Mailing-List: contact common-dev-help@hadoop.apache.org; run by ezmlm
Precedence: bulk
Date: Fri, 27 Jan 2017 21:26:24 +0000 (UTC)
From: "Daryn Sharp (JIRA)" <jira@apache.org>
To: common-dev@hadoop.apache.org
Message-ID: <JIRA.13038539.1485552343000.22010.1485552384503@Atlassian.JIRA>
In-Reply-To: <JIRA.13038539.1485552343000@Atlassian.JIRA>
References: <JIRA.13038539.1485552343000@Atlassian.JIRA> <JIRA.13038539.1485552343072@jira-lw-us.apache.org>
Subject: [jira] [Created] (HADOOP-14032) Reduce fair call queue priority
 inversion
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Transfer-Encoding: 7bit
archived-at: Fri, 27 Jan 2017 21:26:30 -0000

Daryn Sharp created HADOOP-14032:
------------------------------------

             Summary: Reduce fair call queue priority inversion
                 Key: HADOOP-14032
                 URL: https://issues.apache.org/jira/browse/HADOOP-14032
             Project: Hadoop Common
          Issue Type: Sub-task
          Components: ipc
    Affects Versions: 2.7.0
            Reporter: Daryn Sharp
            Assignee: Daryn Sharp


The fcq's round robin multiplexer actually rewards abusive users.  Queue consumers scan for a call from the roving multiplexer index to the lowest prio ring before wrapping around to the higher prio rings.

Let's take a fcq with 4 priority rings.  Multiplexer shares per index are 8, 4, 2, 1.  
All well behaved clients are operating in ring 0.  Bad client floods the server and drops to the lowest prio.  Unfortunately the service order gives 8 shares to the good clients, followed by 4+2+1=7 shares to the bad client.


--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

---------------------------------------------------------------------
To unsubscribe, e-mail: common-dev-unsubscribe@hadoop.apache.org
For additional commands, e-mail: common-dev-help@hadoop.apache.org