Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
From:   Dave Chiluk <chiluk+linux@indeed.com>
To:     Ben Segall <bsegall@google.com>, Phil Auld <pauld@redhat.com>,
        Peter Oskolkov <posk@posk.io>,
        Peter Zijlstra <peterz@infradead.org>,
        Ingo Molnar <mingo@redhat.com>, cgroups@vger.kernel.org,
        linux-kernel@vger.kernel.org, Brendan Gregg <bgregg@netflix.com>,
        Kyle Anderson <kwa@yelp.com>,
        Gabriel Munos <gmunoz@netflix.com>,
        John Hammond <jhammond@indeed.com>,
        Cong Wang <xiyou.wangcong@gmail.com>
Subject: 
Date:   Mon, 24 Jun 2019 10:50:03 -0500
Message-Id: <1561391404-14450-1-git-send-email-chiluk+linux@indeed.com>
In-Reply-To: <1558121424-2914-1-git-send-email-chiluk+linux@indeed.com>
References: <1558121424-2914-1-git-send-email-chiluk+linux@indeed.com>
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk


Changelog v4
 - Rewrote patch to return all quota when cfs_b has very litte.
 - Removed documentation changes, as bursting is no longer possible with this
   new solution.

After the suggestion from Ben Segall to set min_cfs_rq_runtime=0, I came up
this in an attempt to balance the desire leave runtime on the per-cpu queues
with the desire to use this quota on other per-cpu rq.

Basically we now check the cfs_b on each return, and decide if all the remaining
time should be returned or to leave min_cfs_rq_runtime on the per-cpu queue
based on how much time is remaining on the cfs_b.  As a result this mostly
gives us the benefits of both worlds.