Re: [Qemu-devel] [PATCH V11 1/5] throttle: Add a new throttling API implementing continuous leaky bucket.

[Paolo Bonzini]

Il 02/09/2013 12:16, Benoît Canet ha scritto:
>> If op_size is non-zero, iops limits are merely a fixed proportion of bps
>> > limits, which means the lower set of the two is applied and the higher 
>> > skipped.
>> > I understand the amazon uses op_size like accounting for big IO requests, 
>> > but
>> > we don't do it condionally on io size or anything here, so that once user 
>> > sets
>> > op_size, it simply kicks either bps_{,rd,wr} or iops_{,rd,wr} out the 
>> > game, is
>> > that true?
> It will combine with iops_{,rd,wr}.

I think what Fam asked is this:

what is the difference between bps=524288 and iops=1/op_size=524288?

And I think the answer should be that bs=524288 will allow many requests
up to 524288 bytes, while with iops all requests should be rounded up to
the op_size.

However, I think this is not what the code is doing right now, isn't it?

Paolo