On Mon, 2013-04-15 at 07:47 +0000, Counihan, Tom wrote: > > Your request is about making Noop or Deadline to be the default. I am > > generally fine with it, but I'd like to make it more consciously. > > Makes perfect sense. What was it about CFQ that made it default originally?
I do not know. I think it is just the oldest I/O scheduler and it is traditionally used as the default one. It allows setting I/O priorities (see 'man ionice') which is very helpful. BTW, I've just submitted the kernel change which enables the deadline scheduler. > > I'd probably put it this way: the I/O scheduler which allows for the fastest > > system boot should be made the default one in Tizen IVI. Do you agree? > > I would also add wear levelling and file system consistency as a criteria. This would be a good criteria if you could measure it, which I doubt very much... > What I mean is under the scenario of cranking, where power flux will > be drastic and result in an immediate reboot. I think I/O scheduler is the least important thing here. This is, in my thinking, about: 1. Dow the SW handles their data. Does it keep the crucial data always synced? Does it modify the crucial data in an atomic manner? Does it crash if a not-so-crucial file is truncated or corrupted or you are resilient and re-construct the file on-the-fly? Were power cuts tested? How extensively. 2. Are the barriers enabled so 'fsync()' really makes the data to the media, not just to the HW cache? 3. Is FTL on the SSD/eMMC is good enough so that in case of power loss it does not lose the data which has already been "synced" by the SW? And of course in case of power cuts some data is _always_ lost, e.g., the non-synced one, and some files can always get inconsistent e.g., when you modify them in-place instead of doing atomic copy-modify-renamle. This is really not about I/O scheduler. This is a lot more complex topic, I think. > On the topic of Fastboot - I need to take a look, but is sreadahead employed? Unfortunately, I do not know much about this project, but I believe we do not use this. -- Best Regards, Artem Bityutskiy _______________________________________________ IVI mailing list [email protected] https://lists.tizen.org/listinfo/ivi
