> Situation: > In the case that the aoe target backing storage has errors on a > section of, for example, a raid5 array, the vblade hangs for a period > of time and makes the aoe initiator wait or rexmit other concurrent IO > requests until vblade gets past the slow/bad section of the raid5 array. > > Theoretical Fix: > Could adding in a worker thread pool alleviate some of the "blocking" > factor since vblade is single threaded and essentially runs > sequentially? Where would locking have to be placed (if at all) and > could this cause corruption?
Possibly. It depends on how i/o gets mapped through to the disks. Theoretically you could access other disks in the raid while one is stuck trying to retrieve/remap a bad sector. The raid software would have to be programmed to permit it. There shouldn't be any ordering issues processing i/o. That's taken care of by the client. If the aoe initiator driver is given the i/o, it's fair game to complete it out of order. The user of the AoE block device (on the client end) would use i/o barriers if it needs to guarantee ordering of some kind. Hasn't someone made a multithreaded linux aoe target? Seems like I recall one. Sam ------------------------------------------------------------------------------ Join us December 9, 2009 for the Red Hat Virtual Experience, a free event focused on virtualization and cloud computing. Attend in-depth sessions from your desk. Your couch. Anywhere. http://p.sf.net/sfu/redhat-sfdev2dev _______________________________________________ Aoetools-discuss mailing list Aoetools-discuss@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/aoetools-discuss
