Hi Steve, I occurs when only using the SQL output, i.e. old text only replaced with new SQL only. Adding text output to the latter unfortunately slows things down even more due to the additional formatting/formula evaluation etc.
In a typical use-case the text formatting could be done after the simulation is done, so it would hopefully not be a problem. Dumping with high frequency with SQL and text would cause issues with performance. It's not perfect at this point, but by having a broader audience we can hopefully fix these issues much faster. Andreas On 23/01/2013 16:15, "Steve Reinhardt" <[email protected]> wrote: >Hi Andreas, > >Sorry I have not had time to look at your patches, but I fully support >what >you're doing in principle. > >One clarification about the slowdown: does this occur even when you use >text output, or does it occur only when you use SQL? That is, will users >who stick with text output be affected, or are you merely lamenting that >SQL is slower than text output when that shouldn't necessarily be the >case? > If it's the latter, I'm absolutely fine with it; if it's the former, then >I'm a bit more concerned. > >Thanks, > >Steve > > >On Wed, Jan 23, 2013 at 7:10 AM, Andreas Hansson ><[email protected]>wrote: > >> Hi all, >> >> The short version: >> >> A week ago I posted a number of review requests that build on top of >> Nate's initial work to move the stats from C++ to Python. The reviews, >> request 1636 to 1647, has not gotten any comments so far, and I would >>like >> to know if people are generally happy with these patches being pushed. >>Are >> you happy to move to a Python-based stats system with SQL output even >> though it currently involves a performance hit? >> >> The longer version: >> >> The main benefit of the new stats system is the flexibility in the stats >> processing, with the SQL output being the first example of this. The old >> text output is still available as an option, but for any post processing >> the SQL output removes the need to create complex regular expressions >>and >> parse strings back into numbers/vectors/distributions etc. >> >> There is already an API available for querying the database, and our >>hope >> is that this will be expanded upon by the users going forward. There is >> also a possibility to add functionality that would allow dumping stats >>for >> only certain components etc. >> >> Lots of exciting new features and possibilities in other words, but >>also a >> slight hit in performance. Although it has already been optimised >>compared >> to the initial implementation, it is still more costly than the old text >> output. For a detailed simulation, with dumps every 100k instructions, >>the >> simulation time goes from 11 s to 19 s. Our hope is to get this pushed >>out >> to get a broader audience helping to improve both the speed and the >> functionality. >> >> Thanks, >> >> Andreas >> >> >> >> -- IMPORTANT NOTICE: The contents of this email and any attachments are >> confidential and may also be privileged. If you are not the intended >> recipient, please notify the sender immediately and do not disclose the >> contents to any other person, use it for any purpose, or store or copy >>the >> information in any medium. Thank you. >> >> _______________________________________________ >> gem5-dev mailing list >> [email protected] >> http://m5sim.org/mailman/listinfo/gem5-dev >> >_______________________________________________ >gem5-dev mailing list >[email protected] >http://m5sim.org/mailman/listinfo/gem5-dev > -- IMPORTANT NOTICE: The contents of this email and any attachments are confidential and may also be privileged. If you are not the intended recipient, please notify the sender immediately and do not disclose the contents to any other person, use it for any purpose, or store or copy the information in any medium. Thank you. _______________________________________________ gem5-dev mailing list [email protected] http://m5sim.org/mailman/listinfo/gem5-dev
