Releasev2x

Notes for the v2.X series

Estimated timeline

• June '15 master branched to 2.0.0
• July '16 2.0.0 is released (yes 1 year and 1 month later!)
• September '16 2.0.1 release
• September 30, feature freeze for 2.1.0
• Late Oct./early November '16 2.1.0 release (driven by OSHEM 1.3 compliance, PMIX 2.0 embedded)
• Early '17 2.2.0 release

Must-have Features for 2.0.X

• Thread safety (MPI_THREAD_MULTIPLE) support
  • need to verify which BTLs are thread safe (via testing vs stating) (DONE)
  • need more testing (non-blocking collectives, one-sided, MPI I/O, etc.)
  • need to document what is not thread safe (DONE)
  • performance improvements when using MPI_THREAD_MULTIPLE (i.e., TEST/WAIT improvements) - may wait for a publication before committing (DONE)
• MPI-3.1 Compliance
  • Ticket #349 (MPI_Aint_add) - 2.0.X candidate (DONE)
  • Ticket #369 (same_disp_info key for MPI_Win_create) - 2.0.X candidate, maybe (DONE)
  • Ticket #273 (non blocking coll I/O, non trivial). This is dependent on moving libnbc core out of libnbc component. (DONE)
  • Ticket #404 (MPI_Aint_diff) - 2.0.X candidate (DONE)
  • Ticket #357 (MPI_Initialized, MPI_Query_thread, MPI_Thread_is_main) always thread safe (probably just verify with a test to see this is true now for OMPI thread models) (DONE)
• MPI-3 Errata Items
  • Ticket #428 - non-profile-able mpi routines
  • Ticket #419 - neighborhood coll can’t handle non-symmetric general graphs, improve error checking
  • Ticket #424 - check fortran9X/F08 interface for MPI_IMPROBE (DONE)
  • Ticket #415, Ticket #345, Ticket #388 (DONE), Ticket #390 more fortran interface stuff
• Coverity cleanup (IN PROGRESS, down to ~260) (never ending)
• Scalable startup work (smarter add_proc in the OB1 PML), needs more work (DONE)
  • Sparse groups (DONE)
  • Additional PMIx features (issue 394)
• ROMIO refresh - need to be using a released ROMIO package (DONE)
• Fix Java bindings garbage collection issues (DONE)
• Hwloc 1.11.3 final (DONE)
• CUDA extension (to add MPIX_CUDA_IS_AWESOME to mpi.h) and MPI_T Cvar for run-time query of whether CUDA is supported in this OMPI (DONE)
• Add MPI 3 features to Java bindings (DONE)

Must-have Features for the 2.1 Series

• PMIx 2.0 integration (Intel, MLNX) - partially in master Issue #2072. Will require porting to the v2.x branch. (this may not happen, may use PMIx 1.2 instead) - PR to bring PMIx v1.2 into OMPI v2.1
• OpenSHMEM 1.3 compliance (MLNX) - issue #2109 tracking multiple PRs (DONE)
• mpool/rcache rewrite (LANL) - already in master (DONE)
• HCOLL datatypes (MLNX) - already in master (DONE)
• Better CMA support detection / make CMA the default in vader (LANL) - [PR #2216] (https://github.com/open-mpi/ompi/pull/2216)
• BTL/OpenIB across different subnets PR #1043 (DONE)
• usNIC BTL thread safety PR #1326 and libfabric 1.4 support (DONE)
• Lustre performance PR #987 (DONE)
• mpirun command line options ehancements PR #1317 (DONE)
• pending PRs (Nathan's free list work) (DONE)
• Multi-rail performance in OB1? What happened? (WONT FIX)
• Support for thread-based asynchronous progress for BTLs (DONE)
• Improved story on out-of-the-box performance, particularly for collectives. Ideally some kind of auto-tune type of mechanism. (otopo project) (kind of DONE, not otopo)

Desirable-to-have Features for 2.1 (but may go in to a 2.2)

• Awesome bug fixes from IBM
• direct AMO support in OSC/RDMA (compartmentalized) - partially in master (DONE)
• TCP latency went up and bandwidth(rendezvous) went way down. What happened? Maybe in 2.x series sometime... (AWS)
• better CUDA support (UTK/Nvidia working on this, BTL, ob1 specific, follow-on for a collectives)
• MPI I/O support MPI_THREAD_MULTIPLE (IBM, UH)
• New coll tuned component (UTK)
• Missing MPI I/O routines in ROMIO: MPI_File_iread_all, etc. (relies on generalized request MPICH-specific extensions) (IBM, UH)
• Configury enhancements for modified library names, sonames, etc. (IBM) (DONE) PR #2139

Desirable-to-have Features for 2.1/2.2 (but may go in to a future release)

• Rationalized configuration for Cray XE/XC (DONE)
• platform file for using OFI MTL on Cray XC/KNL
• usNIC stuff
  • conversion to libfabric (DONE)
• simplified verbs BTL for iWarp? (NOT GOING TO HAPPEN)
• Mellanox stuff (see above)
• OMPI commands (mpirun, orte_info, etc.): deprecate all single-dash options except for the sacrosanct ones (-np, etc.). Print a stderr warning for all the deprecated options.
  • Note that MPI-3.1 8.8 mpiexec mentions: -soft, -host, -arch, -wdir, -path, -file (MOVED TO 3.x)

Score-P integration (won't hit 2.0.0, but will get in 2.x) (MOVED TO 3.x)

libfabric support (Intel MTL, Cisco BTL, others) (DONE in 1.10)

Memkind support both for MPI_Alloc_mem and Open MPI internal (DONE)

• Note that there is not the right info infrastructure to allow apps to use memkind via MPI_alloc_mem

Nathan Hjelmn's BTL 3.0 changes (DONE)

MPI-4 features (maybe as extensions?)

• endpoints proposal (3.0 or later)
• ULFM (as of June 2015, Ralph/George are coordinating so that ORTE can give ULFM what it needs) (Not for 2.1)
• MPI T extensions - event stuff to replace Peruse (3.0)

Better interop with OpenMP 4 placement - esp. for nested OMP parallelism

OFI MTL support MPI_THREAD_MULTIPLE - may already be thread safe

OFI OSC component (probably will not happen)

Issue with pipelined algorithms in collectives don't comply with MPI standard (placeholder) see [Issue #1763] (https://github.com/open-mpi/ompi/issues/1763)

multi-target configury

• install of both thread safe/thread single with a single configure option

Features that are already in master that made it in to v2.0

• Switch to using OMPI I/O as default
• Switch to vader as default for shared memory BTL
• PSM2 MTL

Terminating support

• Cray XT legacy items (ESS alps component, etc.) (DONE - although new ess/alps for Cray XE/XC)
• MX BTL (DONE)
• What other BTLs to delete? SCIF?
• Clean up README (DONE)
• Delete coll hierarch component
• coll ML disabled
• Delete VampirTrace interface
• Deprecate mpif77/mpif90: print a stderr warning

Testing

• What do we want to test?
  • More thread safety tests - non blocking collectives, etc.
  • OMPI I/O tests, refresh from HDF group? (DONE)

Stale code check - Opal

!INCLUDE "opal_mca_owner.md"

Stale code check - ORTE

!INCLUDE "orte_mca_owner.md"

Stale code check - OMPI

!INCLUDE "ompi_mca_owner.md"

Stale code check - Opal

Framework	Component	Owner	Status
shmem	sysv	LANL	maintenance
shmem	mmap	LANL	maintenance
shmem	posix	LANL	maintenance
shmem	base	LANL	maintenance
backtrace	none	SNL	maintenance
backtrace	execinfo	SNL	maintenance
backtrace	printstack	SNL	maintenance
backtrace	base	project	maintenance
crs	none	UTK	maintenance
crs	criu	CISCO	maintenance
crs	self	UTK	maintenance
crs	dmtcp	UBrit.Columbia	unmaintained
crs	base	project	maintenance
pstat	linux	INTEL	maintenance
pstat	test	INTEL	maintenance
if	bsdx_ipv4	INTEL	maintenance
if	bsdx_ipv6	INTEL	maintenance
if	linux_ipv6	INTEL	maintenance
if	solaris_ipv6	nobody	maintenance
if	posix_ipv4	INTEL	maintenance
if	base	project	active
pmix	s1	INTEL	active
pmix	s2	INTEL	active
pmix	cray	LANL	active
installdirs	env	SNL	maintenance
installdirs	config	SNL	maintenance
installdirs	base	project	active
hwloc	external	CISCO	maintenance
hwloc	hwloc1110	INTEL	maintenance
hwloc	base	project	maintenance
reachable	netlink	INTEL	unmaintained
reachable	weighted	INTEL	unmaintained
reachable	base	INTEL	unmaintained
event	external	CISCO	maintenance
event	libevent2022	INTEL	active
event	base	project	maintenance
allocator	basic	NVIDIA	maintenance
allocator	bucket	NVIDIA	maintenance
allocator	base	project	maintenance
timer	solaris	nobody	unmaintained
timer	linux	SNL	maintenance
timer	darwin	SNL	unmaintained
timer	aix	IBM?	unmaintained
timer	altix	SNL?	unmaintained
timer	base	SNL	maintenance
compress	gzip	project	maintenance
compress	bzip	project	maintenance
compress	base	project	maintenance
rcache	vma	LANL	maintenance
memcpy	base	project	maintenance
common	sm	UTK	maintenance
common	verbs	MELLANOX	maintenance
common	ugni	LANL	active
common	cuda	NVIDIA	active
common	libfabric	Intel	active
memchecker	valgrind	HLRS?	unmaintained
memchecker	base	project	unmaintained
dstore	hash	project	active
dstore	base	project	active
sec	basic	INTEL	maintenance
sec	munge	INTEL	active
sec	keystone	INTEL	maintenance
sec	base	INTEL	active
btl	tcp	UTK	active
btl	sm	UTK	active
btl	usnic	CISCO	active
btl	template	project	active
btl	portals4	SNL	active?
btl	scif	LANL	maintenance
btl	self	UTK	active
btl	ugni	LANL	active
btl	smcuda	NVIDIA	active
btl	vader	LANL	active
btl	openib	Chelsio	maintenance
btl	base	btlowners	active
mpool	sm	LANL	maintenance
mpool	gpusm	NVIDIA	maintenance
mpool	rgpusm	NVIDIA	maintenance
mpool	grdma	LANL	maintenance
mpool	udreg	LANL	maintenance
mpool	base	project	maintenance
memory	linux	MELLANOX,CISCO	maintenance
memory	malloc_solaris	nobody	unmaintained
memory	base	project	maintenance

Stale code check - ORTE

Framework	Component	Owner	Status
oob	tcp	INTEL	maintenance
oob	alps	LANL	active
oob	usock	INTEL	maintenance
oob	ud	MELLANOX	maintenance
oob	base	project	maintenance
rtc	hwloc	INTEL	maintenance
rtc	freq	INTEL	maintenance
rtc	omp	INTEL	active
rtc	base	INTEL	maintenance
schizo	ompi	INTEL	active
schizo	base	INTEL	active
filem	raw	INTEL	maintenance
filem	base	INTEL	maintenance
dfs	app	INTEL	maintenance
dfs	orted	INTEL	maintenance
dfs	base	INTEL	maintenance
rmaps	rank_file	INTEL	maintenance
rmaps	lama	CISCO	maintenance
rmaps	round_robin	INTEL	maintenance
rmaps	seq	INTEL	maintenance
rmaps	staged	INTEL	maintenance
rmaps	resilient	INTEL	maintenance
rmaps	mindist	MELLANOX	maintenance
rmaps	ppr	INTEL	maintenance
rmaps	base	INTEL	maintenance
routed	binomial	INTEL	maintenance
routed	debruijn	LANL?	unmaintained
routed	direct	INTEL	active
routed	base	INTEL	maintenance
errmgr	default_app	INTEL	maintenance
errmgr	default_orted	INTEL	maintenance
errmgr	default_tool	INTEL	maintenance
errmgr	default_hnp	INTEL	maintenance
errmgr	base	INTEL	maintenance
plm	isolated	INTEL	maintenance
plm	tm	INTEL	maintenance
plm	rsh	INTEL	maintenance
plm	alps	LANL	maintenance
plm	slurm	INTEL	maintenance
plm	lsf	INTEL	maintenance
plm	base	project	maintenance
ess	tm	INTEL	maintenance
ess	tool	INTEL	maintenance
ess	pmi	INTEL	maintenance
ess	hnp	INTEL	maintenance
ess	singleton	INTEL	maintenance
ess	alps	LANL	maintenance
ess	env	Intel	maintenance
ess	slurm	INTEL	maintenance
ess	lsf	INTEL	maintenance
ess	base	project	maintenance
rml	oob	INTEL	maintenance
rml	ftrm	?	unmaintained
rml	base	INTEL	maintenance
snapc	full	nobody	unmaintained
snapc	base	nobody	unmaintained
sstore	central	nobody	unmaintained
sstore	stage	nobody	unmaintained
sstore	base	nobody	unmaintained
odls	alps	LANL	active
odls	default	INTEL	maintenance
odls	base	project	maintenance
state	app	INTEL	active
state	novm	INTEL	active
state	staged_orted	INTEL	active
state	tool	INTEL	active
state	dvm	INTEL	active
state	hnp	INTEL	active
state	orted	INTEL	active
state	staged_hnp	INTEL	active
state	base	INTEL	active
grpcomm	rcd	INTEL	maintenance
grpcomm	direct	INTEL	maintenance
grpcomm	brks	INTEL	maintenance
grpcomm	base	INTEL	maintenance
ras	tm	INTEL	maintenance
ras	loadleveler	IBM	maintenance
ras	alps	LANL	active
ras	simulator	INTEL	maintenance
ras	slurm	INTEL	maintenance
ras	lsf	INTEL	maintenance
ras	gridengine	INTEL	unmaintained
ras	base	INTEL	maintenance
common	alps	LANL	active
iof	mr_hnp	INTEL	maintenance
iof	tool	INTEL	maintenance
iof	hnp	INTEL	maintenance
iof	mr_orted	INTEL	maintenance
iof	orted	INTEL	maintenance
iof	base	INTEL	maintenance

Stale code check - OMPI

Framework	Component	Owner	Status
fbtl	pvfs2	UH	active
fbtl	posix	UH	active
fbtl	plfs	UH	active
fbtl	base	UH	active
pubsub	orte	INTEL	maintenance
pubsub	pmi	INTEL	maintenance
pubsub	base	INTEL	maintenance
osc	sm	LANL	maintenance
osc	rdma	LANL	active?
osc	portals4	SNL	active
osc	pt2pt	LANL	active
osc	base	LANL	active
fs	pvfs2	UH	active
fs	lustre	UH	active
fs	ufs	UH	active
fs	plfs	UH	active
fs	base	UH	active
pml	ob1	LANL	active
pml	yalla	MELLANOX	active
pml	v	UTK	maintenance
pml	bfo	NVIDIA	unmaintained
pml	cm	SNL	maintenance
pml	crcpw	nobody	unmaintained
pml	base	project	active
dpm	orte	INTEL	maintenance
dpm	base	INTEL	maintenance
topo	basic	UTK	maintenance
topo	example	UTK	maintenance
topo	base	UTK	maintenance
vprotocol	pessimist	UTK	maintenance
vprotocol	example	UTK	maintenance
vprotocol	base	UTK	maintenance
coll	libnbc	project	active
coll	inter	UH	maintenance
coll	sm	nobody	unmaintained
coll	fca	MELLANOX	active
coll	hcoll	MELLANOX	active
coll	portals4	SNL	active
coll	self	CISCO	maintenance
coll	cuda	NVIDIA	maintenance
coll	basic	UH	maintenance
coll	ml	ORNL?	unmaintained
coll	demo	project	maintenance
coll	tuned	UTK	maintenance
coll	base	project	maintenance
bml	r2	SNL	maintenance
bml	base	project	maintenance
io	romio314	LANL/RIST	active
io	ompio	UH	active
io	base	project	maintenance
bcol	iboffload	ORNL	unmaintained
bcol	basesmuma	ORNL	unmaintained
bcol	ptpcoll	ORNL	unmaintained
bcol	base	ORNL	unmaintained
sharedfp	sm	UH	maintenance
sharedfp	lockedfile	UH	maintenance
sharedfp	individual	UH	maintenance
sharedfp	addproc	UH	maintenance
sharedfp	base	UH	maintenance
sbgp	p2p	ORNL	unmaintained
sbgp	basesmuma	ORNL	unmaintained
sbgp	basesmsocket	ORNL	unmaintained
sbgp	ibnet	ORNL	unmaintained
sbgp	base	ORNL	unmaintained
op	x86	INTEL	maintenance
op	example	project	maintenance
op	base	project	unmaintained
mtl	mxm	MELLANOX	active
mtl	psm	INTEL	active
mtl	portals4	SNL	active
mtl	ofi	INTEL	active
mtl	base	project	active
fcoll	dynamic	UH	active
fcoll	static	UH	active
fcoll	individual	UH	active
fcoll	two_phase	UH	active
fcoll	base	UH	active
crcp	bkmrk	nobody	unmaintained
crcp	base	nobody	unmaintained