On 10/18/13 4:36 PM, Prakash Surya wrote: > Other than what Matt brought up, it would be nice to have a way to > override the default values without recompiling. But we can add that to > the Linux port as needed. I don't really know how you guys support that > sort of thing.
On the Linux port I think the guys add sysctl's to change these values. In Illumos we simply use mdb and modify the in-core value at runtime (that's essentially what /etc/system does). > Also, it _might_ be useful to move statements like: > > Using the default settings these values translate to ~1 MB of hash > tables for each 1 GB of physical memory. > > closer to the variable declarations just to try and ensure the comment > is updated if the default values are ever changed. But maybe I'm being > overly pedantic. Then it would be further away from the general theory statement. I don't know, I have no strong preference either way. > Do you have any before/after values for the max chain length from your > testing? It should be easy(?) to grab it from the arc stats and would be > nice to have numbers to justify the change. Yeah, just grab see the hash_ arcstats on your systems (kstat -n arcstats | grep hash_). > Looks good, overall. Thanks. Cool, thanks! Cheers, -- Saso _______________________________________________ developer mailing list developer@open-zfs.org http://lists.open-zfs.org/mailman/listinfo/developer
