on response time, it has a good trace methodology and graphs many relations derived from file system traces that we got by adding a circular trace buffer to our Sys V RA81 device driver. The measured data was voluminous, but it may be saved somewhere. I can check. %T Disk Response Time Measurements %A Thomas D. Johnson %A Jonathan M. Smith %A Eric S. Wilson %B Proceedings, Winter 1987 USENIX Technical Conference %C Washington, DC %P 147-162 %D January, 1987 -Jonathan