[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [Qemu-block] [Qemu-devel] [PATCH v2 1/2] block/accounting: introduce
From: |
Vladimir Sementsov-Ogievskiy |
Subject: |
Re: [Qemu-block] [Qemu-devel] [PATCH v2 1/2] block/accounting: introduce latency histogram |
Date: |
Thu, 8 Mar 2018 21:58:10 +0300 |
User-agent: |
Mozilla/5.0 (X11; Linux x86_64; rv:52.0) Gecko/20100101 Thunderbird/52.6.0 |
08.03.2018 21:21, Vladimir Sementsov-Ogievskiy wrote:
08.03.2018 21:14, Vladimir Sementsov-Ogievskiy wrote:
08.03.2018 20:31, Eric Blake wrote:
On 03/06/2018 09:32 AM, Stefan Hajnoczi wrote:
On Wed, Feb 07, 2018 at 03:50:36PM +0300, Vladimir
Sementsov-Ogievskiy wrote:
Introduce latency histogram statics for block devices.
For each accounted operation type latency region [0, +inf) is
divided into subregions by several points. Then, calculate
hits for each subregion.
Signed-off-by: Vladimir Sementsov-Ogievskiy
<address@hidden>
---
According to Wikipedia and Mathworld, "intervals" and "bins" are
commonly used terms:
https://en.wikipedia.org/wiki/Histogram
http://mathworld.wolfram.com/Histogram.html
I suggest:
typedef struct {
/* The following histogram is represented like this:
*
* 5| *
* 4| *
* 3| * *
* 2| * * *
* 1| * * * *
* +------------------
* 10 50 100
*
* BlockLatencyHistogram histogram = {
* .nbins = 4,
* .intervals = {10, 50, 100},
* .bins = {3, 1, 5, 2},
* };
The name 'intervals' is still slightly ambiguous: does it hold the
boundary point (0-10 for 10 slots, 10-50 for 40 slots, 50-100, for
50 slots, then 100-INF) or is it the interval size of each slot
(first bin is 10 slots for 0-10, next bin is 50 slots wide so 10-60,
next bin is 100 slots wide so 60-160, everything else is 160-INF).
But the ascii-art diagram plus the text is sufficient to resolve the
intent if you keep that name (I don't have a suggestion for a better
name).
Hm. these numbers are actually boundary points of histogram
intervals, not intervals itself. And, wiki says "The bins are usually
specified as consecutive, non-overlapping intervals of a variable.",
so, intervals are bins.
So, what about:
1. interval_boundaries
2. bin_boundaries
3. boundaries
(and same names with s/_/-/ for qapi)
Also, now I doubt, is it a good idea to share same bin boundaries for
each io type.
so, for qmp command, what about:
boundaries - optional, default boundaries for all io operations
boundaries-read - boundaries for read
boundaries-write - boundaries for write
...
so, call without any boundaries: drop _all_ histograms
call with only specific boundaries-*: set or reset _only_ corresponding
specific histograms
call with only boundaries parameter: set or reset _all_ histograms
call with boundaries parameter and some of (or all, but it is not
useful) specific boundaries-*: set or reset _all_ histograms, some to
default and some to specific.
--
Best regards,
Vladimir