Re: aborting IO requests

From: Henrik Nordstrom <hno@dont-contact.us>
Date: 03 Dec 2002 14:33:23 +0100

tis 2002-12-03 klockan 08.06 skrev Robert Collins:

> More specifically, what I meant was that the store object close routines
> appear to take different approaches in aufs and diskd. The end behaviour
> may be consistent.

Yes. diskd has the benefit of being able to fully serialize requests
using an abstract filehandle. What has been queued to the diskd daemon
is guaranteed to be executed in the same order, and you can queue writes
while the open/create is taking place.

asyncio does not have this property. asyncio operates based on the UNIX
filedescriptor and there is no guarantee of serialization of the request
queue.
 
For reads this is not a big problem as the sio is only closed if the
operation should be aborted.

For writes it is a big problem.

Regards
Henrik
Received on Tue Dec 03 2002 - 06:33:26 MST

This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:19:00 MST