From asr32@cornell.edu Mon Apr 24 23:23:22 2006 Return-Path: X-Spam-Checker-Version: SpamAssassin 3.1.0 (2005-09-13) on sundial.cs.cornell.edu X-Spam-Status: No, score=-1.8 required=5.0 tests=AWL,BAYES_00, DATE_IN_PAST_24_48 autolearn=no version=3.1.0 X-Spam-Level: Received: from authusersmtp.mail.cornell.edu (granite1.mail.cornell.edu [128.253.83.141]) by sundial.cs.cornell.edu (8.11.7-20031020/8.11.7/M-3.25) with ESMTP id k3P3NM205976 for ; Mon, 24 Apr 2006 23:23:22 -0400 (EDT) Received: from dreadnought.cornell.edu (r253240123.resnet.cornell.edu [128.253.240.123]) (authenticated bits=0) by authusersmtp.mail.cornell.edu (8.13.1/8.12.10) with ESMTP id k3P3NIU8004276 (version=TLSv1/SSLv3 cipher=DHE-RSA-AES256-SHA bits=256 verify=NOT) for ; Mon, 24 Apr 2006 23:23:22 -0400 (EDT) Message-Id: <6.2.1.2.2.20060422234654.02ffdab8@postoffice8.mail.cornell.edu> X-Mailer: QUALCOMM Windows Eudora Version 6.2.1.2 Date: Sun, 23 Apr 2006 03:38:33 -0400 To: egs+summary@cs.cornell.edu From: Ari Rabkin Subject: PAPER 24 Mime-Version: 1.0 Content-Type: text/plain; charset="us-ascii"; format=flowed Seti@Home: Seti@Home is a system for massively distributed analysis of radio telescope data by idle hosts at the edge of the internet. It runs as a screensaver on user desktops. The analysis is mostly Fourier decomposition--CPU intensive, not requiring much storage, and readily repeated and verified to control for error. Code and data is distributed under signature. Seti@Home is build on top of the BOINC toolkit for distributed computation. BOINC is most applicable for CPU-bound tasks that can be divided into small data chunks, since bandwidth to edge nodes is limited. The central server must hand out every work unit, imposing substantial burden. The system has a number of worrisome security aspects. First, users must trust the supplier of the code--there is no sandboxing, so a malicious application can compromise user machines. Second, an attacker might be able to "upgrade" a user's machine back to a previously released (and signed) version of the application code, in which vulnerabilities have been found. BOINC would be much improved by sandboxing. CCOF: CCOF is a system for "cluster computing on the fly" -- for assigning cycle-intensive tasks to idle machines. CCOF offers algorithms for large workpile batch compute tasks, as well as point-of-presence tasks. The large workpile algorithm involves building a CAN and assigning nodes to locations based on their timezone, so that tasks can be given to idle machines during local night. The intent of this is to ensure that machines donating cycles are likely to be idle for long periods. The insight motivating the CCOF system is that machines are likely to be idle for long blocks during night hours. I'm writing this summary at 3:30 am, so this seems like an undue assumption. Moreover, it assumes that machines have their timezone set correctly. As its underlying substrate, it uses a CAN. This makes me suspect that CCOF doesn't really exist, since I am unaware of any CAN implementations. Ari Rabkin asr32@cornell.edu Risley Hall 454 3-2842 The resources of civilization are not yet exhausted. --William Gladstone From gp72@cornell.edu Tue Apr 25 00:04:04 2006 Return-Path: X-Spam-Checker-Version: SpamAssassin 3.1.0 (2005-09-13) on sundial.cs.cornell.edu X-Spam-Status: No, score=-2.1 required=5.0 tests=AWL,BAYES_00 autolearn=ham version=3.1.0 X-Spam-Level: Received: from penguin.cs.cornell.edu (penguin.cs.cornell.edu [128.84.96.11]) by sundial.cs.cornell.edu (8.11.7-20031020/8.11.7/M-3.25) with ESMTP id k3P443214231 for ; Tue, 25 Apr 2006 00:04:03 -0400 (EDT) Received: from postoffice10.mail.cornell.edu ([132.236.56.14]) by penguin.cs.cornell.edu with Microsoft SMTPSVC(6.0.3790.1830); Tue, 25 Apr 2006 00:03:12 -0400 Received: from orpheus3.dataserver.cornell.edu (orpheus3.dataserver.cornell.edu [128.253.161.167]) by postoffice10.mail.cornell.edu (8.12.10/8.12.6) with ESMTP id k3P43AEf018022; Tue, 25 Apr 2006 00:03:11 -0400 (EDT) Message-ID: <1652260983.1145937789754.JavaMail.webber@orpheus3.dataserver.cornell.edu> Date: Tue, 25 Apr 2006 00:03:09 -0400 (EDT) From: Gopal Parameswaran To: egs+summary@cs.cornell.edu Subject: PAPER 24 Cc: gp72@cornell.edu Mime-Version: 1.0 Content-Type: text/plain; charset=UTF-8 X-Mailer: uPortal WEB email client 3.0 X-OriginalArrivalTime: 25 Apr 2006 04:03:12.0441 (UTC) FILETIME=[2BC68A90:01C6681D] Content-Transfer-Encoding: 8bit X-MIME-Autoconverted: from quoted-printable to 8bit by sundial.cs.cornell.edu id k3P443214231 Cluster Computing on the Fly This paper discusses the application of peer to peer in cluster computing and harvesting cycles from ordinary users in an open access, non institutional environment and discusses the different forms of harnessing the idle computer cycles and then concludes with a scheduling method based on usages on night time free times for machines. As the author states the goal of this paper is to develop a scheduling infrastructure that can support automatic scheduling for p2p cycle sharing applications viz. Infinite workpile applications such as SETI which require a massive amount of computation time and usually operates under a master slave mode based scheduling of jobs, workpile applications with deadlines which are similar to the infinite workpiles but are on a smaller scale and are driven by deadlines and Tree and point of presence based applications. However the peer to peer community’s main contributions would come as the author suggests in an open P2p cycle sharing environment where different hosts can share their cycles and take advantage of the vast resources of a global network of machines especially that is disparate in terms of free cycle times due to the different time zones resulting in different periods of inactivity for the machines. The authors approach relies on distributed schedulers with localized schedulers scheduling tasks across a local geographic region that constitutes a time zone with a global application scheduler that looks into scheduling requirements and managing and verifying results. Their concept called wave scheduler seeks to capture cycles from the millions of machines idle at night by following time zones which are represented in a CAN overlay with time zone being chosen as one of the dimensions in the d-dimensional mesh. When a host joins the network and wishes to share its cycles it randomly selects a node label and joins the network in the time zone in which it is. The application scheduler knows which time zones are night time zones and decides on the number of hosts that a particula ion of the scheduling if it was done partly with the local scheduler with the application scheduler deciding on the ration of allocation to different time zones and letting the local scheduler decide on the actual local scheduling in a time zone would have resulted in a more intelligent control. The authors have explored an area of application for peer to peer which I believe could be the future of computing for the ordinary users. From pjk25@cornell.edu Tue Apr 25 01:16:21 2006 Return-Path: X-Spam-Checker-Version: SpamAssassin 3.1.0 (2005-09-13) on sundial.cs.cornell.edu X-Spam-Status: No, score=-2.2 required=5.0 tests=AWL,BAYES_00 autolearn=unavailable version=3.1.0 X-Spam-Level: Received: from penguin.cs.cornell.edu (penguin.cs.cornell.edu [128.84.96.11]) by sundial.cs.cornell.edu (8.11.7-20031020/8.11.7/M-3.25) with ESMTP id k3P5GL228814 for ; Tue, 25 Apr 2006 01:16:21 -0400 (EDT) Received: from authusersmtp.mail.cornell.edu ([128.253.83.141]) by penguin.cs.cornell.edu with Microsoft SMTPSVC(6.0.3790.1830); Tue, 25 Apr 2006 01:15:21 -0400 Received: from [10.0.1.3] (cpe-69-207-37-155.twcny.res.rr.com [69.207.37.155]) (authenticated bits=0) by authusersmtp.mail.cornell.edu (8.13.1/8.12.10) with ESMTP id k3P5FKLR015638 (version=TLSv1/SSLv3 cipher=RC4-SHA bits=128 verify=NOT) for ; Tue, 25 Apr 2006 01:15:21 -0400 (EDT) Mime-Version: 1.0 (Apple Message framework v749.3) To: egs+summary@cs.cornell.edu Message-Id: <6F5A542A-94D5-4099-B427-AF3324151F59@cornell.edu> Content-Type: multipart/signed; micalg=sha1; boundary=Apple-Mail-16-131214043; protocol="application/pkcs7-signature" From: Philip Kuryloski Subject: PAPER 24 Date: Tue, 25 Apr 2006 01:15:38 -0400 X-Mailer: Apple Mail (2.749.3) X-OriginalArrivalTime: 25 Apr 2006 05:15:21.0331 (UTC) FILETIME=[3FFEBC30:01C66827] --Apple-Mail-16-131214043 Content-Transfer-Encoding: 7bit Content-Type: text/plain; charset=US-ASCII; delsp=yes; format=flowed CLUSTER COMPUTING ON THE FLY: The authors propose a distributed P2P system which attempts to schedule the use of client machines throughout the internet at night in their local timezone using an overlay network. Participating hosts are quizzed to evaluate the level at which they can be trusted as members of the cycle sharing network. The system handles all aspects of scheduling at both the global and local level, which the authors call generally a "P2P Scheduling System". The authors see four broad classes of applications: infinite workpile, deadline driven workpile, tree based search, and point of presence, all off which they attempt to handle. These essentially reflect different degrees of coordination between donating nodes. The primary disparity between traditional grid computing and P2P clustering is that there is an implied amount of trust or value in the jobs being submitted (otherwise users would not participate). In an open system, such guarantees are not implicit. Furthermore, there is potentially less confidence or trust in other participating nodes. Also, there is potentially a much larger variation in the resources needed by particular jobs, making management, finding, and scheduling of resources more difficult. --Apple-Mail-16-131214043 Content-Transfer-Encoding: base64 Content-Type: application/pkcs7-signature; name=smime.p7s Content-Disposition: attachment; filename=smime.p7s MIAGCSqGSIb3DQEHAqCAMIACAQExCzAJBgUrDgMCGgUAMIAGCSqGSIb3DQEHAQAAoIIGFjCCAs8w ggI4oAMCAQICAw+L7TANBgkqhkiG9w0BAQQFADBiMQswCQYDVQQGEwJaQTElMCMGA1UEChMcVGhh d3RlIENvbnN1bHRpbmcgKFB0eSkgTHRkLjEsMCoGA1UEAxMjVGhhd3RlIFBlcnNvbmFsIEZyZWVt YWlsIElzc3VpbmcgQ0EwHhcNMDUwOTI2MTc1NzM0WhcNMDYwOTI2MTc1NzM0WjBDMR8wHQYDVQQD ExZUaGF3dGUgRnJlZW1haWwgTWVtYmVyMSAwHgYJKoZIhvcNAQkBFhFwamsyNUBjb3JuZWxsLmVk dTCCASIwDQYJKoZIhvcNAQEBBQADggEPADCCAQoCggEBALFkgyhHSufUWaYxKh+wvUSDmrM8cViE JjeRS7Ssdd5tf0ckH2iwktNuSkxSsavsmAY+8zahwJjwk/JWTVyOGW/QsjgA5zoTJeAz4ah/QcKZ hou20lN6NvlFZWA43b4/jwtpVVa2RMS1fitkEs7YA9N16akGyXCpJR2i6EVTk7tx8/zf7i7bqg4t tmbJaySQyMQ4QV1O+F00m+zms0WZN5XRDqPwU2/WZfUE5BK/pGLkkFheBGSJJssuOsct8ctup0AI fJLlLZZhBCEdNeM2x9KfQEm+Tk3Ty0zl0pOewe7oW9vgwBJ2LwTVurzQ7qXeq1VhkDmkQJOwjcxM ssGAPXMCAwEAAaMuMCwwHAYDVR0RBBUwE4ERcGprMjVAY29ybmVsbC5lZHUwDAYDVR0TAQH/BAIw ADANBgkqhkiG9w0BAQQFAAOBgQBanW/NR5+pfeGOS7lM21kLObfzzGKtHvTFZ/RS0cSgWSKaCZfx aLLhqC9EFFFxh0b4wn0zCTv4CQWhrpaPZZC7oroP70kqWypQdjFbQ2rlLrVVS8pE4gtjZnRPFMr0 BEH+1K7kWB6kTHvg2eI1EotCI92yARGzlzKrXjPonHppijCCAz8wggKooAMCAQICAQ0wDQYJKoZI hvcNAQEFBQAwgdExCzAJBgNVBAYTAlpBMRUwEwYDVQQIEwxXZXN0ZXJuIENhcGUxEjAQBgNVBAcT CUNhcGUgVG93bjEaMBgGA1UEChMRVGhhd3RlIENvbnN1bHRpbmcxKDAmBgNVBAsTH0NlcnRpZmlj YXRpb24gU2VydmljZXMgRGl2aXNpb24xJDAiBgNVBAMTG1RoYXd0ZSBQZXJzb25hbCBGcmVlbWFp bCBDQTErMCkGCSqGSIb3DQEJARYccGVyc29uYWwtZnJlZW1haWxAdGhhd3RlLmNvbTAeFw0wMzA3 MTcwMDAwMDBaFw0xMzA3MTYyMzU5NTlaMGIxCzAJBgNVBAYTAlpBMSUwIwYDVQQKExxUaGF3dGUg Q29uc3VsdGluZyAoUHR5KSBMdGQuMSwwKgYDVQQDEyNUaGF3dGUgUGVyc29uYWwgRnJlZW1haWwg SXNzdWluZyBDQTCBnzANBgkqhkiG9w0BAQEFAAOBjQAwgYkCgYEAxKY8VXNV+065yplaHmjAdQRw nd/p/6Me7L3N9VvyGna9fww6YfK/Uc4B1OVQCjDXAmNaLIkVcI7dyfArhVqqP3FWy688Cwfn8R+R NiQqE88r1fOCdz0Dviv+uxg+B79AgAJk16emu59l0cUqVIUPSAR/p7bRPGEEQB5kGXJgt/sCAwEA AaOBlDCBkTASBgNVHRMBAf8ECDAGAQH/AgEAMEMGA1UdHwQ8MDowOKA2oDSGMmh0dHA6Ly9jcmwu dGhhd3RlLmNvbS9UaGF3dGVQZXJzb25hbEZyZWVtYWlsQ0EuY3JsMAsGA1UdDwQEAwIBBjApBgNV HREEIjAgpB4wHDEaMBgGA1UEAxMRUHJpdmF0ZUxhYmVsMi0xMzgwDQYJKoZIhvcNAQEFBQADgYEA SIzRUIPqCy7MDaNmrGcPf6+svsIXoUOWlJ1/TCG4+DYfqi2fNi/A9BxQIJNwPP2t4WFiw9k6GX6E sZkbAMUaC4J0niVQlGLH2ydxVyWN3amcOY6MIE9lX5Xa9/eH1sYITq726jTlEBpbNU1341YheILc IRk13iSx0x1G/11fZU8xggLnMIIC4wIBATBpMGIxCzAJBgNVBAYTAlpBMSUwIwYDVQQKExxUaGF3 dGUgQ29uc3VsdGluZyAoUHR5KSBMdGQuMSwwKgYDVQQDEyNUaGF3dGUgUGVyc29uYWwgRnJlZW1h aWwgSXNzdWluZyBDQQIDD4vtMAkGBSsOAwIaBQCgggFTMBgGCSqGSIb3DQEJAzELBgkqhkiG9w0B BwEwHAYJKoZIhvcNAQkFMQ8XDTA2MDQyNTA1MTUzOFowIwYJKoZIhvcNAQkEMRYEFGDheGhgoPus FsV2r1Kb81mEZIfVMHgGCSsGAQQBgjcQBDFrMGkwYjELMAkGA1UEBhMCWkExJTAjBgNVBAoTHFRo YXd0ZSBDb25zdWx0aW5nIChQdHkpIEx0ZC4xLDAqBgNVBAMTI1RoYXd0ZSBQZXJzb25hbCBGcmVl bWFpbCBJc3N1aW5nIENBAgMPi+0wegYLKoZIhvcNAQkQAgsxa6BpMGIxCzAJBgNVBAYTAlpBMSUw IwYDVQQKExxUaGF3dGUgQ29uc3VsdGluZyAoUHR5KSBMdGQuMSwwKgYDVQQDEyNUaGF3dGUgUGVy c29uYWwgRnJlZW1haWwgSXNzdWluZyBDQQIDD4vtMA0GCSqGSIb3DQEBAQUABIIBAFwYOnUUSJ7R P4pXNNsphN0UDNYFXaRqe6ZbTWcGDOYRDq1rl2C5Q0vlb4AUyWf1V82AkbulFN0EeNxSvDNIxmPG H0mdJNV8uSc4SF1ikBBLyV0asgjiGouFgD3J63LLyKsRPL+MCyGkk6TJTKJDHi7BHhWAf6lAKfg2 r2dZEMJVPd0Ynfa0ByHOByAJrazbgac3nfrCaVXpCtcxI1ehKRmEbszM86zcUgRTCiXMSeyuOpdc R6PxTNHuL2xfo73UPncrsmSgokHuMopSd1uWYBgh92nSuNd/FQ84zGved2z99lz3gtIMZNQzrjDx z6X6219Z5cnH5HvUjCEZoRyGSqwAAAAAAAA= --Apple-Mail-16-131214043-- From pjk25@cornell.edu Tue Apr 25 01:36:06 2006 Return-Path: X-Spam-Checker-Version: SpamAssassin 3.1.0 (2005-09-13) on sundial.cs.cornell.edu X-Spam-Status: No, score=-2.2 required=5.0 tests=AWL,BAYES_00 autolearn=ham version=3.1.0 X-Spam-Level: Received: from iago.cs.cornell.edu (iago.cs.cornell.edu [128.84.96.10]) by sundial.cs.cornell.edu (8.11.7-20031020/8.11.7/M-3.25) with ESMTP id k3P5a6202931 for ; Tue, 25 Apr 2006 01:36:06 -0400 (EDT) Received: from authusersmtp.mail.cornell.edu ([128.253.83.141]) by iago.cs.cornell.edu with Microsoft SMTPSVC(6.0.3790.1830); Tue, 25 Apr 2006 01:35:05 -0400 Received: from [10.0.1.3] (cpe-69-207-37-155.twcny.res.rr.com [69.207.37.155]) (authenticated bits=0) by authusersmtp.mail.cornell.edu (8.13.1/8.12.10) with ESMTP id k3P5Z5bZ016681 (version=TLSv1/SSLv3 cipher=RC4-SHA bits=128 verify=NOT) for ; Tue, 25 Apr 2006 01:35:05 -0400 (EDT) Resent-Message-Id: <6F5A542A-94D5-4099-B427-AF3324151F59@cornell.edu> Mime-Version: 1.0 (Apple Message framework v749.3) Content-Type: text/plain; charset=US-ASCII; delsp=yes; format=flowed Resent-Date: Tue, 25 Apr 2006 01:35:23 -0400 Message-Id: <5808CDF1-3318-4BA2-9111-074F30940627@cornell.edu> Content-Transfer-Encoding: 7bit Resent-To: egs+summary@cs.cornell.edu From: Philip Kuryloski Subject: PAPER 24 Resent-From: Philip Kuryloski Date: Tue, 25 Apr 2006 01:15:38 -0400 To: egs+summary@cs.cornell.edu X-Mailer: Apple Mail (2.749.3) X-OriginalArrivalTime: 25 Apr 2006 05:35:06.0043 (UTC) FILETIME=[022370B0:01C6682A] CLUSTER COMPUTING ON THE FLY: The authors propose a distributed P2P system which attempts to schedule the use of client machines throughout the internet at night in their local timezone using an overlay network. Participating hosts are quizzed to evaluate the level at which they can be trusted as members of the cycle sharing network. The system handles all aspects of scheduling at both the global and local level, which the authors call generally a "P2P Scheduling System". The authors see four broad classes of applications: infinite workpile, deadline driven workpile, tree based search, and point of presence, all off which they attempt to handle. These essentially reflect different degrees of coordination between donating nodes. The primary disparity between traditional grid computing and P2P clustering is that there is an implied amount of trust or value in the jobs being submitted (otherwise users would not participate). In an open system, such guarantees are not implicit. Furthermore, there is potentially less confidence or trust in other participating nodes. Also, there is potentially a much larger variation in the resources needed by particular jobs, making management, finding, and scheduling of resources more difficult. CCOF handles this via quizzing of hosts. Scheduling is handled by the Wave Scheduler, named because it follows the local nighttime around the globe continuously, tracking nodes via a CAN DHT. Nodes forward workload to a random neighbor in the next time zone, causing a surge of activity to the next time zone. Although the potential to donate cycles in exchange for access to a global supercomputer sounds promising, CCOF is not without certain shortcomings. Primarily, it is difficult for the average user to generate a task which can run in a manner that sees benefit from such a computing resource. Thus, it is difficult to see a strong personal computational benefit from participating in such a system. Also, the cause which I donate to is now less identifiable and less trustable. The systems primary method of scheduling relies on using nighttime cycles. I have difficulty believing that this accurately reflects the availability of resources. Also, the system mentions reliance on a centralized certificate authority to authenticate system members, making in not truly P2P. From ns253@cornell.edu Tue Apr 25 08:56:29 2006 Return-Path: X-Spam-Checker-Version: SpamAssassin 3.1.0 (2005-09-13) on sundial.cs.cornell.edu X-Spam-Status: No, score=-2.4 required=5.0 tests=AWL,BAYES_00 autolearn=ham version=3.1.0 X-Spam-Level: Received: from authusersmtp.mail.cornell.edu (granite1.mail.cornell.edu [128.253.83.141]) by sundial.cs.cornell.edu (8.11.7-20031020/8.11.7/M-3.25) with ESMTP id k3PCuS218688 for ; Tue, 25 Apr 2006 08:56:28 -0400 (EDT) Received: from localhost (cpe-69-207-49-126.twcny.res.rr.com [69.207.49.126]) (authenticated bits=0) by authusersmtp.mail.cornell.edu (8.13.1/8.12.10) with ESMTP id k3PCuRPu021580 (version=TLSv1/SSLv3 cipher=DHE-RSA-AES256-SHA bits=256 verify=NOT) for ; Tue, 25 Apr 2006 08:56:28 -0400 (EDT) Date: Tue, 25 Apr 2006 08:56:28 -0400 From: Niranjan Sivakumar To: egs+summary@cs.cornell.edu Subject: PAPER 24 Message-Id: <20060425085628.90327854.ns253@cornell.edu> Organization: Cornell Law School X-Mailer: Sylpheed version 2.2.4 (GTK+ 2.8.13; i686-pc-linux-gnu) Mime-Version: 1.0 Content-Type: text/plain; charset=US-ASCII Content-Transfer-Encoding: 7bit Niranjan Sivakumar Cluster Computing on the Fly: P2P Scheduling of Idle Cycles in the Internet Cluster Computing on the Fly (CCOF) is a system that is designed to take advantage of idle CPU cycles of heterogeneous network members, The system provides a "quizzing" mechanism to determine trust levels of member nodes. CCOF also deals with scheduling at the local host level as well as coordinated scheduling across the network. Some incentives for fairness are considered, but the CCOF model assumes that members are generally "donating" cycles and thus do not necessarily care about cycle-cycle fairness. The proposed CCOF implementation is designed to run over CAN. The CAN network is partitioned into 24 sectors, one for each hour in the day. When nodes are ready to join the network, perhaps at night when the machine may generally be idle, it will select a node label in the zone corresponding to the current hour. An application will choose a subset of nodes to farm its workload out to. When a host leaves the overlay, there is a mechanism to transfer its state to a node that is available in the next zone. Results can be held in a CAN file system in the event that the application requesting results is offline at the time that work is completed and forwarded at a later time. One flaw seen in CCOF is that the authors have not yet dealt with issues relating to DoS attacks and the possibility of malicious users simply scheduling many meaningless tasks to reduce the efficiency of the system. They offload the difficulty of excluding untrusted nodes to the overlay network, but this has not always been dealt with effectively at that level. From nsg7@cornell.edu Tue Apr 25 09:13:11 2006 Return-Path: X-Spam-Checker-Version: SpamAssassin 3.1.0 (2005-09-13) on sundial.cs.cornell.edu X-Spam-Status: No, score=-2.3 required=5.0 tests=AWL,BAYES_00 autolearn=ham version=3.1.0 X-Spam-Level: Received: from postoffice10.mail.cornell.edu (postoffice10.mail.cornell.edu [132.236.56.14]) by sundial.cs.cornell.edu (8.11.7-20031020/8.11.7/M-3.25) with ESMTP id k3PDDB222069 for ; Tue, 25 Apr 2006 09:13:11 -0400 (EDT) Received: from webmail.cornell.edu (hermes21.mail.cornell.edu [132.236.56.20]) by postoffice10.mail.cornell.edu (8.12.10/8.12.6) with ESMTP id k3PDD9Ac004782 for ; Tue, 25 Apr 2006 09:13:09 -0400 (EDT) Received: from 132.236.227.119 by webmail.cornell.edu with HTTP; Tue, 25 Apr 2006 09:13:10 -0400 (EDT) Message-ID: <1858.132.236.227.119.1145970790.squirrel@webmail.cornell.edu> Date: Tue, 25 Apr 2006 09:13:10 -0400 (EDT) Subject: PAPER 24 From: "Nicholas S Gerner" To: egs+summary@cs.cornell.edu User-Agent: SquirrelMail/1.4.5 MIME-Version: 1.0 Content-Type: text/plain;charset=iso-8859-1 Content-Transfer-Encoding: 8bit X-Priority: 3 (Normal) Importance: Normal SETI@home is a massively distributed cycle donation system where data collected by the Search for Extraterrestrial Intelligence project distributes "work units" to client systems which are computer users who donate their computers by installing the SETI@home screensaver. The goal is to identify a signal from an intelligent source by scanning frequency ranges and varying many other parameters. The dataset is very large (many frequencies over a long period of time must be considered), but consists of many independent "work units". These work units each are small (350KB) but require a great deal of computation (many parameter combinations must be considered). Clients request work units as cycles become available using an HTTP-like protocol from a centralized server. The work units are redundantly computed to verify results. If any positive result does come through an additional out-of-band scientific protocol exists to further verify the result. In this way SETI@home was able to utilize an average of 27.36TFLOPS. SETI@home is technically simple, reliable results are obtained through replicated computation, there is no trust measure. But SETI@home does introduce important social advances in public-cycle sharing. While incentives are not technically addressed by SETI@home, they do exist out-of-band. The client application acts as a screen-saver, visualizing result computation. While this may seem trival, developers have found that this is an important incentive for users. SETI@home also maintains a website with rankings of groups donating cycles by amount of cycles donated, so some accounting exists at the server. This incentive mechanism has been successfully attacked, but this poses no direct threat to result reliability (through redundant computation) or throughput. SETI@home is considered a wild success, however the infrastructure (technical and social) is single-purpose and cannot be harnessed by other researchers seeking to take advantage of the resources available to SETI@home. Every such @home project must develop its own infrastructure ( technical and social). Cluster Computing on the Fly is an architecture for cycle-sharing applications. in "Cluster Computing on the Fly: ..." Lo, Zappala, Zhou, Liu and Zhao begin to address some of the problems faced by such an architecture: scheduling, resource discovery, incentives, trust and security. Three classes of cycle-sharing applications are identified: infinite workpile applications where an infinite stream of independent work-units are available for client consumption (as in SETI@home), workpile applications with deadlines where work-units must be computed by a given deadline, tree-based search applications where clients interact to some degree to help produce and bound future work-units, and point-of-presence applications where client position in some space is important (e.g. for measurement studies of the internet). This paper goes on to present an the Wave Scheduling algorithm seeking to provide uninterrupted access to dedicated machines. These machines are organized in a CAN overlay organized by time-zone. hosts are elected for dedication by selecting the CAN zones which lie in the current off-time timezones (e.g. night-time). As time goes on some dedicated hosts are freed as their time-zone becomes on-time (e.g. 8:00am) and new hosts are elected when their zone becomes off-time (e.g. 5:00pm). In this way dedicated hosts are available to the scheduler without impact to the user (since his or her machine is unused during off-time). The SETI@home system seeks to solve a similar problem: utilize cycles while the user's machine is not needed. However, SETI@home does this without an overlay by using a screen-saver. It's not clear that an overlay is needed for such a system. Although the fact of having an overlay identify likely un-used machines is a nice property, this is not the main goal of Wave Scheduling. From km266@cornell.edu Tue Apr 25 11:32:55 2006 Return-Path: X-Spam-Checker-Version: SpamAssassin 3.1.0 (2005-09-13) on sundial.cs.cornell.edu X-Spam-Status: No, score=-1.8 required=5.0 tests=AWL,BAYES_00 autolearn=ham version=3.1.0 X-Spam-Level: Received: from penguin.cs.cornell.edu (penguin.cs.cornell.edu [128.84.96.11]) by sundial.cs.cornell.edu (8.11.7-20031020/8.11.7/M-3.25) with ESMTP id k3PFWs226430 for ; Tue, 25 Apr 2006 11:32:54 -0400 (EDT) Received: from authusersmtp.mail.cornell.edu ([128.253.83.141]) by penguin.cs.cornell.edu with Microsoft SMTPSVC(6.0.3790.1830); Tue, 25 Apr 2006 11:32:13 -0400 Received: from KEVSTOY (cpe-69-207-37-246.twcny.res.rr.com [69.207.37.246]) (authenticated bits=0) by authusersmtp.mail.cornell.edu (8.13.1/8.12.10) with ESMTP id k3PFWDYa022428 (version=TLSv1/SSLv3 cipher=RC4-MD5 bits=128 verify=NOT) for ; Tue, 25 Apr 2006 11:32:13 -0400 (EDT) Message-ID: <000501c6687d$8dfdd0a0$f625cf45@KEVSTOY> Reply-To: "Kevin" From: "Kevin" To: Subject: PAPER 24 Date: Tue, 25 Apr 2006 11:33:08 -0400 MIME-Version: 1.0 Content-Type: text/plain; format=flowed; charset="iso-8859-1"; reply-type=original Content-Transfer-Encoding: 7bit X-Priority: 3 X-MSMail-Priority: Normal X-Mailer: Microsoft Outlook Express 6.00.2900.2869 X-MimeOLE: Produced By Microsoft MimeOLE V6.00.2900.2869 X-OriginalArrivalTime: 25 Apr 2006 15:32:13.0843 (UTC) FILETIME=[6D2BDE30:01C6687D] Cluster Computing on the Fly presents and gives history on cycle-sharing applications. First, they go into the four broad categories: infinite workpile applications, workpiles with deadlines, tree-based search, and point-of-presence applications. Infinite workpile applications, like SETI@home, have a master-slave architecture that follows the standard server-client paradigm. The server sends out CPU-intensive work to be processed by the slaves, which reply with the result once it is computed. Workpiles with deadlines are similar to infinite workpiles but these have deadlines, usually measured in days or weeks. Tree-based applications have many server-client relationships: a child node is your client while your parent node is your master. The entire tree is a cycle-sharing structure. Finally point-of-presence applications consume few cycles but are present all over the place: they might try to try latency, bandwidth, or other things all over the physical internet. The goal of the paper's contribution, wave scheduling, is to help the infinite workpile effectively get cycles during the nighttime. The network is organized such that nodes are in the system if it is nighttime there, sharing their cycles with the community. It seems a bit odd that the authors only want cycles shared at night. While it might be a general trend that people don't use their computers at night, a lot of people probably leave them on during the day while they are at work and other times, detecting idleness might be a better goal here... From asg46@cornell.edu Tue Apr 25 12:08:11 2006 Return-Path: X-Spam-Checker-Version: SpamAssassin 3.1.0 (2005-09-13) on sundial.cs.cornell.edu X-Spam-Status: No, score=-2.3 required=5.0 tests=AWL,BAYES_00 autolearn=ham version=3.1.0 X-Spam-Level: Received: from iago.cs.cornell.edu (iago.cs.cornell.edu [128.84.96.10]) by sundial.cs.cornell.edu (8.11.7-20031020/8.11.7/M-3.25) with ESMTP id k3PG8A205702 for ; Tue, 25 Apr 2006 12:08:10 -0400 (EDT) Received: from postoffice10.mail.cornell.edu ([132.236.56.14]) by iago.cs.cornell.edu with Microsoft SMTPSVC(6.0.3790.1830); Tue, 25 Apr 2006 12:07:07 -0400 Received: from webmail.cornell.edu (hermes21.mail.cornell.edu [132.236.56.20]) by postoffice10.mail.cornell.edu (8.12.10/8.12.6) with ESMTP id k3PG75OU020412 for ; Tue, 25 Apr 2006 12:07:05 -0400 (EDT) Received: from 128.84.98.90 by webmail.cornell.edu with HTTP; Tue, 25 Apr 2006 12:07:06 -0400 (EDT) Message-ID: <1097.128.84.98.90.1145981226.squirrel@webmail.cornell.edu> Date: Tue, 25 Apr 2006 12:07:06 -0400 (EDT) Subject: paper 24 From: "Abhishek Santosh Gupta" To: egs+summary@cs.cornell.edu User-Agent: SquirrelMail/1.4.5 MIME-Version: 1.0 Content-Type: text/plain;charset=iso-8859-1 Content-Transfer-Encoding: 8bit X-Priority: 3 (Normal) Importance: Normal X-OriginalArrivalTime: 25 Apr 2006 16:07:07.0144 (UTC) FILETIME=[4CE04080:01C66882] CLUSTER COMPUTING... deals with harnessing idle compute cycles throughout the Internet. this must include the following activities: overlay management for hosts distributing cycles,resource discovery within the overlay, application based scheduling,local scheduling and meta-level scheduling resource discovery becomes difficult when the resource(compute cycles) is perishable,cannot be shared and is dynamic. 4 search techniques were examined : expanding ring, advertisement based, random walk and rendezvous point. Rendezvous point performed better under light loads and outperformed others when message passing overheads were compared. the authors identify 4 classes of problems 1) infinite workpile applications that consume a huge amount of compute time under a master-slave model. no communication is required b/w slave nodes. 2) workpile applications with deadlines are deadline-driven but the compute cycles required are moderate. 3) tree based search applications require substantial compute cycles with loose coordination among subtasks. e.g. communicating a bound in a search tree 4) Point-of-Presence applications consume minimal cycles but require placement throughout the Internet. e.g. distributed monitoring applications. the authors suggest a wave scheduler which uses a CAN-based DHT. time zones are represented by a d-dimensional mesh. each zone represents a particular night zone. the joining node has the freedom to choose its night zone. when morning comes to a host node, it selects a new target night zone, randomly selects a node in that night zone for migration, and after negotiation the task is migrated to the new zone. results may be returned to the application or stored in the DHT file system and retrieved using DHT lookup. the trust value for each node is determined using a quizzing mechanism. this trust value is used to select a node for a particular task. From sh366@cornell.edu Tue Apr 25 12:33:42 2006 Return-Path: X-Spam-Checker-Version: SpamAssassin 3.1.0 (2005-09-13) on sundial.cs.cornell.edu X-Spam-Status: No, score=-2.4 required=5.0 tests=AWL,BAYES_00 autolearn=ham version=3.1.0 X-Spam-Level: Received: from penguin.cs.cornell.edu (penguin.cs.cornell.edu [128.84.96.11]) by sundial.cs.cornell.edu (8.11.7-20031020/8.11.7/M-3.25) with ESMTP id k3PGXf211830 for ; Tue, 25 Apr 2006 12:33:41 -0400 (EDT) Received: from postoffice10.mail.cornell.edu ([132.236.56.14]) by penguin.cs.cornell.edu with Microsoft SMTPSVC(6.0.3790.1830); Tue, 25 Apr 2006 12:31:00 -0400 Received: from orpheus3.dataserver.cornell.edu (orpheus3.dataserver.cornell.edu [128.253.161.167]) by postoffice10.mail.cornell.edu (8.12.10/8.12.6) with ESMTP id k3PGUxSW009678 for ; Tue, 25 Apr 2006 12:31:00 -0400 (EDT) Message-ID: <2089943011.1145982659203.JavaMail.webber@orpheus3.dataserver.cornell.edu> Date: Tue, 25 Apr 2006 12:30:59 -0400 (EDT) From: Huang Shiang-Jia To: egs+summary@cs.cornell.edu Subject: PAPER 24 Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Transfer-Encoding: 7bit X-Mailer: uPortal WEB email client 3.0 X-OriginalArrivalTime: 25 Apr 2006 16:31:01.0093 (UTC) FILETIME=[A3938150:01C66885] * This paper presents a P2P scheduling system that aims to utilize the available idle compute cycles throughout the Internet. It encompasses all activities, such as overlay construction, resource discovery, local scheduling, application-level scheduling as well as trust and fairness among peers, which are involved in the management of idle cycles. * OOCF differs from SETI@home in its 'automatic' scheduling of four classes of cycle-sharing systems: infinite workpile, deadline-driven workpile, tree-based search and point of presence, in an open environment. From lackhand@gmail.com Tue Apr 25 13:01:50 2006 Return-Path: X-Spam-Checker-Version: SpamAssassin 3.1.0 (2005-09-13) on sundial.cs.cornell.edu X-Spam-Status: No, score=-0.4 required=5.0 tests=AWL,BAYES_00,HTML_00_10, HTML_MESSAGE,RCVD_IN_BL_SPAMCOP_NET autolearn=no version=3.1.0 X-Spam-Level: Received: from penguin.cs.cornell.edu (penguin.cs.cornell.edu [128.84.96.11]) by sundial.cs.cornell.edu (8.11.7-20031020/8.11.7/M-3.25) with ESMTP id k3PH1o218738 for ; Tue, 25 Apr 2006 13:01:50 -0400 (EDT) Received: from pproxy.gmail.com ([64.233.166.176]) by penguin.cs.cornell.edu with Microsoft SMTPSVC(6.0.3790.1830); Tue, 25 Apr 2006 13:00:58 -0400 Received: by pproxy.gmail.com with SMTP id b36so193261pyb for ; Tue, 25 Apr 2006 10:00:57 -0700 (PDT) DomainKey-Signature: a=rsa-sha1; q=dns; c=nofws; s=beta; d=gmail.com; h=received:message-id:date:from:to:subject:mime-version:content-type; b=hw6oUKdTZ3IOe8dAR6eKY4GUEFDU7wLgX0riH1JKpZicSt3peh5KwJxu6w0VX5QyadgYmPjYMgL7BnvfatgsyQQZAPosDTVp6ION8IZJtUbVEi2Jh9vDzk2Vmjo1Rg40TYrQbeWgHYM+loMdrwUrf6PDQ5ljosywEj8aSlVLl9k= Received: by 10.35.49.4 with SMTP id b4mr2428539pyk; Tue, 25 Apr 2006 10:00:57 -0700 (PDT) Received: by 10.35.125.16 with HTTP; Tue, 25 Apr 2006 10:00:57 -0700 (PDT) Message-ID: <9aa7a97d0604251000w18fe0b10p3e054d5796d4ae06@mail.gmail.com> Date: Tue, 25 Apr 2006 13:00:57 -0400 From: "Andrew Cunningham" To: egs+summary@cs.cornell.edu Subject: PAPER 24 MIME-Version: 1.0 X-Security: message sanitized on sundial.cs.cornell.edu See http://www.impsec.org/email-tools/sanitizer-intro.html for details. $Revision: 1.148 $Date: 2004-12-19 11:59:17-08 X-Security: The postmaster has not enabled quarantine of poisoned messages. Content-Type: multipart/alternative; boundary="----=_Part_7615_26869454.1145984457547" X-OriginalArrivalTime: 25 Apr 2006 17:00:58.0142 (UTC) FILETIME=[D2B36BE0:01C66889] ------=_Part_7615_26869454.1145984457547 Content-Type: text/plain; charset=ISO-8859-1 Content-Disposition: inline Content-Transfer-Encoding: quoted-printable Andrew Cunningham arc39 Cluster Computing on the Fly: P2P Scheduling of Idle Cycles in the Internet Virginia Lo, Daniel Zappala, Dayi Zhou, Yuhong Liu, and Shanyu Zhao This system seeks to harvest cycles from ordinary users in an open access, non-institutional environment. It encompasses all activities involved in the management of idle cycles -- overlay construction for hosts donating cycles, resource discovery within the overlay, application-based scheduling, local scheduling on the host node, and meta-level scheduling among a community of application-level schedulers. Four important classes o= f cycle sharing application are identified and given outlines of requirements -- workpile, workpile with deadlines, tree based search, and point-of-presence. They then describe the Wave Scheduler for workpile tasks that exploits the Earth's day-and-night nature to harvest idle cycles, and the Point of Presence scheduler to discover and schedule hosts that meet application-specific requirements for location, topology, and resources. What the paper does not present is a working system that implements these ideals, nor the specifics of task migration, which are of course application specific, but equally, the most important part of the system. What is covered is when and to where the migration must occur. This is useful and an elegant solution, but severely restricts the class of applications which can be run, namely, to those which can be easily migrate= d every twelve hours. Another restriction is that, while mention is made of security concerns, they are not truly addressed -- for a system intended to run application code, a user of this system has no guarantees about the safety of their system or the legitimacy of code, while the entity which provides the code has several guarantees, due to the quizzing subsystem. Thus the security model is flawed in several ways, since there is no incentive other than the purely altruistic one to run this system, but good reasons not to. ------=_Part_7615_26869454.1145984457547 Content-Type: text/html; charset=ISO-8859-1 Content-Disposition: inline Content-Transfer-Encoding: quoted-printable Andrew Cunningham
arc39
    Cluster Computing on the Fly: P2P Scheduling of Idle Cyc= les in the Internet
    Virginia Lo, Daniel Zappala, Dayi Zhou, Yuhong Liu, and = Shanyu Zhao
    
    This system seeks to harvest cycles from ordinary users in an open access, non-institutional environment. It encompasses all activities involved in the management of idle cycles -- overlay construction for hosts donating cycles, resource discovery within the overlay, application-based scheduling, local scheduling on the host node, and meta-level scheduling among a community of application-level schedulers. Four important classes of cycle sharing application are identified and given outlines of requirements -- workpile, workpile with deadlines, tree based search, and point-of-presence. They then describe the Wave Scheduler for workpile tasks that exploits the Earth's day-and-night nature to harvest idle cycles, and the Point of Presence scheduler to discover and schedule hosts that meet application-specific requirements for location, topology, and resources.     What the paper does not present is a working system that implements these ideals, nor the specifics of task migration, which are of course application specific, but equally, the most important part of the system. What is covered is when and to where the migration must occur. This is useful and an elegant solution, but severely restricts the class of applications which can be run, namely, to those which can be easily migrated every twelve hours. Another restriction is that, while mention is made of security concerns, they are not truly addressed -- for a system intended to run application code, a user of this system has no guarantees about the safety of their system or the legitimacy of code, while the entity which provides the code has several guarantees, due to the quizzing subsystem. Thus the security model is flawed in several ways, since there is no incentive other than the purely altruistic one to run this system, but good reasons not to. ------=_Part_7615_26869454.1145984457547-- From victoria@cs.hmc.edu Tue Apr 25 13:27:22 2006 Return-Path: X-Spam-Checker-Version: SpamAssassin 3.1.0 (2005-09-13) on sundial.cs.cornell.edu X-Spam-Status: No, score=-2.3 required=5.0 tests=AWL,BAYES_00 autolearn=ham version=3.1.0 X-Spam-Level: Received: from penguin.cs.cornell.edu (penguin.cs.cornell.edu [128.84.96.11]) by sundial.cs.cornell.edu (8.11.7-20031020/8.11.7/M-3.25) with ESMTP id k3PHRL224459 for ; Tue, 25 Apr 2006 13:27:21 -0400 (EDT) Received: from turing.cs.hmc.edu ([134.173.42.99]) by penguin.cs.cornell.edu with Microsoft SMTPSVC(6.0.3790.1830); Tue, 25 Apr 2006 13:25:20 -0400 Received: by turing.cs.hmc.edu (Postfix, from userid 34382) id AE02C5326A; Tue, 25 Apr 2006 09:54:24 -0700 (PDT) Date: Tue, 25 Apr 2006 09:54:24 -0700 From: Victoria Krafft To: egs+summary@cs.cornell.edu Subject: PAPER 24 Message-ID: <20060425165424.GA1100@cs.hmc.edu> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline User-Agent: Mutt/1.4.2.1i X-OriginalArrivalTime: 25 Apr 2006 17:25:20.0399 (UTC) FILETIME=[3A4621F0:01C6688D] The CCOF system is designed to support the wide variety of applications which wish to take advantage of the spare cycles available on many computers. The authors have divided these applications into several different types, which require different scheduling algorithms. First, there are infinite workload applications, where there is a huge pile of data to be processed, and no set deadline for completion. Second, there are similar projects which wish to analyze large quantities of data, but have deadlines the data must be analyzed by. Third are tree-based search applications, which assign different sections of the problem to different slave nodes, which must communicate with each other to share new discoveries. Finally, there are point-of-presence applications, where not many CPU cycles are used, but the slave nodes should be scattered across the network. Schedulers for two of these types of projects are presented. For infinite workload applications, a wave scheduling protocol puts the nodes into a CAN-based DHT, where one of the dimensions is timezone, and then work is sent to those computers for which is it currently nighttime. For point-of-presence applications, a couple of possible schemes to elect leaders which will cover the network from the pool of slave nodes. Seti@Home is a system designed to distribute the cost of analyzing radio signals across a large number of computers with unused cycles. The amount of data which needs to be sent is small in comparison to the amount of analysis which is needed, making the bandwidth costs reasonable. -- Victoria Krafft From kelvinso@gmail.com Tue Apr 25 16:58:09 2006 Return-Path: X-Spam-Checker-Version: SpamAssassin 3.1.0 (2005-09-13) on sundial.cs.cornell.edu X-Spam-Status: No, score=-2.5 required=5.0 tests=AWL,BAYES_00 autolearn=ham version=3.1.0 X-Spam-Level: Received: from penguin.cs.cornell.edu (penguin.cs.cornell.edu [128.84.96.11]) by sundial.cs.cornell.edu (8.11.7-20031020/8.11.7/M-3.25) with ESMTP id k3PKw9217695 for ; Tue, 25 Apr 2006 16:58:09 -0400 (EDT) Received: from wproxy.gmail.com ([64.233.184.239]) by penguin.cs.cornell.edu with Microsoft SMTPSVC(6.0.3790.1830); Tue, 25 Apr 2006 16:57:59 -0400 Received: by wproxy.gmail.com with SMTP id i4so1403055wra for ; Tue, 25 Apr 2006 13:57:59 -0700 (PDT) DomainKey-Signature: a=rsa-sha1; q=dns; c=nofws; s=beta; d=gmail.com; h=received:message-id:date:from:to:subject:mime-version:content-type:content-transfer-encoding:content-disposition; b=keUjRfLMOaqmG2lj7/Uo4lHfIN96/IImBT2gotOZnQ+K4JO81qeJXuwvZJUy5mPlTWc7FNMM892LzrOrERAUfavHK7SFzttT4KHAgzOBR94B1CtwZG8IlA/baJYjtCbZB7v47ySQ0FYC/Dby98/O5Mof3wgr4cZhQyrf2ljbR/g= Received: by 10.54.154.8 with SMTP id b8mr4927892wre; Tue, 25 Apr 2006 08:54:50 -0700 (PDT) Received: by 10.54.80.8 with HTTP; Tue, 25 Apr 2006 08:54:50 -0700 (PDT) Message-ID: <6e1ca4560604250854s3bef0c96u99f77907850882aa@mail.gmail.com> Date: Tue, 25 Apr 2006 11:54:50 -0400 From: "Chiu Wah Kelvin So" To: egs+summary@cs.cornell.edu Subject: Paper 24 MIME-Version: 1.0 Content-Type: text/plain; charset=ISO-8859-1 Content-Disposition: inline X-OriginalArrivalTime: 25 Apr 2006 20:57:59.0504 (UTC) FILETIME=[EF4B0D00:01C668AA] Content-Transfer-Encoding: 8bit X-MIME-Autoconverted: from quoted-printable to 8bit by sundial.cs.cornell.edu id k3PKw9217695 The paper, "Cluster Computing on the Fly," presents architecture CCOF to use up the idle compute cycles in the internet. The authors first identify the four major classes of cycle-sharing applications, which include infinite workpile applications, workpile applications with deadlines, tree-based search application, and point-of-presence applications. CCOF support two classes of applications, workpile applications with deadlines and point of presence applications. The authors observe that the night time idle cycles are more likely to not be interrupted from users reclaiming their machines. Therefore, CCOF's Wave Scheduler uses a CAN-based DHT to assign nodes into different time-zone, and it assigns work to nodes in one of the night-time zone. The application can also store the results in the DHT when it is not online. Second, it verifies correctness of results by using two methods for quizzing hosts. It packets the quizzes into similar packet as the application code, and sends it from time to time. Or it includes short quizzes into the application code. The result of the quizzes can be used to compute trust of a host. Finally, it suggests a way to schedule nodes for Point-of-Presence applications by using CAN and Lee distances to elect leaders.