On Wed, Sep 05, 2007 at 02:27:30PM -0700, John Forte wrote: > Nicolas Williams wrote: > >I don't know enough about the libscf API, but if I understand correctly > >that it does not talk to svc.configd for pg updates until the > >transaction is committed, then you should be able to batch all your > >updates into one transaction (or a small number of them). That should > >go a lot faster than one transaction per-prop update. > > > Since we're providing a library and CLI that do not have batch > semantics, I cannot do that. If need be, for a "load configuration" > scenario, we'll provide special operations. But I doubt it will be > required from a customer perspective. As I said, I just want to ensure > that what I was seeing was expected.
Are these new or existing interfaces that you cannot modify to support batch operation? If existing, can't you add new interfaces that do support batch operation? Yes, what you see is expected. If the access time of your disks is, say, 10ms then you can expect scf_transaction_commit() to take no less than 20ms. I think 60ms is probably quite good for a freeware, embedded SQL DB like SQLite. Nico --
