Skip to content

Latest commit

 

History

History
84 lines (62 loc) · 3.42 KB

exchange-test.md

File metadata and controls

84 lines (62 loc) · 3.42 KB
Raw

Nebula-Exchange test result

We use LDBC dataset to test the exchange client import performance.

prepare

  • The Nebula Schema DDL is configed in bench/NEBULA_DDL.

  • The exchange config file is configed in bench/EXCHANGE_CONFIG.

import command

for space sf1, the command is:

spark-submit --master "spark://127.0.0.1:7077" \
--driver-memory=2G \
--num-executors=3 \
--executor-memory=10G \
--executor-cores=20 \
--class com.vesoft.nebula.exchange.Exchange \
nebula-exchange-2.6.0.jar -c app_sf1.conf

for space sf30, the command is:

spark-submit --master "spark://127.0.0.1:7077" \
--driver-memory=2G \
--num-executors=3 \
--executor-memory=30G \
--executor-cores=20 \
--class com.vesoft.nebula.exchange.Exchange \
nebula-exchange-2.6.0.jar -c app_sf30.conf

for space sf100, the command is:

spark-submit --master "spark://127.0.0.1:7077" \
--driver-memory=2G \
--num-executors=3 \
--executor-memory=30G \
--executor-cores=20 \
--class com.vesoft.nebula.exchange.Exchange \
nebula-exchange-2.6.0.jar -c app_sf100.conf

import result

Here is the import result:

When Space has 1 replica, and the auto-compact is enable.

Dataset Data Amount cores executor-memory spark-partition batch size duration speed
LDBC sf1 vertex:3165488 edge:17256029 60 10G 60 2000 56s 360,000/s
LDBC sf30 vertex:88673640 edge:540915215 60 20G 60 2000 7.5min 1,399,086/s
LDBC sf100 vertex:282386021 edge:1775513185 60 30G 60 2000 27min 1,270,303/s

When Space has 1 replica, and the auto-compact is false.

Dataset Data Amount cores executor-memory spark-partition batch size duration speed
LDBC sf1 vertex:3165488 edge:17256029 60 10G 60 2000 49s 416,765/s
LDBC sf30 vertex:88673640 edge:540915215 60 20G 60 2000 6.3min 1,665,578/s
LDBC sf100 vertex:282386021 edge:1775513185 60 30G 60 2000 22min 1,559,014/s

After data import, space sf100 with one replica will take to finish the manual compaction.

When Space has 3 replicas, and the auto-compact is closed.

Dataset Data Amount cores executor-memory spark-partition batch size duration speed
LDBC sf1 vertex:3165488 edge:17256029 60 10G 60 2000 58s 352,095/s
LDBC sf30 vertex:88673640 edge:540915215 60 20G 60 2000 17min 617,243/s
LDBC sf100 vertex:282386021 edge:1775513185 60 30G 60 2000 42min 816,623/s

After data import, space sf100 with three replicas will take 1.1h to finish the manual compaction.

other information

The Spark cluster and nebula cluster are separated

Spark cluster has three workers, nebula cluster has three metad, three graphd and three storaged.

The clusters have 10 Gigabit Network, each nebula machine has 1.5T SSD disk and 256G memory.