Download presentation
Presentation is loading. Please wait.
Published byDonna Watts Modified over 9 years ago
1
Andrew McNab - Manchester HEP - 17 September 2002 UK Testbed Deployment Aim of this talk is to the answer the questions: –“How much of the Testbed has been deployed in the UK?” –“What is in place to help sites and users join and use the Testbed?”
2
Andrew McNab - Manchester HEP - 17 September 2002 Existing “Testbed 0” Many sites have some version of Globus Toolkit installed: –GT1.1.3 or 2.0 built from source locally. –UK HEP GT1.1.3 RPMs –GridPP/EDG GT2.0 RPMs BaBar Grid gatekeepers EDG Testbed site Gavin’s Green Dot Map monitors nominated gatekeepers every ~25 minutes Need to move this to Testbed 1 with resource brokering before we’re really “a Grid”
3
Andrew McNab - Manchester HEP - 17 September 2002 EDG Software Releases Deployment largely driven/limited by EDG software release status and cycles. Testbed 1.2 release orginally intended to be stable for applications (eg Atlas Data Challenge.) 1.2.1 released ~10 days ago for application/“production” testbed sites (now up to 1.2.2 after minor additions.) Still some bugs limiting job sizes, which will need upgrades to fix (current version of Globus does not have these problems.) Uncertainty about “renormalisation” of 1.2 -> 1.3 Due to complexity of installation and configuration, need to use LCFG to install and manage the site. –Many configuration steps implemented in LCFG configuration are not fully documented elsewhere. –Therefore requires dedicated machines for each testbed site.
4
Andrew McNab - Manchester HEP - 17 September 2002 UK Deployment Plan Start with UK-WP6 people (+ other key experts) –Use tb-support@jiscmail.ac.uk mailing list, which anyone can join and is archived. Once have some UK WP6 sites up and a procedure that should work for any site, then ask more sites to test installation procedure, docs etc. Once this is stable, invite all interested sites to install Testbed software: by this point, installation instructions should be clear and not require previous grid installation experience. Support will then be provided by tb-support mailing list on best-effort basis, and by a formal ticket-based system. Sites will be validated using GridPP services (RB, VO, RC) and then proposed to EDG-WP6 for inclusion in EDG Testbed.
5
Andrew McNab - Manchester HEP - 17 September 2002 Testbed Status by site Bristol1.2b10 CE + SE + 2*WN + RC Cambridge1.2b9 CE + SE + 16*WN Imperial1.2.0 CE + SE + 2*WN + RB Liverpool1.2b10 CE + SE + 2*WN Manchester1.2.2 CE + SE + 6*WN + farms RAL1.2.1+ 2 * (CE + SE + 10*WN) UCL1.2.b9 CE + SE + WN This is based on last reported status at the phone conferences. All these sites are accessible via the GridPP Resource Broker at IC RAL and IC also in CERN RB. All installed by LCFG.
6
Andrew McNab - Manchester HEP - 17 September 2002 Testbed Support Web and List http://www.gridpp.ac.uk/tb-support/ on GridPP website. Has links to GridPP-provided Testbed services and installation recipes. The tb-support@jiscmail.ac.uk mailing list is intended to be the primary support and announcements forum. Instructions for joining and access to the posting archives are available at the Testbed webpages. If you want to participate in the Testbed, please get someone from your institute to join this list.
7
Andrew McNab - Manchester HEP - 17 September 2002 Testbed Support phone conferences These are arranged via the tb-support list and advertised on the website and the list. They are intended to be short, frequent, short, informal and short. ~1hr (and shorter if possible) @ ~7-14 days. Currently averaging about one every 2 weeks; different days of the week due to clashes with conferences. Two main purposes: –UK WP6 sees Testbed site by site - early flagging of problems. –Sites can get an overview of the state of the Testbed in the UK, including where they stand in relation to everyone else. –Aim to raise issues and direct effort at them, rather than trying to fix problems during meetings.
8
Andrew McNab - Manchester HEP - 17 September 2002 install.gridpp.ac.uk Installation procedure for LCFG server at each site: –modified RedHat HTTP network install –automates standard LCFG server installation procedure –puts all the commands into a small number of scripts (eg 1) Installation of the LCFG server almost turnkey now. Sites still need to configure local site description files manually. Installation and configuration of testbed elements (CE etc) is then almost entirely automatic. The steps necessary (including the otherwise undocumented “gotchas”) are described in the LCFG pages on the website.
9
Andrew McNab - Manchester HEP - 17 September 2002 GridPP Testbed Services rc.gridpp.ac.uk - Replica Catalog (Bristol) –used to advertise the location of replicas of files vo.gridpp.ac.uk - VO Authorisation Service (Mancheser) –provides lists of experiments’ authorised grid identities to sites. mds.gridpp.ac.uk - Information Service (RAL) –used by sites to advertise their properties and authorised users rb.gridpp.ac.uk - Resource Broker (Imperial) –uses information from MDS and RC to direct jobs to appropriate sites. gppui.gridpp.rl.ac.uk - User Interface node on CSF (RAL) –allows anyone with a certificate and a CSF account to use the EDG and GridPP Testbeds
10
Andrew McNab - Manchester HEP - 17 September 2002 Testbed Support “Bugzilla” Globus and EDG use Bugzilla ticket-based bug- tracking system. UK deployment plan calls for a ticket-based system for GridPP. Have set up an experimental system, which includes site admins in “front line” Still have to decide how this relates to Grid Support Centre system.
11
Andrew McNab - Manchester HEP - 17 September 2002 Bugzilla guidelines http://bugzilla.gridpp.ac.uk/ says: –“The GridPP Bugzilla system is intended for tracking and managing queries about the Testbed in the UK (not just bugs.) Site users wanting to report problems or ask questions via this system should select their "home" site as the "product" in their report (so that the report initially goes to their local site manager.)” –“Site managers reporting or dealing with queries outside their existing experience with the Testbed are encouraged to use the informal tb-support mailing list to attempt to resolve the question before passing it on to one of the Middleware "products" defined in Bugzilla.” –“Please bear in mind that many software problems cannot be solved within the UK. It may be better to report true bugs directly through the Globus Problem Reporting page or the EU DataGrid Bug Tracking page (both of which use Bugzilla too.)”
12
Andrew McNab - Manchester HEP - 17 September 2002 Monitoring Flagging problems is a vital part of keeping a Testbed available. In a changing environment, Testbed configuration will “go stale” if not actively maintained. –This also avoids pre-demo panics! Gavin’s map already monitors Globus functionality. Need monitoring of EDG job submission (Dave C.) and EDG information services (Steve T. using the NorduGrid tools.)
13
Andrew McNab - Manchester HEP - 17 September 2002 Summary Seven sites have EDG 1.2 Testbed software installed. All accessible via the GridPP Resource Broker. Support structures in place for sites: –mailing list, web pages, installation server, regular phone conferences, Bugzilla. Now ready for more sites to get on board...
Similar presentations
© 2024 SlidePlayer.com. Inc.
All rights reserved.