Return-Path: X-Original-To: apmail-horn-dev-archive@minotaur.apache.org Delivered-To: apmail-horn-dev-archive@minotaur.apache.org Received: from mail.apache.org (hermes.apache.org [140.211.11.3]) by minotaur.apache.org (Postfix) with SMTP id 4D24518E91 for ; Mon, 9 Nov 2015 03:27:17 +0000 (UTC) Received: (qmail 96683 invoked by uid 500); 9 Nov 2015 03:27:17 -0000 Delivered-To: apmail-horn-dev-archive@horn.apache.org Received: (qmail 96639 invoked by uid 500); 9 Nov 2015 03:27:17 -0000 Mailing-List: contact dev-help@horn.incubator.apache.org; run by ezmlm Precedence: bulk List-Help: List-Unsubscribe: List-Post: List-Id: Reply-To: dev@horn.incubator.apache.org Delivered-To: mailing list dev@horn.incubator.apache.org Received: (qmail 96614 invoked by uid 99); 9 Nov 2015 03:27:17 -0000 Received: from Unknown (HELO spamd1-us-west.apache.org) (209.188.14.142) by apache.org (qpsmtpd/0.29) with ESMTP; Mon, 09 Nov 2015 03:27:17 +0000 Received: from localhost (localhost [127.0.0.1]) by spamd1-us-west.apache.org (ASF Mail Server at spamd1-us-west.apache.org) with ESMTP id B19A7C1F58 for ; Mon, 9 Nov 2015 03:27:16 +0000 (UTC) X-Virus-Scanned: Debian amavisd-new at spamd1-us-west.apache.org X-Spam-Flag: NO X-Spam-Score: 1.77 X-Spam-Level: * X-Spam-Status: No, score=1.77 tagged_above=-999 required=6.31 tests=[KAM_ASCII_DIVIDERS=0.8, KAM_LAZY_DOMAIN_SECURITY=1, RCVD_IN_MSPIKE_H3=-0.01, RCVD_IN_MSPIKE_WL=-0.01, T_RP_MATCHES_RCVD=-0.01] autolearn=disabled Received: from mx1-us-east.apache.org ([10.40.0.8]) by localhost (spamd1-us-west.apache.org [10.40.0.7]) (amavisd-new, port 10024) with ESMTP id s73k3pPJ995D for ; Mon, 9 Nov 2015 03:27:16 +0000 (UTC) Received: from mail.apache.org (hermes.apache.org [140.211.11.3]) by mx1-us-east.apache.org (ASF Mail Server at mx1-us-east.apache.org) with SMTP id 6DC2F439DD for ; Mon, 9 Nov 2015 03:27:11 +0000 (UTC) Received: (qmail 96361 invoked by uid 99); 9 Nov 2015 03:27:11 -0000 Received: from arcas.apache.org (HELO arcas) (140.211.11.28) by apache.org (qpsmtpd/0.29) with ESMTP; Mon, 09 Nov 2015 03:27:11 +0000 Received: from arcas.apache.org (localhost [127.0.0.1]) by arcas (Postfix) with ESMTP id E51902C14F8 for ; Mon, 9 Nov 2015 03:27:10 +0000 (UTC) Date: Mon, 9 Nov 2015 03:27:10 +0000 (UTC) From: "Edward J. Yoon (JIRA)" To: dev@horn.incubator.apache.org Message-ID: In-Reply-To: References: Subject: [jira] [Updated] (HORN-8) Implementation of Parameter Server MIME-Version: 1.0 Content-Type: text/plain; charset=utf-8 Content-Transfer-Encoding: 7bit X-JIRA-FingerPrint: 30527f35849b9dde25b450d4833f0394 [ https://issues.apache.org/jira/browse/HORN-8?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ] Edward J. Yoon updated HORN-8: ------------------------------ Description: The current implementation works in synchronous way like below (SmallLayeredNeuralNetworkTrainer.java 101 lines): {code} task0 task1 task2 compute updates locally -------------------------------- sends updates to master task -------------------------------- merge updates and broadcast it to every tasks compute updates locally -------------------------------- sends updates to master task -------------------------------- merge updates and broadcast it to every tasks ... (Loop until onvergence) {code} By separating the master, we can support asynchronous parallel SGD. My idea is just using of a task0 (BSPTask) as a server daemon. In this issue ticket, single master is enough at this moment. {code} task0 | task1 .... taskN | | | compute updates locally | Receive |<------ push updates to master task Update1 | +------> fetch updates | | | Receive |<------------------------------------ .. Update2 | +------------------------------------> .. | | {code} was: The current implementation works in synchronous way like below: {code} task0 task1 task2 compute updates locally -------------------------------- sends updates to master task -------------------------------- merge updates and broadcast it to every tasks compute updates locally -------------------------------- sends updates to master task -------------------------------- merge updates and broadcast it to every tasks ... (Loop until onvergence) {code} By separating the master, we can support asynchronous parallel SGD. My idea is just using of a task0 (BSPTask) as a server daemon. In this issue ticket, single master is enough at this moment. {code} task0 | task1 .... taskN | | | compute updates locally | Receive |<------ push updates to master task Update1 | +------> fetch updates | | | Receive |<------------------------------------ .. Update2 | +------------------------------------> .. | | {code} > Implementation of Parameter Server > ---------------------------------- > > Key: HORN-8 > URL: https://issues.apache.org/jira/browse/HORN-8 > Project: Apache Horn > Issue Type: Improvement > Reporter: Edward J. Yoon > > The current implementation works in synchronous way like below (SmallLayeredNeuralNetworkTrainer.java 101 lines): > {code} > task0 task1 task2 > compute updates locally > -------------------------------- sends updates to master task > -------------------------------- merge updates and broadcast it to every tasks > compute updates locally > -------------------------------- sends updates to master task > -------------------------------- merge updates and broadcast it to every tasks > > ... > (Loop until onvergence) > {code} > By separating the master, we can support asynchronous parallel SGD. My idea is just using of a task0 (BSPTask) as a server daemon. In this issue ticket, single master is enough at this moment. > {code} > task0 | task1 .... taskN > | > | > | compute updates locally > | > Receive |<------ push updates to master task > Update1 | > +------> fetch updates > | > | > | > Receive |<------------------------------------ .. > Update2 | > +------------------------------------> .. > | > | > {code} -- This message was sent by Atlassian JIRA (v6.3.4#6332)