R-GMA

1
R-GMA DataGrids Monitoring System 1/7/2003
Werner Nutt (Heriot-Watt
University) ltw.nutt_at_hw.ac.ukgt
2
RGMA Relational Grid Monitoring
Architecture

• Grid Monitoring and Information System
  developed within DataGrid (Work Package
  3)
• Based on the Grid Monitoring Architecture
  of the Global
  Grid Forum
• Code is open source and freely available
  Homepage type wp3 into Google

3
Contributors

• Heriot-Watt, Edinburgh
• Andrew Cooke, Alasdair Gray, Lisha Ma, Werner
  Nutt
• IBM-UK
• James Magowan, Manfred Oevers, Paul Taylor
• Queen Mary, University of London
• Roney Cordenonsi
• CCLRC/PPARC
• Rob Byrom, Laurence Field, Steve Hicks, Manish
  Soni, Antony Wilson, Jason Leake
• Linda Cornwall, Abdeslem Djaoui, Steve Fisher,
  Robin Middleton
• SZTAKI, Hungary
• Peter Kacsuk, Norbert Podhorszki
• Trinity College Dublin
• Brian Coghlan, Stuart Kenny, David OCallaghan

4
Overview

• Grid monitoring Requirements
• The R-GMA approach A virtual monitoring
  database
• Components of R-GMA
• Schema
• Producers and Consumers
• Registry
• Republishers
• Query Planning

5
Major Components of DataGrid
6
WP7 R-GMA Collects Network Monitoring Data
7
The Grid Monitoring Problem

• In a Grid we have
• Computers
• Storage elements
• Network nodes and connections
• Application programmes,
• Monitoring
• What is the current state of the system?
• How did the system behave in the past ?

8
Monitoring Data Come in two Kinds

• A Grid monitoring system makes available two
  kinds of data
• static data pools, e.g., databases on
• network topology, nodes connected
• applications available (versions, licences, ...)
• streams of data, e.g.,
• sensor data (cpu load, network traffic, ...)
• Data streams may give rise to data pools if they
  are archived
• Today R-GMA is tailored towards streams,
• but not pools

9
Examples of Monitoring Queries

• Show me the (average) cpu-load of computers at
  Heriot-Watt!
• Between which nodes was yesterday the average
  transportation time for 1 MB packets higher than
  than 0. seconds?
• For every computing element CE, how many
  computers of CE have currently a cpu-load of no
  more than 30?

10
Grid Monitoring Requirements

• Support for publishing data pools and
  streams
• Support for locating data sources
  (automatic, if possible)
• Queries with different temporal interpretations
  (continuous,
  latest state, history)
• Scalability (there
  may be thousands of data sources)
• Resilience to failure
  (data sources may become unavailable)
• Flexibility (we dont know which queries
  will be posed)

11
Architecture Approach 1 A Monitoring Data
Warehouse

• Idea
• store all data about the Grid status into a huge
  database
• and query it
• Not realistic
• Loading takes time
• Data occupy space
• Connections to the warehouse may fail
• Often monitoring data flow as data streams, and
  queries ask for data streams as output

12
Approach 2 Monitoring with a
Multi-agent System

• The Grid Monitoring Architecture (GMA) of the
  Global Grid Forumdistinguishes between

• Consumers of information
• Producers of information
• Directory Service
• Producers register their supply
• Consumers register their demand

Directory Service mediates between producers and
consumers
13
Questions about GMA

• Which kinds of producers and consumers are there?
• In which language do producers register their
  supply
  and consumers their demand ?
• What is the meaning of a registration?
• How does a consumer find suitable producers?
  And how does a producer find suitable
  consumers?
• Producers have different capabilities to answer
  queries
  (e.g. selections,
  joins, ). Which of them should they register?

14
R-GMA A Virtual Monitoring Data Warehouse

• Language of producers and consumers
  relational queries (SQL)
• Vocabulary Relations in a global schema

• Consumer poses queries over
  global schema
• Producer

• has a type (stream p., database p.)
• publishes relations R1, ,Rk
• for every R, registers a simple view V on the
  global schema

15
Schema Contributions
CPULoad (Global Schema) CPULoad (Global Schema) CPULoad (Global Schema) CPULoad (Global Schema) CPULoad (Global Schema)
Country Site Facility Load Timestamp
UK RAL CDF 0.3 19055711022002
UK RAL ATLAS 1.6 19055611022002
UK GLA CDF 0.4 19055811022002
UK GLA ALICE 0.5 19055611022002
CH CERN ALICE 0.9 19055611022002
CH CERN CDF 0.6 19055511022002
CPULoad (Stream Producer 2) CPULoad (Stream Producer 2) CPULoad (Stream Producer 2) CPULoad (Stream Producer 2) CPULoad (Stream Producer 2)
UK GLA CDF 0.4 19055811022002
UK GLA ALICE 0.5 19055611022002
CPULoad (Stream Producer 1) CPULoad (Stream Producer 1) CPULoad (Stream Producer 1) CPULoad (Stream Producer 1) CPULoad (Stream Producer 1)
UK RAL CDF 0.3 19055711022002
UK RAL ATLAS 1.6 19055611022002
CPULoad (Stream Producer 3) CPULoad (Stream Producer 3) CPULoad (Stream Producer 3) CPULoad (Stream Producer 3) CPULoad (Stream Producer 3)
CH CERN ATLAS 1.6 19055611022002
CH CERN CDF 0.6 19055511022002
16
Contributions are Views
CPULoad (Producer 1) CPULoad (Producer 1) CPULoad (Producer 1) CPULoad (Producer 1) CPULoad (Producer 1)
UK RAL CDF 0.3 19055711022002
UK RAL ATLAS 1.6 19055611022002
SELECT FROM cpuLoad WHERE country UK AND
site RAL
CPULoad (Producer 2) CPULoad (Producer 2) CPULoad (Producer 2) CPULoad (Producer 2) CPULoad (Producer 2)
UK GLA CDF 0.4 19055811022002
UK GLA ALICE 0.5 19055611022002
SELECT FROM cpuLoad WHERE country UK AND
site GLA
17
Keys in the Global Schema

• Network throughput
• tp(src, dest, method, pcktSize, timestamp,
  time)
• Intuitively, tp has the primary key
• (src, dest, method, pcktSize, timestamp).
• We need to know the primary keys
• to understand the global schema
• to answer latest snapshot queries
• Primary keys are declared, but not enforced!
• Although, sometimes they hold globally if
  they hold locally !
  

18
Metaphor Roles and Agents

• R-GMA Clients Grid components or Grid
  applications
• Clients can play the roles of producers or
  consumers
• A client would need special capabilities for a
  role
• Clients are supported in their roles by agents
• Implementation
• APIs for client roles new
  StreamProducer()
• Agents are objects on a Web server

19
Primary Producers

• Database producer
• supports queries over fixed set of tuples (static
  queries)
• can be used to publish a database
• Stream producer
• supports queries over changing set of tuples
  
  (continuous queries)
• supports latest snapshot queries
• offers up-to-date values for each primary key in
  a db
• Today DatabaseProducers and StreamProducers
  in R-GMA are different from the
  above!

20
Communication Modes of Stream Producers

• Stream Producers may offer two communication
  modes for continuous queries
• lossless ( but tuples could become stale)
• lossy ( but tuples are fresh)

Today R-GMAs StreamProducers are resilient and
support lossless communication
21
Republishers Publish Query Answers

• Archiver shows the history of a stream.
• Stream Republisher enables
• merging,
• thinning,
• summarising of streams

22
Republishers in R-GMA Today

• Republishers are called archivers
  (although some of them don't archive
  anything)
• An archiver ( republisher)
• is defined by a query
• consumes only from stream producers
• publishes the query result according to its
  type, using
• a stream producer, or
• a latest snapshot producer, or
• a database producer (which keeps an
  archive)
• Republishers are used to answer complex queries!

23
The Next Step Hierarchies of Stream Republishers
24
Republisher HierarchiesThe Issues

• Republishers are defined by querieshierarchies
  have to be maintained automatically
• new stream producers must only be added
  to republishers at
  lowest level
• hierarchy has to be replanned if a republisher
  fails
• difficult transition from one plan to the other
  without
  loss of tuples
• How well can we describe the content of a
  stream?Possibly need for descriptions that join
• stream relations CPULoad(machineID, load,
  timestamp)
• static relations locatedAt(machineID,
  site)

25
What is the Meaning of a Query in R-GMA?

• Assumption the views of (primary) producers are
  selections on a single relation, i.e., queries of
  the form
• SELECT
• FROM cpu_load
• WHERE machine_id AB123 AND loc hw
  (each producer contributes its parts
  of a relation)
• The virtual database contains the union of
  the data of all the primary producers
• Conceptually, a query is evaluated
• over the entire virtual db

26
Stream Queries can have Various Temporal
Interpretations

• Consider a query over the relation Transport
  Time
• tt(src, dest, pcktSize, method, timestamp, time)
• SELECT FROM tt
• WHERE src ral AND dest bologna
• What is meant? Measurements
• from now ? (Continuous Query)
• up until now ? (History Query)
• right now ? (Latest Snapshot Query)
• Today Queries can
  be flagged with their type

27
Advanced Queries Mixing Temporal Query Types

• Which connections have currently a
  transportation time that is higher than last
  week's average?
  (latest snapshot and history)
• Show me the cpu load of those machines where it
  is lower than yesterday's load average!
• 
  (continuous and history)
• We do not intend to support such
  queries by R-GMA!

28
In R-GMA Query Answering Needs Mediation

• Suppose P1, P2 publish for tp (throughput)
• P1 WHERE src hw
• P2 WHERE src ral AND pcktSize gt 20
• A global consumer poses its query over global
  relations
• SELECT FROM tp WHERE pcktSize gt 10
• A mediator translates this into queries over
  local relations
• SELECT FROM P1.tp WHERE pcktSize gt 10
• UNION
• SELECT FROM P2.tp
• Today R-GMAs mediator handles simple queries
  like the one above

29
Global and Local Consumers

• Global consumers pose queries over global
  relations
• SELECT FROM tp WHERE pcktSize gt 10 ,
• which are translated into queries over local
  relations
• SELECT FROM P1.tp WHERE pcktSize gt 10
• UNION
• SELECT FROM P2.tp
• Local consumers pose queries over local
  relations directly
• SELECT FROM P1.tp WHERE method ping
• Today a consumer can be global or local,
• but local relations cannot be
  referred to explicitly

30
How does the Mediator Find Suitable Publishers?

• P1, P2, P3 publish for tt (Transport Time)
• P1 src hw
• P2 src ral AND pcktSize gt 20
• P3 src ral AND method ping
• Q SELECT FROM tt WHERE src ral AND method
  ping
• We see P1 is not suitable for Q, but P2 and P3
  are. Why?
• src hw AND src ral AND method ping
  is never true
• src ral AND pcktSize gt 20 AND
  is sometimes true
• Satisfiability
  Test!
• Today implemented

31
So Which Publishers Should the Mediator
Ask?

• P2 src ral AND pcktSize gt 20
• P3 src ral AND method ping
• Q SELECT FROM tt WHERE src ral AND
  method ping
• All answers to Q returned by P2 are also returned
  by P3
• whenever
• src ral AND pcktSize gt 20 AND src ral AND
  method ping
• is true, then
• src ral AND method ping AND src ral AND
  method ping
• is true.
• Hence, R-GMA only needs to ask P3
• Entailment
  Test!
• Needed for Republisher Hierarchies!
  (not yet implemented)

32
But What Did the Producers Promise?

• P registers view V
• Does P promise
• some of V ? (sound description)
• all of V? (sound and
  complete description)
• The Entailment Test only makes sense when the
  registered views are sound and complete
  descriptions
• Producers should register completeness flags

33
Why May a Producer not be Complete?

• The language of views is more restricted than the
  language of queriesHence republishers may be
  unable to say exactly what they
  publish
• Archivers may archive in lossy mode
• Producers may lose tuples
• A producer may not know everything
  about
  the real world
• 
  Open to debate

34
Summary (1)

• Monitoring data come in Pools and Streams
• Global Schema
• primary keys
• Types of Stream Queries
• continuous vs. history vs. latest snapshot
• Producers
• DB producers publish database
• stream producers lossless vs.
  lossy communication modes