From: "Eric Blake" <eblake(a)redhat.com>
To: "Peter Krempa" <pkrempa(a)redhat.com>
Cc: libvir-list(a)redhat.com, fromani(a)redhat.com
Sent: Monday, June 15, 2015 6:21:13 PM
Subject: Re: [libvirt] [RFC PATCHv2 1/8] threshold: new
API virDomainBlockSetWriteThreshold
On 06/15/2015 07:19 AM, Peter Krempa wrote:
> On Fri, Jun 12, 2015 at 13:29:25 -0600, Eric Blake wrote:
>> qemu 2.3 added a new QMP command block-set-write-threshold,
>> which allows callers to get an interrupt when a file hits a
>> write threshold, rather than the current approach of repeatedly
>> polling for file allocation. This patch prepares the API for
>> callers to register to receive the event, as well as a way
>> to query the threshold via virDomainListGetStats().
>>
>> +
>> +typedef enum {
>> + /* threshold is thousandth of a percentage (0 to 100000) relative to
>
> You managed to choose a unusual unit. Commonly used ones are 1/1000 and
> 1/1 000 000. Financial world also uses 1/10 000. Your unit of 1/100 000
> is not among:
>
>
https://en.wikipedia.org/wiki/Parts-per_notation#Parts-per_expressions
>
> I'd again suggest to use 1/1 000 000. Or if you want to be uber preciese
> you might choose 1/(2^64 - 1).
Francesco, what precision would you like? Parts per million seems okay
to me, if we want an order of magnitude closer; and I don't think we
need anything beyond that. Or if parts per thousand is sufficient, that
leads to smaller numbers on input. But it's pretty trivial for me to
adjust the code to a different base, for whatever people would like.
We (in oVirt) use very coarse thresholds.
For our current needs, I believe even parts per thousand is sufficient.
Trying to be a bit forward thinking, I believe parts per million is perfectly fine.
Bests,
--
Francesco Romani
RedHat Engineering Virtualization R & D
Phone: 8261328
IRC: fromani