DUROC DynamicallyUpdated Request Online Coallocator

1
DUROC (Dynamically-Updated Request Online
Co-allocator)

• August 8, 2002
• MPI Seminar
• Distributed Computing Communication Laboratory
• Jiyoung song

2
Introduction of DUROC

• What is DUROC ?
• Dynamically-Updated Request Online Co-allocator
• Bring up the distributed pieces of the jobs
• Interface to obtain resources and execute jobs
  across multiple management pools
• Similar to GRAM, with addition of the subjob-add,
  subjob-delete, and barrier-release operation for
  managing resources

3
Globus components
4
Structure of duroc runtime module
Duroc_runtime_module
Duct_runtime_module
Gram_myjob_module
Nexus_module
Globus_io_module
5
Flow of duroc_runtime_module initialization
Globus_duroc_runtime_activate()
s_duroc_runtime_ activate()
s_intra_subjob_init()
s_inter_subjob_init()
s_inter_subjob_duct_init()
globus_duct_runtime_init()
Globus_duroc_runtime_activate()
6
Globus_duroc_runtime_activate()

• Call s_duroc_runtime_activate ()
• Error handling of the call

7
S_duroc_runtime_activate()

• Globus_module_activate()
• GLOBUS_COMMON_MODULE
• GLOBUS_THREAD_MODULE
• GLOBUS_NEXUS_MODULE
• GLOBUS_DUCT_RUNTIME_MODULE
• GLOBUS_GRAM_MYJOB_MODULE
• Call s_intra_subjob_init ()
• Call s_inter_subjob_init ()

8
S_intra_subjob_init()

• Call globus_hashtable_init()
• Create the hash table whose name is
  s_tagged_gram_myjob_hasht

9
S_inter_subjob_init()

• Check gram_rank
• If gram_rank 0 then
• Create the hash table whose name is
  s_inter_subjob_tagged_duct_hasht
• Call s_inter_subjob_duct_init()
• Else
• just return 0

10
DUROC
DUROC
CO-ALLOCATION
Subjob_size4
Subjob_size2
Subjob_size2
Subjob_size4
SUBJOB_INDEX0
SUBJOB_INDEX1
SUBJOB_INDEX2
SUBJOB_INDEX3
0
0
0
0
1
1
1
1
2
2
3
3
Grid4.sogang.ac.kr
Grid1.sogang.ac.kr
Grid2.sogang.ac.kr
Grid3.sogang.ac.kr
11
S_inter_subjob_duct_init()

• Get the environmental variables
• GLOBUS_DUROC_DUCT_CONTACT, GLOBUS_DUROC_DUCT_ID
• Initialize the fifo queue
• s_inter_subjob_duct_fifo
• Initialize mutex and cond
• s_inter_subjob_duct_mutex
• Call globus_duct_runtime_init()
• s_inter_subjob_duct_cond

12
Globus_duct_runtime_init()

• Get the startpoint to the server (maybe
  subjob_index0, gram_rank0)
• globus_duct_runtime_make_startpoint()
• url or LSP(linearized startpoint) depend on
  (const char ) GLOBUS_DUROC_DUCT_CONTACT
• Use globus_io module
• Set the argument
• globus_duct_runtime_t s_inter_subjob_duct_runtime
• Data_port, config_port, callback function
• Make the endpoint(data_port, config_port) to
  communicate
• Bind sp and ep
• Send the rsr(checkin message) to the server
• Handler id CHECKIN_MSG_ID

13
Other functions of duroc runtime module

• globus_duroc_runtime_intra_subjob_rank()
• globus_duroc_runtime_intra_subjob_size()
• globus_duroc_runtime_inter_subjob_structure()
• globus_duroc_runtime_inter_subjob_send()
• globus_duroc_runtime_inter_subjob_receive()
• globus_duroc_runtime_intra_subjob_send()
• globus_duroc_runtime_intra_subjob_receive()

14
globus_duroc_runtime_intra_subjob_rank()

• Obtain the rank of the local subjob process
• Wrapper function
• gram_myjob_rank()
• Error handling

15
Globus_duroc_runtime_intra_subjob_size()

• Obtain the size of the local subjob process
• Wrapper function
• gram_myjob_size()
• Error handling

16
globus_duroc_runtime_inter_subjob_structure()

• The DUROC inter-subjob communication routines can
  only be called on the subjob node where
  globus_duroc_runtime_intra_subjob_rank() reports
  the rank as zero
• Arguments
• Int local_addressp
• Int remote_addressesp
• Initialize local_addressp with the local
  subjob's communication address
• Initialize remote_addressesp with a
  freshly-allocated array containing the remote
  subjobs' communication addresses

17
Flow of globus_duroc_runtime_inter_subjob_structur
e()
globus_duroc_runtime_inter_subjob_structure()
s_inter_subjob_duct_structure()
globus_duct_runtime_structure()
static globus_duct_runtime_t s_inter_subjob_duct_r
untime In globus_duroc_runtime.c
18
globus_duroc_runtime_intra_subjob_send()

• Send a byte-vector to another process in the
  DUROC subjob
• Make the message the form of GLOBUS_DUROC_RUNTIME_
  INTRA_SEND_PROTOCOL_VERSION
• Call gram_myjob_send()
• Arguments
• Int dst_rank
• the rank of the destination process
• Cont char tag
• a nul-terminated string which must match that
  provided to the receive call on the destination
  process
• Example "globus_duroc_runtime run status"
• Globus_byte_t msg

19
globus_duroc_runtime_intra_subjob_receive()

• Receive a byte-vector sent by another process in
  the DUROC subjob
• Messages are queued and reordered if the process
  receives messages with a different tag than the
  one requested by the receiving call (by hash
  table look up)
• Call gram_myjob_receive()

20
globus_duroc_runtime_inter_subjob_send()

• Send a byte-vector to another subjob in the DUROC
  job
• The DUROC inter-subjob communication routines can
  only be called on the subjob node where
  globus_duroc_runtime_intra_subjob_rank() reports
  the rank as zero

21
globus_duroc_runtime_inter_subjob_send()
s_inter_subjob_duct_send()
globus_duct_runtime_send()
22
globus_duct_runtime_send()

• Look up the hash table
• Hash table runtimep-gtremote_data_spst, key
  dst_addr
• Obtain the startpoint to communicate
• Sent the RSR(message) to the context
• Handler id DATA_MSG_ID

23
globus_duroc_runtime_inter_subjob_receive()

• The same as globus_duroc_runtime_intra_subjob()
  except the hash table name
• s_inter_subjob_tagged_duct_hasht

24
(No Transcript)
25
(No Transcript)
26
(No Transcript)