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;
MIME-Version: 1.0
In-Reply-To: <153666073461.19117.1958730317836145457.stgit@localhost.localdomain>
References: <153666041612.19117.14667042009014596105.stgit@localhost.localdomain>
 <153666073461.19117.1958730317836145457.stgit@localhost.localdomain>
From:   Miklos Szeredi <miklos@szeredi.hu>
Date:   Tue, 25 Sep 2018 11:08:55 +0200
Message-ID: <CAJfpegt3ckxAUAdabXKwxgFYdP9=wVSis+G3pztWgqzXVGNFSA@mail.gmail.com>
Subject: Re: [PATCH 3/3] fuse: Use hash table to link processing request
To:     Kirill Tkhai <ktkhai@virtuozzo.com>
Cc:     kuznet@virtuozzo.com, linux-fsdevel@vger.kernel.org,
        linux-kernel@vger.kernel.org
Content-Type: text/plain; charset="UTF-8"
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Tue, Sep 11, 2018 at 12:12 PM, Kirill Tkhai <ktkhai@virtuozzo.com> wrote:
> We noticed the performance bottle neck in FUSE running our
> Virtuozzo storage over rdma. On some types of workload
> we observe 20% of times pent in request_find() in profiler.
> This function is iterating over long requests list, and it
> scales bad.
>
> The patch introduces hash table to reduce the number
> of iterations, we do in this function. Hash generating
> algorithm is taken from hash_add() function, while
> 512 lines table is used to store pending requests.
> This fixes problem and improves the performance.

Pushed to fuse.git#for-next with a number of small changes.   E.g. I
reduced the number of cachlines to 256 to make the hashtable size just
4k.   Was there a scientific reason for choosing 512 as the optimal
number of cache lines?

Thanks,
Miklos