ARSITEKTUR DBMS TERDISTRIBUSI

1
ARSITEKTUR DBMS TERDISTRIBUSI
2
STANDARISASI DBMS

• Berdasarkan Komponen.
• Komponen dari sistem didefinisikan bersama
  dengan keterkaitan antar komponen. Suatu DBMS
  terdiri dari sejumlah komponen, masing-masing
  menyediakan beberapa fungsi.
• Berdasarkan Fungsi.
• Kelas-kelas yang berbeda dari pengguna
  diidentifikasi dan fungsi bahwa sistem akan
  melakukan untuk masing-masing kelas
  didefinisikan. Spesifikasi Sistem dalam kategori
  ini biasanya menentukan struktur hirarki untuk
  kelas pengguna.

3
STANDARISASI DBMS

• Berdasarkan Data.
• Jenis data yang berbeda diidentifikasi, dan
  sebuah kerangka kerja arsitektur ditentukan yang
  mendefinisikan unit fungsional yang akan
  menyadari atau menggunakan data sesuai dengan
  pandangan yang berbeda. Pendekatan (juga disebut
  sebagai pendekatan datalogical) diklaim menjadi
  pilihan lebih baik untuk kegiatan standardisasi.

4
STANDARISASI DBMSARSITEKTUR ANSI / SPARC

• ANSI / SPARC arsitektur diklaim didasarkan pada
  data organisasi. Ia mengakui tiga tampilan
  datatampilan eksternal, yang adalah bahwa dari
  pengguna, yang mungkin programmer, pandangan
  internal, bahwa dari sistem atau mesin dan
  pandangan konseptual, yaitu perusahaan.Untuk
  masing-masing pandangan, definisi skema yang
  tepat diperlukan.

5
STANDARISASI DBMSARSITEKTUR ANSI / SPARC
6
STANDARISASI DBMSARSITEKTUR ANSI / SPARC

• Pada tingkat terendah arsitektur adalah pandangan
  internal, yang berkaitan dengan definisi fisik
  dan organisasi data.Pada ekstrem yang lain
  adalah pandangan eksternal, yang berkaitan dengan
  bagaimana para pemakai memandang
  database.Antara kedua ujung adalah skema
  konseptual, yang merupakan definisi abstrak dari
  database. Ini adalah "dunia nyata" pandangan dari
  perusahaan yang dimodelkan dalam database.

7
STANDARISASI DBMSARSITEKTUR ANSI / SPARC
8
STANDARISASI DBMSARSITEKTUR ANSI / SPARC

• Kotak-kotak persegi merupakan fungsi pengolahan,
  sedangkan segi enam adalah peran administratif.
• Tanda panah menunjukkan data, perintah, program,
  dan aliran deskripsi, sedangkan "Aku" bar
  berbentuk pada mereka merupakan antarmuka.
• Komponen utama yang memungkinkan pandangan
  pemetaan antara data yang berbeda organisasi
  adalah data dictionary / directory (digambarkan
  sebagai segitiga), yang merupakan meta-database.
• Database administrator bertanggung jawab untuk
  menentukan definisi skema internal.
• Peran perusahaan administrator adalah untuk
  mempersiapkan definisi skema konseptual.
• Administrator aplikasi bertanggung jawab untuk
  mempersiapkan skema eksternal untuk aplikasi.

9
STANDARISASI DBMSARSITEKTUR ANSI / SPARC

• Sistem ditandai sehubungan dengan(1) otonomi
  sistem lokal,(2) distribusi,(3) heterogenitas.

10
MODEL ARSITEKTUR UNTUK DISTRIBUSI DBMS Otonomi

• Otonomi mengacu pada distribusi kontrol, tidak
  ada data. Hal ini menunjukkan sejauh mana DBMSs
  individu dapat beroperasi secara
  independen.Tiga alternatifketat
  integrasisemiautonomous sistemisolasi total

11
MODEL ARSITEKTUR UNTUK DISTRIBUSI DBMS Otonomi

• Tight integrasi.Gambar-tunggal seluruh database
  tersedia untuk setiap pengguna yang ingin berbagi
  informasi, yang dapat berada di beberapa
  database. Dari sudut pandang pengguna, data
  secara logis terpusat dalam satu database.
• Semiautonomous sistem.The DBMSs dapat beroperasi
  secara independen. Masing-masing DBMSs menentukan
  bagian mana dari database mereka sendiri, mereka
  akan membuat diakses pengguna DBMSs lain.
• Total isolasi.Sistem DBMSs individu yang berdiri
  sendiri, yang tidak mengetahui tentang keberadaan
  DBMSs lain atau bagaimana berkomunikasi dengan
  mereka.

12
MODEL ARSITEKTUR UNTUK DISTRIBUSI DBMS Otonomi

• Distribusi mengacu pada distribusi data. Tentu
  saja, kita sedang mempertimbangkan distribusi
  fisik data melalui beberapa situs, pengguna
  melihat data sebagai salah satu kolam renang
  logis.Dua alternatifclient / server
  distribusipeer-to-peer distribusi (distribusi
  penuh)

13
MODEL ARSITEKTUR UNTUK DISTRIBUSI DBMS
Distribusi

• Client / distribusi server.Klien / server
  distribusi konsentrat tugas manajemen data pada
  server sedangkan klien fokus pada penyediaan
  lingkungan aplikasi termasuk user interface.
  Tugas komunikasi yang dibagi antara mesin klien
  dan server. Client / server DBMSs merupakan usaha
  pertama di mendistribusikan fungsionalitas.
• Peer-to-peer distribusi.Tidak ada perbedaan dari
  mesin klien versus server. Setiap mesin memiliki
  fungsionalitas penuh DBMS dan dapat berkomunikasi
  dengan mesin lainnya untuk mengeksekusi query dan
  transaksi.

14
MODEL ARSITEKTUR UNTUK DISTRIBUSI DBMS
Heterogenitas

• Heterogenitas dapat terjadi dalam berbagai
  bentuk dalam sistem terdistribusi, mulai bentuk
  heterogenitas perangkat keras dan perbedaan dalam
  jaringan protokol untuk variasi dalam manajer
  data.Mewakili data dengan alat pemodelan yang
  berbeda menciptakan heterogenitas karena kekuatan
  ekspresif yang melekat dan keterbatasan model
  data individu. Heterogenitas dalam bahasa query
  tidak hanya melibatkan penggunaan paradigma yang
  sama sekali berbeda akses data dalam model data
  yang berbeda, tetapi juga mencakup perbedaan
  dalam bahasa bahkan ketika sistem individu
  menggunakan model data yang sama.

15
ARCHITECTURAL MODELS FOR DISTRIBUTED DBMSs -
ALTERNATIVES

• The dimensions are identified as A (autonomy),
  D (distribution) and H (heterogeneity).
• The alternatives along each dimension are
  identified by numbers as 0, 1 or 2.
• A0 - tight integration D0 - no distribution
• A1 - semiautonomous systems D1 - client /
  server systems
• A2 - total isolation D2 - peer-to-peer systems
• H0 - homogeneous systems
• H1 - heterogeneous systems

16
ARCHITECTURAL MODELS FOR DISTRIBUTED DBMSs -
ALTERNATIVES

• (A0, D0, H0)
• If there is no distribution or heterogeneity,
  the system is a set of multiple DBMSs that are
  logically integrated.
• (A0, D0, H1)
• If heterogeneity is introduced, one has multiple
  data managers that are heterogeneous but provide
  an integrated view to the user.
• (A0, D1, H0)
• The more interesting case is where the database
  is distributed even though an integrated view of
  the data is provided to users (client / server
  distribution).

17
ARCHITECTURAL MODELS FOR DISTRIBUTED DBMSs -
ALTERNATIVES

• (A0, D2, H0)
• The same type of transparency is provided to the
  user in a fully distributed environment. There is
  no distinction among clients and servers, each
  site providing identical functionality.
• (A1, D0, H0)
• These are semiautonomous systems, which are
  commonly termed federated DBMS. The component
  systems in a federated environment have
  significant autonomy in their execution, but
  their participation in the federation indicate
  that they are willing to cooperate with other in
  executing user requests that access multiple
  databases.

18
ARCHITECTURAL MODELS FOR DISTRIBUTED DBMSs -
ALTERNATIVES

• (A1, D0, H1)
• These are systems that introduce heterogeneity
  as well as autonomy, what we might call a
  heterogeneous federated DBMS.
• (A1, D1, H1)
• System of this type introduce distribution by
  pacing component systems on different machines.
  They may be referred to as distributed,
  heterogeneous federated DBMS.
• (A2, D0, H0)
• Now we have full autonomy. These are
  multidatabase systems (MDBS). The components have
  no concept of cooperation. Without heterogeneity
  and distribution, an MDBS is an interconnected
  collection of autonomous databases.

19
ARCHITECTURAL MODELS FOR DISTRIBUTED DBMSs -
ALTERNATIVES

• (A2, D0, H1)
• These case is realistic, maybe even more so than
  (A1, D0, H1), in that we always want to built
  applications which access data from multiple
  storage systems with different characteristics.
• (A2, D1, H1) and (A2, D2, H1)
• These two cases are together, because of the
  similarity of the problem. They both represent
  the case where component databases that make up
  the MDBS are distributed over a number of sites -
  we call this the distributed MDBS.

20
DISTRIBUTED DBMS ARCHITECTURE

• Client / server systems - (Ax, D1, Hy)
• Distributed databases - (A0, D2, H0)
• Multidatabase systems - (A2, Dx, Hy)

21
DISTRIBUTED DBMS ARCHITECTURECLIENT / SERVER
SYSTEMS

• This provides two-level architecture which make
  it easier to manage the complexity of modern
  DBMSs and the complexity of distribution.
• The server does most of the data management work
  (query processing and optimization, transaction
  management, storage management).
• The client is the application and the user
  interface (management the data that is cached to
  the client, management the transaction locks).

22
DISTRIBUTED DBMS ARCHITECTURECLIENT / SERVER
SYSTEMS

• This architecture is quite common in relational
  systems where the communication between the
  clients and the server(s) is at the level of SQL
  statements.

23
DISTRIBUTED DBMS ARCHITECTURECLIENT / SERVER
SYSTEMS

• Multiple client - single server
• From a data management perspective, this is not
  much different from centralized databases since
  the database is stored on only one machine (the
  server) which also hosts the software to manage
  it. However, there are some differences from
  centralized systems in the way transactions are
  executed and caches are managed.
• Multiple client - multiple server
• In this case, two alternative management
  strategies are possible either each client
  manages its own connection to the appropriate
  server or each client knows of only its home
  server which then communicates with other
  servers as required.

24
DISTRIBUTED DBMS ARCHITECTUREPEER-TO-PEER
DISTRIBUTED SYSTEMS

• The physical data organization on each machine
  may be different.
• Local internal scheme (LIS) - is an individual
  internal schema definition at each site.
• Global conceptual schema (GCS) - describes the
  enterprise view of the data.
• Local conceptual schema (LCS) - describes the
  logical organization of data at each site.
• External schemas (ESs) - support user
  applications and user access to the database.

25
DISTRIBUTED DBMS ARCHITECTUREPEER-TO-PEER
DISTRIBUTED SYSTEMS
26
DISTRIBUTED DBMS ARCHITECTUREPEER-TO-PEER
DISTRIBUTED SYSTEMS

• In these case, the ANSI/SPARC model is extended
  by the addition of global directory / dictionary
  (GD/D) to permits the required global mappings.
  The local mappings are still performed by local
  directory / dictionary (LD/D). The local database
  management components are integrated by means of
  global DBMS functions. Local conceptual schemas
  are mappings of global schema onto each site.

27
DISTRIBUTED DBMS ARCHITECTUREPEER-TO-PEER
DISTRIBUTED SYSTEMS

• The detailed components of a distributed DBMS.
• Two major components
• user processor
• data processor

28
DISTRIBUTED DBMS ARCHITECTUREPEER-TO-PEER
DISTRIBUTED SYSTEMS

• User processor
• user interface handler - is responsible for
  interpreting user commands as they come in, and
  formatting the result data as it is sent to the
  user,
• semantic data controller - uses the integrity
  constraints and authorizations that are defined
  as part of the global conceptual schema to check
  if the user query can be processed,
• global query optimizer and decomposer -
  determines an execution strategy to minimize a
  cost function, and translates the global queries
  in local ones using the global and local
  conceptual schemas as well as global directory,
• distributed execution monitor - coordinates the
  distributed execution of the user request.

29
DISTRIBUTED DBMS ARCHITECTUREPEER-TO-PEER
DISTRIBUTED SYSTEMS

• Data processor
• local query optimizer - is responsible for
  choosing the best access path to access any data
  item,
• local recovery manager - is responsible for
  making sure that the local database remains
  consistent even when failures occur,
• run-time support processor - physically accesses
  the database according to the physical commands
  in the schedule generated by the query optimizer.
  This is the interface to the operating system and
  contains the database buffer (or cache) manager,
  which is responsible for maintaining the main
  memory buffers and managing the data accesses.

30
DISTRIBUTED DBMS ARCHITECTUREMDBS ARCHITECTURE

• Models using a Global Conceptual Schema (GCS)
• The GCS is defined by integrating either the
  external schemas of local autonomous databases or
  parts of their local conceptual schemas.
  If the heterogeneity exists in the system, then
  two implementation alternatives exists unilingual
  and multilingual.
• Models without a Global Conceptual Schema (GCS)
• The existence of a global conceptual schema in a
  multidatabase system is a controversial issue.
  There are researchers who even define a
  multidatabase management system as one that
  manages several databases without the global
  schema.

31
DISTRIBUTED DBMS ARCHITECTUREMDBS ARCHITECTURE -
models using a GCS
32
DISTRIBUTED DBMS ARCHITECTUREMDBS ARCHITECTURE -
models using a GCS

• A unilingual multi-DBMS requires the users to
  utilize possibly different data models and
  languages when both a local database and the
  global database are accessed.
• Any application that accesses data from multiple
  databases must do so by means of an external view
  that is defined on the global conceptual schema.
• One application may have a local external schema
  (LES) defined on the local conceptual schema as
  well as a global external schema (GES) defined on
  the global conceptual schema.

33
DISTRIBUTED DBMS ARCHITECTUREMDBS ARCHITECTURE -
models using a GCS

• An alternative is multilingual architecture,
  where the basic philosophy is to permit each user
  to access the global database by means of an
  external schema, defined using the language of
  the users local DBMS.
• The multilingual approach obviously makes
  querying the databases easier from the users
  perspective. However, it is more complicated
  because we must deal with translation of queries
  at run time.

34
DISTRIBUTED DBMS ARCHITECTUREMDBS ARCHITECTURE -
models without a GCS
35
DISTRIBUTED DBMS ARCHITECTUREMDBS ARCHITECTURE -
models without a GCS

• The architecture identifies two layers the local
  system layer and the multidatabase layer on top
  of it.
• The local system layer consists of a number of
  DBMSs, which present to the multidatabase layer
  the part of their local database they are willing
  to share with users of the other databases. This
  shared data is presented either as the actual
  local conceptual schema or as a local external
  schema definition.
• The multidatabase layer consist of a number of
  external views, which are constructed where each
  view may be defined on one local conceptual
  schema or on multiple conceptual schemas. Thus
  the responsibility of providing access to
  multiple databases is delegated to the mapping
  between the external schemas and the local
  conceptual schemas.

36
DISTRIBUTED DBMS ARCHITECTUREMDBS ARCHITECTURE -
models without a GCS

• The MDBS provides a layer of software that runs
  on top of these individual DBMSs and provides
  users with the facilities of accessing various
  databases.
• Fig. represents a nondistributed multi-DBMS. If
  the system is distributed, we would need to
  replicate the multidatabase layer to each site
  where there is a local DBMS that participates in
  the system.

37
DISTRIBUTED DBMS ARCHITECTUREGLOBAL DIRECTORY
ISSUE

• The global directory includes information about
  the location of the fragments as well as the
  makeup of the fragments.
• The directory is itself a database that contains
  meta-data about the actual data stored in the
  database.
• We have three dimensions
• 1.type 2.location 3.replication

38
DISTRIBUTED DBMS ARCHITECTUREGLOBAL DIRECTORY
ISSUE

• Type
• A directory maybe either global to the entire
  database or local to each site. In other words,
  there might be a single directory containing
  information about all the data in the database,
  or a number of directories, each containing the
  information stored at one site.
• Location
• The directory maybe maintained centrally at one
  site, or in a distributed fashion by distributing
  it over a number of sites.
• Replication
• There maybe a single copy of the directory or
  multiply copies.

39
DISTRIBUTED DBMS ARCHITECTUREGLOBAL DIRECTORY
ISSUE

• These three dimensions are orthogonal to one
  another. The unrealistic combination have been
  designed by a question mark.

40
DISTRIBUTED DATABASE DESIGN
41
DISTRIBUTED DATABASE DESIGN

• The organization of distributed systems can be
  investigated along three orthogonal dimensions
• 1. Level of sharing
• 2. Behavior of access patterns
• 3. Level of knowledge on access pattern behavior

42
DISTRIBUTED DATABASE DESIGN

• Level of sharing
• no sharing - each application and its data
  execute at one site,
• data sharing - all the programs are replicated at
  all the sites, but data files are not,
• data plus program sharing - both data and
  programs may be shared.
• Behavior of access patterns
• static - access patterns of user requests do not
  change over time,
• dynamic - access patterns of user requests change
  over time.
• Level of knowledge on access pattern behavior
• complete information - the access patterns can
  reasonably be predicted and do not deviate
  significantly from the predictions,
• partial information - there are deviations from
  the predictions.

43
ALTERNATIVE DESIGN STRATEGIES

• Two major strategies that have been identified
  for designing distributed databases are
• the top-down approach
• the bottom-up approach

44
ALTERNATIVE DESIGN STRATEGIESTOP-DOWN DESIGN
PROCESS
45
ALTERNATIVE DESIGN STRATEGIESTOP-DOWN DESIGN
PROCESS

• view design - defining the interfaces for end
  users,
• conceptual design - is the process by which the
  enterprise is examined to determine entity types
  and relationships among these entities. One can
  possibly divide this process into to related
  activity groups
• entity analysis - is concerned with determining
  the entities, their attributes, and the
  relationships among these entities,
• functional analysis - is concerned with
  determining the fundamental functions with which
  the modeled enterprise is involved.

46
ALTERNATIVE DESIGN STRATEGIESTOP-DOWN DESIGN
PROCESS

• distributions design - design the local
  conceptual schemas by distributing the entities
  over the sites of the distributed system. The
  distribution design activity consists of two
  steps
• fragmentation
• allocation
• physical design - is the process, which maps the
  local conceptual schemas to the physical storage
  devices available at the corresponding sites,
• observation and monitoring - the results is some
  form of feedback, which may result in backing up
  to one of the earlier steps in the design.

47
ALTERNATIVE DESIGN STRATEGIESBOTTOM-UP DESIGN
PROCESS

• Top-down design is a suitable approach when a
  database system is being designed from scratch.
• If a number of databases already exist, and the
  design task involves integrating them into one
  database - the bottom-up approach is suitable for
  this type of environment. The starting point of
  bottom-up design is the individual local
  conceptual schemas. The process consists of
  integrating local schemas into the global
  conceptual schema.

48
DISTRIBUTION DESIGN ISSUESREASONS FOR
FRAGMENTATION

• The important issue is the appropriate unit of
  distribution. For a number of reasons it is only
  natural to consider subsets of relations as
  distribution units.
• If the applications that have views defined on a
  given relation reside at different sites, two
  alternatives can be followed, with the entire
  relation being the unit of distribution. The
  relation is not replicated and is stored at only
  one site, or it is replicated at all or some of
  the sites where the applications reside.
• The fragmentation of relations typically results
  in the parallel execution of a single query by
  dividing it into a set of subqueries that operate
  on fragments. Thus, fragmentation typically
  increases the level of concurrency and therefore
  the system throughput.

49
DISTRIBUTION DESIGN ISSUESREASONS FOR
FRAGMENTATION

• There are also the disadvantages of
  fragmentation
• if the application have conflicting requirements
  which prevent decomposition of the relation into
  mutually exclusive fragments, those applications
  whose views are defined on more than one fragment
  may suffer performance degradation,
• the second problem is related to semantic data
  control, specifically to integrity checking.

50
DISTRIBUTION DESIGN ISSUESFRAGMENTATION
ALTERNATIVES

• The are clearly two alternatives
• horizontal fragmentation
• vertical fragmentation
• The fragmentation may, of course, be nested. If
  the nestings are of different types, one gets
  hybrid fragmentation.

51
DISTRIBUTION DESIGN ISSUESDEGREE OF FRAGMENTATION

• The extent to which the database should be
  fragmented is an important decision that affects
  the performance of query execution.
• The degree of fragmentation goes from one
  extreme, that is, not to fragment at all, to the
  other extreme, to fragment to the level of
  individual tuples (in the case of horizontal
  fragmentation) or to the level of individual
  attributes (in the case of vertical
  fragmentation).

52
DISTRIBUTION DESIGN ISSUESCORRECTNESS RULES OF
FRAGMENTATION

• Completeness
• If a relation instance R is decomposed into
  fragments R1,R2, ..., Rn, each data item that can
  be found in R can also be found in one or more of
  Ris. This property is also important in
  fragmentation since it ensures that the data in a
  global relation is mapped into fragments without
  any loss.
• Reconstruction
• If a relation R is decomposed into fragments
  R1,R2, ..., Rn, it should be possible to define a
  relational operator ? such that
• R ?Ri, ? Ri?FR
• The reconstructability of the relation from its
  fragments ensures that constraints defined on the
  data in the form of dependencies are preserved.

53
DISTRIBUTION DESIGN ISSUESCORRECTNESS RULES OF
FRAGMENTATION

• Disjointness
• If a relation R is horizontally decomposed into
  fragments R1,R2, ..., Rn and data item di is in
  Rj, it is not in any other fragment Rk (k ? j).
  This criterion ensures that the horizontal
  fragments are disjoint. If relation R is
  vertically decomposed, its primary key attributes
  are typically repeated in all its fragments.
  Therefore, in case of vertical partitioning,
  disjointness is defined only on the nonprimary
  key attributes of a relation.

54
DISTRIBUTION DESIGN ISSUESALLOCATION ALTERNATIVES

• The reasons for replication are reliability and
  efficiency of read-only queries.
• Read-only queries that access the same data items
  can be executed in parallel since copies exist on
  multiple sites.
• The execution of update queries cause trouble
  since the system has to ensure that all the
  copies of the data are updated properly.
• The decisions regarding replication is a
  trade-off which depends on the ratio of the
  read-only queries to the update queries.

55
DISTRIBUTION DESIGN ISSUESALLOCATION ALTERNATIVES

• A nonreplicated database (commonly called a
  partitioned database) contains fragments that are
  allocated to sites, and there is only one copy of
  any fragment on the network.
• In case of replication, either the database
  exists in its entirety at each site (fully
  replicated database), or fragments are
  distributed to the sites in such a way that
  copies of a fragment may reside in multiple sites
  (partially replicated database).

56
DISTRIBUTION DESIGN ISSUESALLOCATION ALTERNATIVES
57
DISTRIBUTION DESIGN ISSUESINFORMATION
REQUIREMENTS

• The information needed for distribution design
  can be divided into four categories
• database information,
• application information,
• communication network information,
• computer system information.

58
DISTRIBUTION DESIGN ISSUES FRAGMENTATION

• Horizontal fragmentation partitions a relation
  along its tuples
• Two versions of horizontal fragmentation
• Primary horizontal fragmentation of relation is
  performed using predicates that are defined on
  that relation
• Derived fragmentation is the partitioning of
  relation that results from predicates being
  defined on another relation

59
DISTRIBUTION DESIGN ISSUES FRAGMENTATION

• Vertical fragmentation partitions a relation into
  a set of smaller relations so that many of users
  aplications will run on only one fragment
• Vertical fragmentation is inherently more
  complicated than horizontal partitioning

60
DISTRIBUTION DESIGN ISSUESALLOCATION

• Allocation problem
• there are set of fragments F F1, F2, ... , Fn
  and network consisiting of sites S S1, S2,
  ... , Sm on wich sets aplications Q q1, q2,
  ... , qq is running
• The allocation problem involves finding the
  optimal distribution of F to S

61
DISTRIBUTION DESIGN ISSUESALLOCATION

• One of important issues that need to be discussed
  is the definition of optimality
• The optimality can be defined with respects of
  two measures Dowdy and Foster, 1982
• Minimal cost. The cost consists of the cost of
  storing each Fi at the site Sj, the cost of
  quering Fi at Sj, the cost of updating Fi, at all
  sites it is stored, and cost of data
  comunication. The allocation problem,then,
  attempts to find an alocations scheme that
  minimizes cost function.

62
DISTRIBUTION DESIGN ISSUESALLOCATION

• Perfomance. The allocation strategy is designed
  to maintain a performance mertic. Two well-known
  are to minimize the response time and to maximize
  the system throughput at each site