Return-Path: X-Original-To: apmail-cassandra-commits-archive@www.apache.org Delivered-To: apmail-cassandra-commits-archive@www.apache.org Received: from mail.apache.org (hermes.apache.org [140.211.11.3]) by minotaur.apache.org (Postfix) with SMTP id B1B9E17422 for ; Wed, 24 Jun 2015 21:54:05 +0000 (UTC) Received: (qmail 30212 invoked by uid 500); 24 Jun 2015 21:54:05 -0000 Delivered-To: apmail-cassandra-commits-archive@cassandra.apache.org Received: (qmail 30166 invoked by uid 500); 24 Jun 2015 21:54:05 -0000 Mailing-List: contact commits-help@cassandra.apache.org; run by ezmlm Precedence: bulk List-Help: List-Unsubscribe: List-Post: List-Id: Reply-To: dev@cassandra.apache.org Delivered-To: mailing list commits@cassandra.apache.org Received: (qmail 29905 invoked by uid 99); 24 Jun 2015 21:54:05 -0000 Received: from arcas.apache.org (HELO arcas.apache.org) (140.211.11.28) by apache.org (qpsmtpd/0.29) with ESMTP; Wed, 24 Jun 2015 21:54:05 +0000 Date: Wed, 24 Jun 2015 21:54:05 +0000 (UTC) From: "Joshua McKenzie (JIRA)" To: commits@cassandra.apache.org Message-ID: In-Reply-To: References: Subject: [jira] [Created] (CASSANDRA-9648) Warn if power profile is not High Performance on Windows MIME-Version: 1.0 Content-Type: text/plain; charset=utf-8 Content-Transfer-Encoding: 7bit X-JIRA-FingerPrint: 30527f35849b9dde25b450d4833f0394 Joshua McKenzie created CASSANDRA-9648: ------------------------------------------ Summary: Warn if power profile is not High Performance on Windows Key: CASSANDRA-9648 URL: https://issues.apache.org/jira/browse/CASSANDRA-9648 Project: Cassandra Issue Type: Improvement Reporter: Joshua McKenzie Assignee: Joshua McKenzie Priority: Minor Fix For: 2.2.x Windows' power profiles have a pretty marked impact on application performance and the CPU frequency throttling is fairly aggressive even in balanced mode. As we have a large number of threads with varying work rather than a single busy thread-per-core, the scheduler on Windows sees enough downtime to constantly struggle w/our user-space operations and the frequency on the system will jump up and down even when fully saturated from a stress. I've done some benchmarking of the "Balanced" vs. "High Performance" power profiles - [link to performance numbers|https://docs.google.com/spreadsheets/d/1YS8VtdZAgyec-mcnSgtNhQH9LiHstOaiMtlppvEIIM8/edit#gid=0]. Note: reads are not saturating the box (or even impacting resources at all really) as the CPU's on both stress and node are sitting around 4% usage. Still have something to figure out there on 2.2. We have a few ways we can approach this - for the 1st (warn), here's a branch with warning during startup if non-High Performance power profile detected: [here|https://github.com/apache/cassandra/compare/trunk...josh-mckenzie:check_power_plan]. Alternatively we could get more aggressive and actually attempt a powercfg /s to the GUID of the High Performance power profile or refuse to start Cassandra if we're not in the performance profile. I also briefly pursued using Sigar to query this information however the documentation for the library is no longer available (or at least I couldn't find it). -- This message was sent by Atlassian JIRA (v6.3.4#6332)