[Gluster-devel] general performance translator recommendations

[Dan Parsons]

Regarding the various performance boosters like, readhead,  
writebehind, io-threads, io-cache, what are their general recommended  
usages, given systems with lots of extra RAM?

(1) Should every server volume, including ns, get readahead/ 
writebehind/iot/ioc?

(2) Should very client volume, including ns get readahead/writebehind/ 
iot/ioc?

(3) If I'm using unify, should I put io-threads on the volumes under  
it, or on top of it? It seems like I'd get better performance by using  
it under unify, but then I'll have a lot of extra threads going  
around, as opposed to using it on top, assuming I specify 8 threads  
per sub-volume.

Dan Parsons