From: BlaisorBlade <bla...@ya...> - 2004-06-02 18:00:52
|
Alle 01:23, gioved=EC 27 maggio 2004, roland ha scritto: > for at least two of the uml-pid`s you should see minor and major pagefaul= ts > (minflt/majflt) - i would say there is a releationship between those > pagefaults and the ping delays. (i had turned off swap completely - so my > host was not able to swap at all!) That is not very good - if the latency problem comes from UML being swapped= =20 out, why do you disable swapping? Actually there is some "swapping", in a=20 certain sense, so you could observe the major page faults (major means=20 "requiring I/O", minor means "no IO": this can happen for COW pages, to mar= k=20 as dirty a clean page). When swapping is disabled, the pages belonging to a memory mapping, that wa= s=20 not modified since they were read from the disk, can be swapped out. I.e. the executable pages, actually, are created by mmap'ing the executable= ;=20 since they are not modified, they can be removed from memory. Also the data pages which come from the executable can be treated the same= =20 way, if they are still "clean" (i.e. not modified). [ I actually checked the code works this way, i.e. if you try to read a glo= bal=20 var, even if it's mmaped with PROT_WRITE, in the page-tables it is marked a= s=20 read-only; when you try to write it, it will be marked as read-write and th= e=20 page will become dirty. ] =2D-=20 Paolo Giarrusso, aka Blaisorblade Linux registered user n. 292729 |