Data in analysis.db

This is a table schema file. You are advised to use MindStudio Insight to view the file or use a database development tool such as Navicat Premium to open the file. The profile data summarized by the current .db file is as follows:

CommAnalyzerBandwidth

Table 1 Format

Field

Type

Description

hccl_op_name

TEXT

Large communication operator name, for example, hcom_broadcast__303_1_1.

group_name

TEXT

Hash ID of the communicator, for example, 3915571125887837303.

transport_type

TEXT

Transmission type, including LOCAL, SDMA, and RDMA.

transit_size

NUMERIC

Amount of data to be transmitted, in MB.

transit_time

NUMERIC

Transmission duration, in milliseconds.

bandwidth

NUMERIC

Bandwidth, in GB/s.

large_packet_ratio

NUMERIC

Proportion of large data packets.

package_size

NUMERIC

Size of a communication data packet transmitted at a time, in MB.

count

NUMERIC

Number of communication transmission times.

total_duration

NUMERIC

Total data transmission duration.

step

TEXT

Step to which the operator belongs, for example, step 12.

type

TEXT

Operator type. The value can be Collective or P2P.

CommAnalyzerTime

Table 2 Format

Field

Type

Description

hccl_op_name

TEXT

Communication operator name.

group_name

TEXT

Group of communication operators.

start_timestamp

NUMERIC

Start timestamp of communication, in μs.

elapse_time

NUMERIC

Total operator communication duration, in milliseconds.

transit_time

NUMERIC

Communication duration, in milliseconds. If the communication duration is too long, a link may be faulty.

wait_time

NUMERIC

Waiting duration, in milliseconds. Before establishing communication between nodes, ensure that the synchronization between the two nodes is complete.

synchronization_time

NUMERIC

Synchronization duration, in milliseconds. It is the duration required for synchronization between nodes.

idle_time

NUMERIC

Idle time, in milliseconds. Idle time (idle_time) = Total operator communication duration (elapse_time) – Communication duration (transit_time) – Waiting duration (wait_time)

step

TEXT

Step to which the operator belongs.

type

TEXT

Operator type. The value can be Collective or P2P.

CommAnalyzerMatrix

Table 3 Format

Field

Type

Description

hccl_op_name

TEXT

Simplified operator name after matrix analysis, for example, send-top1.

group_name

TEXT

Hash ID of the communicator, for example, 3915571125887837303.

src_rank

TEXT

Rank ID of the source data, for example, 0.

dst_rank

TEXT

Rank ID of the destination data, for example, 1.

transport_type

TEXT

Transmission type, including LOCAL, SDMA, and RDMA.

transit_size

NUMERIC

Amount of data to be transmitted, in MB.

transit_time

NUMERIC

Transmission duration, in milliseconds.

bandwidth

NUMERIC

Bandwidth, in GB/s.

step

TEXT

Step to which the operator belongs, for example, step 12.

type

TEXT

Operator type. The value can be Collective or P2P.

op_name

TEXT

Original name of the operator, for example, hcom_broadcast__303_1_1.

StepTraceTime

Table 4 Format

Field

Type

Description

deviceId

INTEGER

Device ID.

step

TEXT

Step number, for example, 12.

computing

NUMERIC

Compute time (ms)

communication

NUMERIC

Communication duration, in milliseconds.

overlapped

NUMERIC

Time for overlapping computation and communication, in milliseconds.

communication_not_overlapped

NUMERIC

Not overlapped communication time, in milliseconds.

free

NUMERIC

Free time, in milliseconds.

stage

NUMERIC

Time excluding the time of receiving data in a step, in milliseconds.

bubble

NUMERIC

Time for receiving data in a step, in milliseconds.

communication_not_overlapped_and_exclude_receive

NUMERIC

Not overlapped communication time minus the time used for receiving data, in milliseconds.