On 10 Mar 2013, at 19:03, folkert wrote: > Are they mapped to "barriers"?
Yes > And can I now safely buffer all writes until the barrier is/FUA is > invoked and force them on the platters? Yes, that's what any linux block driver is permitted to do. And your SATA drive may do just that. > Unfortunately the client does not tell you if it will be using barriers > so that you can use this buffering. Any Linux block driver is permitted to do that buffering, so yes, the upper layer should expect it. FWIW if you use barrier support, you *don't* want to be using O_SYNC or the sync option as everything will be treated as flushed. What your server might do (I haven't read the code) that I think nbd-server doesn't currently do is cope properly with O_DIRECT which should speed things up by omitting a copy, if you are going to a real block device. That's non-trivial to fix in the current nbd due to the alignment requirements of O_DIRECT. -- Alex Bligh ------------------------------------------------------------------------------ Symantec Endpoint Protection 12 positioned as A LEADER in The Forrester Wave(TM): Endpoint Security, Q1 2013 and "remains a good choice" in the endpoint security space. For insight on selecting the right partner to tackle endpoint security challenges, access the full report. http://p.sf.net/sfu/symantec-dev2dev _______________________________________________ Nbd-general mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/nbd-general
