Charts from the evaluation of Asynchronous Replication...

Post on 20-May-2020

18 views 0 download

Transcript of Charts from the evaluation of Asynchronous Replication...

Copyright @ 2016 Oracle and/or its affiliates. All rights reserved.

Vítor Oliveira (vitor.s.p.oliveira@oracle.com)Senior Performance Engineer

12th December 2016

Charts from the evaluation ofAsynchronous Replication, Group Replication and Galera

Part 1: Summary Charts

2

Copyright @ 2016 Oracle and/or its affiliates. All rights reserved.

Safe Harbor Statement

The following is intended to outline our general product direction. It is intended for information purpose only, and may not be incorporated into any contract. It is not a commitment to deliver any material, code, or functionality, and should not be relied up in making purchasing decisions. The development, release and timing of any features or functionality described for Oracle's product remains at the sole discretion of Oracle.

3

Copyright @ 2016 Oracle and/or its affiliates. All rights reserved.

Contents

1. Sysbench OLTP RW

2. Sysbench Update Index

3. Flow-control effects

For details refer to:http://mysqlhighavailability.com/performance-evaluation-mysql-5-7-group-replication/

4

Copyright @ 2016 Oracle and/or its affiliates. All rights reserved.

1. OLTP RW, Maximum throughput

3 members 5 members 7 members 9 members0

2 500

5 000

7 500

10 000

12 500

15 000

Single-master Maximum Throughput: Sysbench RW

Asynchronous (non-durable) Group Replication (non-durable) Galera (non-durable)

Asynchronous (durable) Group Replication (durable) Galera (durable)

Group Size

ma

xim

m s

ust

ain

ed

thro

ug

hp

ut (

tra

nsa

ctio

ns

pe

r se

con

d)

5

Copyright @ 2016 Oracle and/or its affiliates. All rights reserved.

1.1. OLTP RW, Throughput by Clients

8 16 32 64 128 2560

2 500

5 000

7 500

10 000

12 500

15 000

Throughput by Number of Clients: Sysbench RW (3 members)

Group Replication (non-durable) Galera (non-durable)

Group Replication (durable) Galera (durable)

total number of clients/threads

ma

xim

um

su

sta

ine

d th

rou

gh

pu

t (tr

an

sact

ion

s p

er

seco

nd

)

higher is better

6

Copyright @ 2016 Oracle and/or its affiliates. All rights reserved.

1.1. OLTP RW, Throughput by Clients

8 16 32 64 128 2560

2 500

5 000

7 500

10 000

12 500

15 000

Throughput by Number of Clients: Sysbench RW (5 members)

Group Replication (non-durable) Galera (non-durable)

Group Replication (durable) Galera (durable)

total number of clients/threads

ma

xim

um

su

sta

ine

d th

rou

gh

pu

t (tr

an

sact

ion

s p

er

seco

nd

)

higher is better

7

Copyright @ 2016 Oracle and/or its affiliates. All rights reserved.

1.1. OLTP RW, Throughput by Clients

8 16 32 64 128 2560

2 500

5 000

7 500

10 000

12 500

15 000

Throughput by Number of Clients: Sysbench RW (7 members)

Group Replication (non-durable) Galera (non-durable)

Group Replication (durable) Galera (durable)

total number of clients/threads

ma

xim

um

su

sta

ine

d th

rou

gh

pu

t (tr

an

sact

ion

s p

er

seco

nd

)

higher is better

8

Copyright @ 2016 Oracle and/or its affiliates. All rights reserved.

1.1. OLTP RW, Throughput by Clients

8 16 32 64 128 2560

2 500

5 000

7 500

10 000

12 500

15 000

Throughput by Number of Clients: Sysbench RW (9 members)

Group Replication (non-durable) Galera (non-durable)

Group Replication (durable) Galera (durable)

total number of clients/threads

ma

xim

um

su

sta

ine

d th

rou

gh

pu

t (tr

an

sact

ion

s p

er

seco

nd

)

higher is better

9

Copyright @ 2016 Oracle and/or its affiliates. All rights reserved.

1.2. OLTP RW, Latency by Clients

8 16 32 64 128 2561

10

100

1000

Single-master Latency: Sysbench RW(3 members)

Group Replication (durable) Galera (durable) Group Replication (non-durable) Galera (non-durable)

number of clients (threads)

clie

nt

95

% tr

an

sact

ion

late

ncy

(m

s)

lower is better

logarithmic scale

10

Copyright @ 2016 Oracle and/or its affiliates. All rights reserved.

1.2. OLTP RW, Latency by Clients

8 16 32 64 128 2561

10

100

1000

Single-master Latency: Sysbench RW(5 members)

Group Replication (durable) Galera (durable) Group Replication (non-durable) Galera (non-durable)

number of clients (threads)

clie

nt

95

% tr

an

sact

ion

late

ncy

(m

s)

lower is better

logarithmic scale

11

Copyright @ 2016 Oracle and/or its affiliates. All rights reserved.

1.2. OLTP RW, Latency by Clients

8 16 32 64 128 2561

10

100

1000

Single-master Latency: Sysbench RW(7 members)

Group Replication (durable) Galera (durable) Group Replication (non-durable) Galera (non-durable)

number of clients (threads)

clie

nt

95

% tr

an

sact

ion

late

ncy

(m

s)

lower is better

logarithmic scale

12

Copyright @ 2016 Oracle and/or its affiliates. All rights reserved.

1.2. OLTP RW, Latency by Clients

8 16 32 64 128 2561

10

100

1000

Single-master Latency: Sysbench RW(9 members)

Group Replication (durable) Galera (durable) Group Replication (non-durable) Galera (non-durable)

number of clients (threads)

clie

nt

95

% tr

an

sact

ion

late

ncy

(m

s)

lower is better

logarithmic scale

13

Copyright @ 2016 Oracle and/or its affiliates. All rights reserved.

Contents

1. Sysbench OLTP RW

2. Sysbench Update Index

3. Flow-control effects

For details refer to:

http://mysqlhighavailability.com/performance-evaluation-mysql-5-7-group-replication/

14

Copyright @ 2016 Oracle and/or its affiliates. All rights reserved.

2. Update Indexed, Maximum Throughput

3 members 5 members 7 members 9 members0

10 000

20 000

30 000

40 000

50 000

Single-master Maximum Throughput: Sysbench Update Indexed

Asynchronous (non-durable) Group Replication (non-durable) Galera (non-durable)

Asynchronous (durable) Group Replication (durable) Galera (durable)

Group Size

sust

ain

ed

thro

ug

hp

ut (

up

da

tes

pe

r se

con

d)

15

Copyright @ 2016 Oracle and/or its affiliates. All rights reserved.

2.1. Update Indexed, Throughput by Clients

8 16 32 64 128 2560

10 000

20 000

30 000

40 000

50 000

Throughput by Number of Clients: Sysbench Update Indexed (3 members)

Group Replication (non-durable) Galera (non-durable)

Group Replication (durable) Galera (durable)

total number of clients/threads

ma

xim

um

su

sta

ine

d th

rou

gh

pu

t (u

pd

ate

s p

er

seco

nd

)

higher is better

16

Copyright @ 2016 Oracle and/or its affiliates. All rights reserved.

2.1. Update Indexed, Throughput by Clients

8 16 32 64 128 2560

10 000

20 000

30 000

40 000

50 000

Throughput by Number of Clients: Sysbench Update Indexed (5 members)

Group Replication (non-durable) Galera (non-durable)

Group Replication (durable) Galera (durable)

total number of clients/threads

ma

xim

um

su

sta

ine

d th

rou

gh

pu

t (u

pd

ate

s p

er

seco

nd

)

higher is better

17

Copyright @ 2016 Oracle and/or its affiliates. All rights reserved.

2.1. Update Indexed, Throughput by Clients

8 16 32 64 128 2560

10 000

20 000

30 000

40 000

50 000

Throughput by Number of Clients: Sysbench Update Indexed (7 members)

Group Replication (non-durable) Galera (non-durable)

Group Replication (durable) Galera (durable)

total number of clients/threads

ma

xim

um

su

sta

ine

d th

rou

gh

pu

t (u

pd

ate

s p

er

seco

nd

)

higher is better

18

Copyright @ 2016 Oracle and/or its affiliates. All rights reserved.

2.1. Update Indexed, Throughput by Clients

8 16 32 64 128 2560

10 000

20 000

30 000

40 000

50 000

Throughput by Number of Clients: Sysbench Update Indexed (9 members)

Group Replication (non-durable) Galera (non-durable)

Group Replication (durable) Galera (durable)

total number of clients/threads

ma

xim

um

su

sta

ine

d th

rou

gh

pu

t (u

pd

ate

s p

er

seco

nd

)

higher is better

19

Copyright @ 2016 Oracle and/or its affiliates. All rights reserved.

2.2. Update Indexed, Latency by Clients

8 16 32 64 128 2561

10

100

1000

Single-master Latency: Sysbench Update Indexed(3 members)

Group Replication (durable) Galera (durable) Group Replication (non-durable) Galera (non-durable)

number of clients (threads)

clie

nt

95

% tr

an

sact

ion

late

ncy

(m

s)

logarithmic scale

lower is better

20

Copyright @ 2016 Oracle and/or its affiliates. All rights reserved.

2.2. Update Indexed, Latency by Clients

8 16 32 64 128 2561

10

100

1000

Single-master Latency: Sysbench Update Indexed(5 members)

Group Replication (durable) Galera (durable) Group Replication (non-durable) Galera (non-durable)

number of clients (threads)

clie

nt

95

% tr

an

sact

ion

late

ncy

(m

s)

logarithmic scale

lower is better

21

Copyright @ 2016 Oracle and/or its affiliates. All rights reserved.

2.2. Update Indexed, Latency by Clients

8 16 32 64 128 2561

10

100

1000

Single-master Latency: Sysbench Update Indexed(7 members)

Group Replication (durable) Galera (durable) Group Replication (non-durable) Galera (non-durable)

number of clients (threads)

clie

nt

95

% tr

an

sact

ion

late

ncy

(m

s)

logarithmic scale

lower is better

22

Copyright @ 2016 Oracle and/or its affiliates. All rights reserved.

2.2. Update Indexed, Latency by Clients

8 16 32 64 128 2561

10

100

1000

Single-master Latency: Sysbench Update Indexed(9 members)

Group Replication (durable) Galera (durable) Group Replication (non-durable) Galera (non-durable)

number of clients (threads)

clie

nt

95

% tr

an

sact

ion

late

ncy

(m

s)

logarithmic scale

lower is better

23

Copyright @ 2016 Oracle and/or its affiliates. All rights reserved.

Contents

1. Sysbench OLTP RW

2. Sysbench Update Index

3. Flow-control effects

For details refer to:

http://mysqlhighavailability.com/performance-evaluation-mysql-5-7-group-replication/

24

Copyright @ 2016 Oracle and/or its affiliates. All rights reserved.

3. Flow-control effects

8 16 32 64 128 256 8 16 32 64 128 256Durable settings Non-durable settings

0

2 500

5 000

7 500

10 000

12 500

15 000

Throughput varying Flow-control: Sysbench OLTP RW(3 members)

Group Replication: default settings flow-control disabled flow-control=1000

Galera: default settings flow-control disabled flow-control=1000

number of client threads

tota

l tra

nsa

ctio

ns

pe

r se

con

d (

TP

S)

25

Copyright @ 2016 Oracle and/or its affiliates. All rights reserved.

3. Flow-control effects

8 16 32 64 128 256 8 16 32 64 128 256Durable settings Non-durable settings

0

2 500

5 000

7 500

10 000

12 500

15 000

Throughput varying Flow-control: Sysbench OLTP RW(5 members)

Group Replication: default settings flow-control disabled flow-control=1000

Galera: default settings flow-control disabled flow-control=1000

number of client threads

tota

l tra

nsa

ctio

ns

pe

r se

con

d (

TP

S)

26

Copyright @ 2016 Oracle and/or its affiliates. All rights reserved.

3. Flow-control effects

8 16 32 64 128 256 8 16 32 64 128 256Durable settings Non-durable settings

0

2 500

5 000

7 500

10 000

12 500

15 000

Throughput varying Flow-control: Sysbench OLTP RW(7 members)

Group Replication: default settings flow-control disabled flow-control=1000

Galera: default settings flow-control disabled flow-control=1000

number of client threads

tota

l tra

nsa

ctio

ns

pe

r se

con

d (

TP

S)

27

Copyright @ 2016 Oracle and/or its affiliates. All rights reserved.

3. Flow-control effects

8 16 32 64 128 256 8 16 32 64 128 256Durable settings Non-durable settings

0

2 500

5 000

7 500

10 000

12 500

15 000

Throughput varying Flow-control: Sysbench OLTP RW(9 members)

Group Replication: default settings flow-control disabled flow-control=1000

Galera: default settings flow-control disabled flow-control=1000

number of client threads

tota

l tra

nsa

ctio

ns

pe

r se

con

d (

TP

S)

28

Copyright @ 2016 Oracle and/or its affiliates. All rights reserved.

3. Flow-control effects

8 16 32 64 128 256 8 16 32 64 128 256Durable settings Non-durable settings

0

7 500

15 000

22 500

30 000

37 500

45 000

Throughput varying Flow-control: Sysbench Update Indexed(3 members)

Group Replication: default settings flow-control disabled flow-control=1000

Galera: default settings flow-control disabled flow-control=1000

number of client threads

tota

l tra

nsa

ctio

ns

pe

r se

con

d (

TP

S)

29

Copyright @ 2016 Oracle and/or its affiliates. All rights reserved.

3. Flow-control effects

8 16 32 64 128 256 8 16 32 64 128 256Durable settings Non-durable settings

0

7 500

15 000

22 500

30 000

37 500

45 000

Throughput varying Flow-control: Sysbench Update Indexed(5 members)

Group Replication: default settings flow-control disabled flow-control=1000

Galera: default settings flow-control disabled flow-control=1000

number of client threads

tota

l tra

nsa

ctio

ns

pe

r se

con

d (

TP

S)

30

Copyright @ 2016 Oracle and/or its affiliates. All rights reserved.

3. Flow-control effects

8 16 32 64 128 256 8 16 32 64 128 256Durable settings Non-durable settings

0

7 500

15 000

22 500

30 000

37 500

45 000

Throughput varying Flow-control: Sysbench Update Indexed(7 members)

Group Replication: default settings flow-control disabled flow-control=1000

Galera: default settings flow-control disabled flow-control=1000

number of client threads

tota

l tra

nsa

ctio

ns

pe

r se

con

d (

TP

S)

31

Copyright @ 2016 Oracle and/or its affiliates. All rights reserved.

3. Flow-control effects

8 16 32 64 128 256 8 16 32 64 128 256Durable settings Non-durable settings

0

7 500

15 000

22 500

30 000

37 500

45 000

Throughput varying Flow-control: Sysbench Update Indexed(9 members)

Group Replication: default settings flow-control disabled flow-control=1000

Galera: default settings flow-control disabled flow-control=1000

number of client threads

tota

l tra

nsa

ctio

ns

pe

r se

con

d (

TP

S)

Copyright © 2012, Oracle and/or its affiliates. All rights reserved.

32