I've finally pulled all of the histogram patches together and updated the original
patch series with a set of re-based changes.
Thanks to Paul Luse for his help in making this happen.
There is more work to be done, but I wanted to put these patches upstream, even though
they are incomplete.
(Just in case someone else would like to help move this work forward. Feel free to do
that, if you are inclined).
1. The histogram_ut tests fail because of the following error.
+ valgrind --leak-check=full --error-exitcode=2
==40355== ==40355== Conditional jump or move depends on uninitialised value(s)
However, I see this failure on my internal testbed even on the master branch.
So this might have nothing to do with this patch. (This is where I can use some help from
1. We need to address some of Jim's comments, which I’ve forwarded in this email
From: Jim Harris <james.r.harris(a)intel.com>
Date: Thursday, February 1, 2018 at 10:07 AM
To: John Meneghini <John.Meneghini(a)netapp.com>, "Luse, Paul E"
<paul.e.luse(a)intel.com>, "Chandrashekar, Girish"
<Girish.Chandrashekar(a)netapp.com>, "Verkamp, Daniel"
Subject: Re: SPDK histogram patch
Hi John and Girish,
I still have a few questions on this patch. I’ll put these in GerritHub as well.
1. I’m concerned about the accuracy of the data on a multi-threaded NVMf target app.
Currently I don’t see any synchronization on the global histogram data structures. I
think this needs histogram data structures per thread, and then iterators that aggregate
the data when a user requests the updated histogram data.
2. It would be great if this was more dynamic, so that it does not have to be compiled
in with a separate flag.
3. Have you measured the time it takes to iterate over a fairly well populated
histogram and send its data over JSON-RPC? Just want to make sure that using JSON-RPC
for the raw histogram data is not going to result in too much overhead. I’ve wondered if
we should just be sending the raw data using some kind of encoding, and let the client app
work out the ranges and counts instead.
From: "Luse, Paul E" <paul.e.luse(a)intel.com>
Date: Friday, January 26, 2018 at 12:27 PM
To: John Meneghini <John.Meneghini(a)netapp.com>, "Chandrashekar, Girish"
Subject: SPDK histogram patch
I’ve got this all up to date again following some unrelated rpc changes that required a
CI is still WIP but here it is: https://review.gerrithub.io/#/c/363114/
And here is the old one that can be abandoned now or when you get a chance: