Why is mandatory to upgrade our Aurora clusters to the newer instance family? Let's figure out together why!
The R4 and R5 instances are optimized for memory-intensive database workloads and offer better price per GiB.
R4
2.3 GHz Intel Xeon E5-2686 v4 Processor
Intel AVX, Intel AVX2, Intel Turbo
EBS Optimized
Enhanced Networking
DDR4 Memory
R5
Up to 768GiB of memory per instance
Intel Xeon Platinum 800 series speed up to 3.1 GHz
Powered by the AWS Nitro System
Uo to 3.1GHz Intel Xeon Patinum 8000 Processor
Intel AVX, Intel AVX2, Intel Turbo
Running the benchmark tests
Benchmark EC2
I used the IAM AuroraBenchmark to create three EC2 r4.4xlarge instances to use to execute the sysbench commands.
Ireland: ami-f45d1983
Ireland: ami-f45d1983
3 EC2 instances r4.4xlarge
Aurora cluster
2 nodes
2 nodes
5.6.mysql_aurora.1.22.2 db.r4.large
5.6.mysql_aurora.1.22.2 db.r5.large
Sysbench installation
sudo yum -y install bzr automake mysql-devel libtool
cd /opt/benchmark_tests/test_suits/sysbench
export W='database-1.cluster-cfjc3gydmz8d.eu-west-1.rds.amazonaws.com'
export R='database-1.cluster-ro-cfjc3gydmz8d.eu-west-1.rds.amazonaws.com'
Sysbench test-data
./sysbench --test=tests/db/oltp.lua --mysql-host=$W --mysql-port=3306 --mysql-user=admin --mysql-password=******** --mysql-db=sbtest --mysql-table-engine=innodb --oltp-table-size=25000 --oltp-tables-count=250 --db-driver=mysql prepare
Sysbench read-heavy test
./sysbench --test=tests/db/oltp.lua --mysql-host=$W --oltp-tables-count=250 --mysql-user=admin --mysql-password=******* --mysql-port=3306 --db-driver=mysql --oltp-table-size=25000 --mysql-db=sbtest --max-requests=0 --oltp_simple_ranges=0 --oltp-distinct-ranges=0 --oltp-sum-ranges=0 --oltp-order-ranges=0 --max-time=600 --oltp-read-only=on --num-threads=32 run
Sysbench write-heavy test
./sysbench --test=tests/db/oltp.lua --mysql-host=$W --oltp-tables-count=250 --mysql-user=admin --mysql-password=******** --mysql-port=3306 --db-driver=mysql --oltp-table-size=25000 --mysql-db=sbtest --max-requests=0 --max-time=600 --oltp_simple_ranges=0 --oltp-distinct-ranges=0 --oltp-sum-ranges=0 --oltp-order-ranges=0 --oltp-point-selects=0 --num-threads=64 --rand-type=uniform run
Results
Only the most performance results of each of the three nodes are reported
Number of threads: 64 for each of the three EC2 instances (total 192)
Summary
R4 family
---- ONLY READ (second execution)
read/write requests: 22,176 per sec
response time:
min: 8.73ms
avg: 86.57ms
max: 275.38ms
approx. 95 percentile: 117.06ms
---- ONLY WRITE (fitfh execution)
read/write requests: 5,937 per sec
response time:
min: 28.47ms
avg: 129.33ms
max: 443.73ms
approx. 95 percentile: 189.94ms
R5 family
---- ONLY READ (third execution)
read/write requests: 39,567 per sec
response time:
min: 6.73ms
avg: 48.52ms
max: 420.96ms
approx. 95 percentile: 60.88ms
---- ONLY WRITE (fourth execution)
read/write requests: 10,140 per sec.)
response time:
min: 15.39ms
avg: 75.72ms
max: 244.90ms
approx. 95 percentile: 107.13ms
Conclusion
The R5 instances deliver a 10% price per GiB improvement and a ~20% increased CPU performance over R4.
The read and write workloads gain a lot of benefits of from the newer hardware and the increased CPU perf, the writer cluster is able to almost double the load. As well as the statements return quicker to the clients.