You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
We routinely read some very large things on almost every invocation (available_filter_functions, kallmodsyms, module.ranges etc). We are doing all this with the default stdio buffer size, which is 1K. We should at least be using PAGE_SIZE: using more than that is risky for /proc but for stuff read out of /lib/modules etc we should probably go all the way to 64K.
This is a bit fiddly since it means we need to manage the stdio buffer lifetimes ourselves, so it's worth trying on one big one in /proc first to see if it has any effect. But it might well be worthwhile in the upcoming performance push.
The text was updated successfully, but these errors were encountered:
Wait until we have a better idea of where dtrace spends most of its time. We want to do this in a larger context of performance optimization in general.
(Reported by Leah Neukirchen.)
We routinely read some very large things on almost every invocation (available_filter_functions, kallmodsyms, module.ranges etc). We are doing all this with the default stdio buffer size, which is 1K. We should at least be using PAGE_SIZE: using more than that is risky for /proc but for stuff read out of /lib/modules etc we should probably go all the way to 64K.
This is a bit fiddly since it means we need to manage the stdio buffer lifetimes ourselves, so it's worth trying on one big one in /proc first to see if it has any effect. But it might well be worthwhile in the upcoming performance push.
The text was updated successfully, but these errors were encountered: