On Oct 21, 2009, at 11:16 AM, Les Mikesell <lesmikesell at gmail.com> wrote: > Ross Walker wrote: >> On Oct 21, 2009, at 9:47 AM, Rainer Duffner <rainer at ultra-secure.de> >> wrote: >> >>> Chan Chung Hang Christopher schrieb: >>>> I suspect so. After all, it is just seen as a disk as far as md is >>>> concerned and it will do the same normal thing if you unplugged a >>>> single >>>> disk from the array. >>>> >>> >>> But the latency over the net is much higher. >>> Who knows if the kernel can handle this in all situations? >> >> I'm sure the kernel can handle the slowness, it's the cache >> consistency one has to be careful with in these setups. With so many >> caching devices in the chain, one must make sure the write and read >> cache is consistent throughout. > > Journaled file systems should take care of the consistency issues. I'm not talking data missing due to target/device failure, I'm talking about wrong data being returned from cache because the caches between the head server and backing store don't agree. No journal can help that, ZFS would help identify it, but not prevent, nor repair it as wrong data would keep coming back. -Ross