On 9/19/18 3:06 PM, Dennis Zhou wrote: > On Wed, Sep 19, 2018 at 01:59:53PM -0600, Jens Axboe wrote: >> On 9/19/18 1:56 PM, Dennis Zhou wrote: >>> On Wed, Sep 19, 2018 at 12:49:15PM -0600, Jens Axboe wrote: >>>> On 9/19/18 12:25 PM, Dennis Zhou wrote: >>>>> As we explore stacking traces, it is nice to be able to scale a trace to >>>>> understand how the traces end up interacting. >>>>> >>>>> This patch adds scaling by letting the user pass in percentages to scale >>>>> a trace by. When passed '--merge_blktrace_scalars="100", the trace is >>>>> ran at 100% speed. If passed 50%, this will halve the trace timestamps. >>>>> The new option takes in a comma separated list that index-wise pairs >>>>> with the passed files in "--read_iolog". >>>> >>>> How is this different than replay_time_scale? >>>> >>> >>> I think merge_blktrace_scalars is a trace building parameter whereas >>> replay_time scale is a runtime parameter. merge_blktrace_scalars is an >>> index-paired list with the logs passed to --read_iolog allowing for each >>> trace to be independently scaled. replay_time_scale happens at runtime >>> and scales the entire trace uniformly. And because replay_time_scale >>> happens at runtime, I'm not sure repurposing the numbers would be super >>> intuitive. >> >> Not sure I see the difference, if you just allow replay_time_scale to >> take multiple values (one for each trace)? >> > > I'm imagining if I reused replay_time_scale, I could use those numbers > for merging, but then I'd have to reset it so that it doesn't affect the > trace a second time during runtime. I feel like this gets a little weird > as we're saying if you are merging, replay_time_scale gets applied > during the merge, otherwise it gets applied during runtime. Reusing also > makes it become a many (merging) to one (runtime) parameter change too > as merging works on multiple files (main thread) while runtime runs a > single file (worker thread). > > Additionally, I don't think it's unreasonable to want to store the > merged trace in realtime and then run the merged trace at a different > pace, which would require the merge time and runtime knobs to be > different. I guess that makes sense, especially as a sub-option to merging. Just ensure it's all properly documented :-) -- Jens Axboe
