Opened 19 months ago

#3561 new bug

massively parallel "recheck" operations perform far below system optimum

Reported by: brainchild Owned by:
Priority: minor Milestone: needs verified
Component: Core Version: 2.0.5
Keywords: recheck, batch, parallel Cc:

Description

Presently, when a large number of torrents are simultaneously submitted for "recheck", the application will dispatch the operation immediately for a very large number of them (I am observing more than 60). I suggest that such a large number of simultaneous operations will create I/O bottlenecks on most consumer hardware, and that 4, 3, or even 2 simultaneous operations is appropriate for optimal throughput. For a relatively inexpensive checksum algorithm, I/O more than CPU is likely to limit throughput, in which case, each additional file dispatched for concurrent processing, above a small number, is likely to degrade performance.

As a point of reference, I have tested on a consumer storage device that is normally capable of read speeds exceeding 200 Mb/s from its redundant magnetic array, but for the batch operations (performed locally) generally has been limited to below one tenth that rate, with negligible CPU usage and extremely high system load.

For solid-state storage, results may be different, but are unlikely to be helped by such an enormous number of concurrent operations.

Change History (0)

Note: See TracTickets for help on using tickets.