Skip to content

Commit

Permalink
Merge pull request #1119 from SpiNNakerManchester/provenance
Browse files Browse the repository at this point in the history 
Provenance straight to and from the database.
  • Loading branch information
@Christian-B
Christian-B authored Oct 25, 2021
2 parents 632dd5d + 41358db commit ae17160
Show file tree
Hide file tree
Showing 20 changed files with 495 additions and 508 deletions.
26 changes: 14 additions & 12 deletions spynnaker/pyNN/external_devices_models/machine_munich_motor_device.py
View file
Original file line number Diff line number Diff line change
Expand Up @@ -26,12 +26,11 @@
from spinn_front_end_common.abstract_models.impl import (
ProvidesKeyToAtomMappingImpl)
from spinn_front_end_common.interface.provenance import (
ProvidesProvenanceDataFromMachineImpl)
ProvidesProvenanceDataFromMachineImpl, ProvenanceWriter)
from spinn_front_end_common.interface.simulation import simulation_utilities
from spinn_front_end_common.utilities.constants import (
SYSTEM_BYTES_REQUIREMENT, SIMULATION_N_BYTES, BYTES_PER_WORD)
from spinn_front_end_common.utilities.utility_objs import (
ExecutableType, ProvenanceDataItem)
from spinn_front_end_common.utilities.utility_objs import ExecutableType
from spynnaker.pyNN.exceptions import SpynnakerException


Expand Down Expand Up @@ -124,17 +123,20 @@ def _n_additional_data_items(self):
@overrides(
ProvidesProvenanceDataFromMachineImpl.parse_extra_provenance_items)
def parse_extra_provenance_items(
self, label, names, provenance_data):
self, label, x, y, p, provenance_data):
n_buffer_overflows, = provenance_data

yield ProvenanceDataItem(
names + [self.INPUT_BUFFER_FULL_NAME], n_buffer_overflows,
(n_buffer_overflows > 0),
f"The input buffer for {label} lost packets on "
f"{n_buffer_overflows} occasions. "
"This is often a sign that the system is running too quickly for "
"the number of neurons per core. Please increase the timer_tic "
"or time_scale_factor or decrease the number of neurons per core.")
with ProvenanceWriter() as db:
db.insert_core(x, y, p, self.INPUT_BUFFER_FULL_NAME,
n_buffer_overflows)
if n_buffer_overflows > 0:
db.insert_report(
f"The input buffer for {label} lost packets on "
f"{n_buffer_overflows} occasions. "
"This is often a sign that the system is running too "
"quickly for the number of neurons per core. "
"Please increase the timer_tic or time_scale_factor "
"or decrease the number of neurons per core.")

@inject_items({"routing_info": "RoutingInfos"})
@overrides(
Expand Down
30 changes: 1 addition & 29 deletions spynnaker/pyNN/extra_algorithms/algorithms_metadata.xml
View file
Original file line number Diff line number Diff line change
Expand Up @@ -107,10 +107,6 @@
<param_name>routing_infos</param_name>
<param_type>RoutingInfos</param_type>
</parameter>
<parameter>
<param_name>provenance_data_objects</param_name>
<param_type>RouterCompressorProvenanceItems</param_type>
</parameter>
<parameter>
<param_name>read_expander_iobuf</param_name>
<param_type>SynapticExpanderReadIOBuf</param_type>
Expand All @@ -129,12 +125,8 @@
<param_name>executable_targets</param_name>
<token part="BitFieldData">DataLoaded</token>
</required_inputs>
<optional_inputs>
<param_name>provenance_data_objects</param_name>
</optional_inputs>
<outputs>
<token part="MulticastRoutesLoaded">DataLoaded</token>
<param_type>RouterCompressorProvenanceItems</param_type>
</outputs>
</algorithm>
<algorithm name="SpynnakerMachineBitFieldUnorderedRouterCompressor">
Expand Down Expand Up @@ -178,10 +170,6 @@
<param_name>routing_infos</param_name>
<param_type>RoutingInfos</param_type>
</parameter>
<parameter>
<param_name>provenance_data_objects</param_name>
<param_type>RouterCompressorProvenanceItems</param_type>
</parameter>
<parameter>
<param_name>read_expander_iobuf</param_name>
<param_type>SynapticExpanderReadIOBuf</param_type>
Expand All @@ -200,12 +188,8 @@
<param_name>executable_targets</param_name>
<token part="BitFieldData">DataLoaded</token>
</required_inputs>
<optional_inputs>
<param_name>provenance_data_objects</param_name>
</optional_inputs>
<outputs>
<token part="MulticastRoutesLoaded">DataLoaded</token>
<param_type>RouterCompressorProvenanceItems</param_type>
</outputs>
</algorithm>
<algorithm name="SpynnakerMachineBitFieldPairRouterCompressor">
Expand Down Expand Up @@ -252,10 +236,6 @@
<param_name>routing_infos</param_name>
<param_type>RoutingInfos</param_type>
</parameter>
<parameter>
<param_name>provenance_data_objects</param_name>
<param_type>RouterCompressorProvenanceItems</param_type>
</parameter>
</input_definitions>
<required_inputs>
<param_name>executable_finder</param_name>
Expand All @@ -270,12 +250,8 @@
<param_name>read_expander_iobuf</param_name>
<token part="BitFieldData">DataLoaded</token>
</required_inputs>
<optional_inputs>
<param_name>provenance_data_objects</param_name>
</optional_inputs>
<outputs>
<token part="MulticastRoutesLoaded">DataLoaded</token>
<param_type>RouterCompressorProvenanceItems</param_type>
</outputs>
</algorithm>
<algorithm name="OnChipBitFieldGenerator">
Expand Down Expand Up @@ -390,13 +366,9 @@
<python_module>spynnaker.pyNN.extra_algorithms.redundant_packet_count_report</python_module>
<python_class>RedundantPacketCountReport</python_class>
<input_definitions>
<parameter>
<param_name>provenance_items</param_name>
<param_type>PlacementsProvenanceItems</param_type>
</parameter>
</input_definitions>
<required_inputs>
<param_name>provenance_items</param_name>
<token>PlacementsProvenanceItems</token>
</required_inputs>
</algorithm>
<algorithm name="SpYNNakerConnectionHolderGenerator">
Expand Down
198 changes: 97 additions & 101 deletions spynnaker/pyNN/extra_algorithms/redundant_packet_count_report.py
View file
Original file line number Diff line number Diff line change
Expand Up @@ -14,129 +14,125 @@
# along with this program. If not, see <http://www.gnu.org/licenses/>.
import os
import logging
import statistics
from collections import defaultdict

from spinn_utilities.log import FormatAdapter
from spinn_front_end_common.interface.provenance import (
ProvenanceReader, ProvenanceWriter)
from spinn_front_end_common.utilities.globals_variables import (
report_default_directory)
from spynnaker.pyNN.models.neuron import PopulationMachineVertex

logger = FormatAdapter(logging.getLogger(__name__))

REDUNDANCY_BY_CORE = (
"CREATE VIEW IF NOT EXISTS redundancy_by_core AS "
"SELECT pro.x, pro.y, pro.p, core_name, "
" received, filtered, invalid, failed, "
" filtered + invalid + failed AS redunant, "
" filtered + invalid + failed + received AS total, "
" (filtered + invalid + failed) * 100.0 / "
" (filtered + invalid + failed + received) AS percent "
"FROM"
" (SELECT x, y, p, total as received, core_name "
" FROM core_stats_view "
f" WHERE description = \"{PopulationMachineVertex.SPIKES_PROCESSED}\") "
" AS pro "
"LEFT JOIN "
" (SELECT x, y, p, total AS filtered "
" FROM core_stats_view "
" WHERE description = "
f" \"{PopulationMachineVertex.BIT_FIELD_FILTERED_PACKETS}\") "
f" AS bit "
"LEFT JOIN "
" (SELECT x, y, p, total AS invalid "
" FROM core_stats_view "
" WHERE description = "
f" \"{PopulationMachineVertex.INVALID_MASTER_POP_HITS}\") AS inv "
"LEFT JOIN "
" (SELECT x, y, p, total AS failed "
" FROM core_stats_view "
" WHERE description = "
f" \"{PopulationMachineVertex.GHOST_SEARCHES}\") AS fai "
"WHERE pro.x = bit.x AND pro.y = bit.y AND pro.p = bit.p "
" AND pro.x = inv.x AND pro.y = inv.y AND pro.p = inv.p "
" AND pro.x = fai.x AND pro.y = fai.y AND pro.p = fai.p")


REDUNDANCY_SUMMARY = (
"CREATE VIEW IF NOT EXISTS redundancy_summary AS "
"SELECT SUM(total), MAX(total), MIN(total), AVG(total), "
" SUM(redunant), MAX(redunant), MIN(redunant), AVG(redunant), "
" MAX(percent), MIN(percent), AVG(percent), "
" SUM(redunant) * 100.0 / SUM(total) as global_percent "
"FROM redundancy_by_core")


class RedundantPacketCountReport(object):

_FILE_NAME = "redundant_packet_count.rpt"
_N_PROV_ITEMS_NEEDED = 4
_MAX = 100

_CORE_LEVEL_MSG = (
"core {} \n\n {} packets received.\n {} were detected as "
"redundant packets by the bitfield filter.\n {} were detected as "
"having no targets after the DMA stage.\n {} were detected as "
"packets which we should not have received in the first place. \n"
" Overall this makes a redundant percentage of {}\n")

_SUMMARY_LEVEL_MSG = (
"overall, the core with the most incoming packets had {} packets.\n"
" The core with the least incoming packets had {} packets.\n"
" The core with the most redundant packets had {} packets.\n"
" The core with the least redundant packets had {} packets.\n"
" The average incoming and redundant were {} and {}.\n"
" The max and min percentages of redundant packets are {}"
" and {}. \n"
" The average redundant percentages from each core were {} "
"accordingly.\n"
" The total packets flown in system was {}"
)

def __call__(self, provenance_items):
def __call__(self):
"""
:param provenance_items:
:type provenance_items:
list(~spinn_front_end_common.utilities.utility_objs.ProvenanceDataItem)
:return:
"""
file_name = os.path.join(report_default_directory(), self._FILE_NAME)

self._create_views()
try:
with open(file_name, "w") as f:
self._write_report(f, provenance_items)
self._write_report(f)
except Exception: # pylint: disable=broad-except
logger.exception("Generate_placement_reports: Can't open file"
" {} for writing.", self._FILE_NAME)

def _write_report(self, output, provenance_items):
data = defaultdict(dict)

overall_entries = list()
overall_redundant = list()
overall_redundant_percentage = list()

for provenance_item in provenance_items:
last_name = provenance_item.names[-1]
key = provenance_item.names[0]
if ((last_name == PopulationMachineVertex.GHOST_SEARCHES) or
(last_name ==
PopulationMachineVertex.BIT_FIELD_FILTERED_PACKETS) or
(last_name ==
PopulationMachineVertex.INVALID_MASTER_POP_HITS) or
(last_name == PopulationMachineVertex.SPIKES_PROCESSED)):

# add to store
data[key][last_name] = provenance_item.value

# accum enough data on a core to do summary
if len(data[key].keys()) == self._N_PROV_ITEMS_NEEDED:

# total packets received
total = (
data[key][PopulationMachineVertex.GHOST_SEARCHES] +
data[key][
PopulationMachineVertex.
BIT_FIELD_FILTERED_PACKETS] +
data[key][
PopulationMachineVertex.INVALID_MASTER_POP_HITS] +
data[key][PopulationMachineVertex.SPIKES_PROCESSED])

# total redundant packets
redundant = (
data[key][PopulationMachineVertex.GHOST_SEARCHES] +
data[key][
PopulationMachineVertex.
BIT_FIELD_FILTERED_PACKETS] +
data[key][
PopulationMachineVertex.INVALID_MASTER_POP_HITS])

percentage = 0
if total != 0:
percentage = (self._MAX / total) * redundant

# add to the trackers for summary data
overall_entries.append(total)
overall_redundant.append(redundant)
overall_redundant_percentage.append(percentage)

output.write(self._CORE_LEVEL_MSG.format(
key, total,
data[key][
PopulationMachineVertex.
BIT_FIELD_FILTERED_PACKETS],
data[key][PopulationMachineVertex.GHOST_SEARCHES],
data[key][
PopulationMachineVertex.INVALID_MASTER_POP_HITS],
percentage))

# do summary
if len(overall_entries) != 0:
output.write(self._SUMMARY_LEVEL_MSG.format(
max(overall_entries), min(overall_entries),
max(overall_redundant), min(overall_redundant),
statistics.mean(overall_entries),
statistics.mean(overall_redundant),
max(overall_redundant_percentage),
min(overall_redundant_percentage),
statistics.mean(overall_redundant_percentage),
sum(overall_entries)))
else:
output.write("was no data to summarise")
def _create_views(self):
with ProvenanceWriter() as db:
with db.transaction() as cur:
cur.execute(REDUNDANCY_BY_CORE)
cur.execute(REDUNDANCY_SUMMARY)

def _write_report(self, output):
reader = ProvenanceReader()
for data in reader.run_query("select * from redundancy_by_core"):
(x, y, p, source, received, filtered, invalid, failed,
redundant, total, percent) = data
output.write(f"\ncore {source} \n")
output.write(f" {total} packets received. \n")
output.write(f" {redundant} were detected as "
"redundant packets by the bitfield filter. \n")
output.write(
f" {filtered} were detected as having no targets "
f"after the DMA stage. \n")
output.write(
f" {invalid} were detected as packets which "
f"we should not have received in the first place. \n")
output.write(f" Overall this makes a redundant percentage of "
f"{percent}\n")
data = reader.run_query("select * from redundancy_summary")
(sum_total, max_total, min_total, avg_total,
sum_reduant, max_redundant, min_redundant, avg_redundant,
max_percent, min_percent, avg_percent, global_percent) = data[0]
output.write(f"\nThe total packets flown in system was "
f"{sum_total}\n")
output.write(f" The max, min and avergae per core was {max_total}, "
f"{min_total}, {avg_total} packets.\n")
output.write(f"The total redundant packets was "
f"{sum_reduant}\n")
output.write(f" The max, min and avergae per core was "
f"{max_redundant}, {min_redundant}, {avg_redundant} "
f"packets.\n")
output.write(
f"The percentages of redundant packets was {global_percent}\n")
output.write(f" The max and min percentages of redundant "
f"packets are {max_percent} and {min_percent}. \n")
output.write(
f" The average redundant percentages from each "
f"core were {avg_percent} accordingly.\n")


if __name__ == '__main__':
print(REDUNDANCY_BY_CORE)
print(REDUNDANCY_SUMMARY)
Loading

0 comments on commit ae17160

Please sign in to comment.