RANCHER-582 Investigate memory optimised instances usage for Rancher clusters

RANCHER-582 Investigate memory optimised instances usage for Rancher clusters

Difference between instance types

Table with compare technical characteristics EC2 instances that can be applied to our infrastructure

Result of investigation shown in table below.  I tested fourth instance types on our infrastructure for compare how many nodes we will create on test application on different utilization.

Currently in use

m5ad.xlarge, m5.xlarge, m5a.xlarge, m5d.xlarge

Class instance

Type instance 

CPU

Memory

Network Bandwidth (Gigabit)

Storage

Processor

price for 1 hour for 1 instance
(On Demand)

scale-down-utilization-threshold=0,5
count of nodes

scale-down-utilization-threshold=0,75
count of nodes

m5

m5ad.xlarge

4

16

10

1 * 150 NVMe SSD

AMD EPYC 7000

0,206

 

 

m5.xlarge

4

16

10

EBS only

Intel Xeon® Platinum 8175M

0,192

5

5

m5a.xlarge

4

16

10

EBS only

AMD EPYC 7000

0,172

 

 

m5d.xlarge

4

16

10

1 * 150 NVMe SSD

Intel Xeon® Platinum 8175M

0,226

 

 

r5

r5.large

2

16

10

EBS only

Intel Xeon® Platinum 8000 (Skylake 8175M или Cascade Lake 8259CL)

0,126

7

6

r5.xlarge

4

32

10

EBS only

Intel Xeon® Platinum 8000 (Skylake 8175M или Cascade Lake 8259CL)

0,252

 

 

r5a.xlarge

4

32

10

EBS only

AMD EPYC 7000 (AMD EPYC 7571)

0,226

3

3

r5ad.xlarge

4

32

10

1 * 150 NVMe SSD

AMD EPYC 7000 (AMD EPYC 7571)

0,262

 

 

r5d.xlarge

4

32

10

1 * 150 NVMe SSD

Intel Xeon® Platinum 8000 (Skylake 8175M или Cascade Lake 8259CL)

0,288

 

 

r6

r6a.large

2

16

12,5

EBS only

AMD EPYC (AMD EPYC 7R13)

0,113

 

 

r6a.xlarge

4

32

12,5

EBS only

AMD EPYC (AMD EPYC 7R13)

0,226

 

 

r6i.xlarge

4

32

12,5

EBS only

 Intel Xeon Scalable (Ice Lake 8375C)

0,252

3

3

r6i.large

2

16

12,5

EBS only

 Intel Xeon Scalable (Ice Lake 8375C)

0,126

 

 

r6in.large

2

16

25

EBS only

 Intel Xeon Scalable (Ice Lake 8375C)

0,174

 

 

r6in.xlarge

4

32

30

EBS only

 Intel Xeon Scalable (Ice Lake 8375C)

0,348

 

 

Features of different instances class

Information was taken from official aws documentation https://aws.amazon.com/ec2/instance-types/?nc1=h_ls.

m5

  • Up to 3.1 GHz Intel Xeon Scalable processor (Skylake 8175M or Cascade Lake 8259CL) with new Intel Advanced Vector Extension (AVX-512) instruction set

  • Up to 25 Gbps network bandwidth using Enhanced Networking

  • Instance storage offered via EBS or NVMe SSDs that are physically attached to the host server

m5a

  • AMD EPYC 7000 series processors (AMD EPYC 7571) with an all core turbo clock speed of 2.5 GHz

  • Up to 20 Gbps network bandwidth using Enhanced Networking

  • Instance storage offered via EBS or NVMe SSDs that are physically attached to the host server

r5

  • Up to 3.1 GHz Intel Xeon® Platinum 8000 series processors (Skylake 8175M or Cascade Lake 8259CL) with new Intel Advanced Vector Extension (AVX-512) instruction set

  • With R5d instances, local NVMe-based SSDs are physically connected to the host server and provide block-level storage that is coupled to the lifetime of the R5 instance

r5a

  • AMD EPYC 7000 series processors (AMD EPYC 7571) with an all core turbo clock speed of 2.5 GHz

  • Up to 20 Gbps network bandwidth using Enhanced Networking

  • Instance storage offered via EBS or NVMe SSDs that are physically attached to the host server

  • With R5ad instances, local NVMe-based SSDs are physically connected to the host server and provide block-level storage that is coupled to the lifetime of the R5a instance

r6a

  • Up to 3.6 GHz 3rd generation AMD EPYC processors (AMD EPYC 7R13)

  • Up to 35% better compute price performance over R5a instances

  • Up to 50 Gbps of networking speed

r6i

  • Up to 3.5 GHz 3rd generation Intel Xeon Scalable processors (Ice Lake 8375C)

  • Up to 15% better compute price performance over R5 instances

  • Up to 20% higher memory bandwidth per vCPU compared to R5 instances

  • Up to 50 Gbps of networking speed

  • With R6id instances, up to 7.6 TB of local NVMe-based SSDs are physically connected to the host server and provide block-level storage that is coupled to the lifetime of the R6i instance

r6in

  • Up to 3.5 GHz 3rd Generation Intel Xeon Scalable processors (Ice Lake 8375C)

  • Up to 20% higher memory bandwidth per vCPU compared to R5n and R5dn instances

  • Up to 200 Gbps of networking speed, which is up to 2x compared to R5n and R5dn instances

  • Up to 80 Gbps of EBS bandwidth, which is up to 1.3x more than R5b instances

  • With R6idn instances, up to 7.6 TB of local NVMe-based SSDs are physically connected to the host server and provide block-level storage that is coupled to the R6idn instance lifetime

Compare total price on tested instance types

Total price for 1 hour

In table below we can see the total price for run one test application on tmp cluster for each instance type, so total price for m5.xlarge is the base price that we are paying now.

 

Currently in use

Currently in use

m5.xlarge

Type instance 

price for 1 hour for 1 instance
(On Demand)

Total count nodes

Total price

m5.xlarge

0,192

5

0.96

r5.large

0,126

7

0.882

r5a.xlarge

0,226

3

0.678

r6i.xlarge

0,252

3

0.756

Total discount

In table below we can see total discount if we will run the test application on different instance type. So, as you can see for r5a.xlarge we will pay less on about 30% because price for AMD less on 10% than on Intel and we are creating less nodes then for m5.xlarge , so this gives to us opportunity to save money.

Type instance 

Discount if compare with m5.xlarge

Type instance 

Discount if compare with m5.xlarge

r5.large

~10%

r5a.xlarge

~30%

r6i.xlarge

~20%

Compare utilization on Rancher

Files with screenshots of utilization CPU/RAM that we can see in rancher.

Calculate total cost for 3 years for different EC2 plans

Calculate total cost for 3 years for 1 instance

Table below show the total cost that we will spend for 3 years using different EC2 payment options for 1 instance.

 

 

Standard RI

Save plan

Type instance 

On Demand

no upfront

Partial upfront

all upfront

no upfront

Partial upfront

all upfront

m5.xlarge

4976,64

2181

2007,64

1897

2549,16

2365,2

2312,64

r5.large

3265,92

1419

1319

1245

1734,48

1603,26

1576,8

r5a.xlarge

5857,92

2575

2370,6

2233

3101,04

2864,44

2811,96

r6i.xlarge

6531,84

3004

2781,68

2615

3454,56

3198,48

3134,68

Calculate total cost for 3 years for amount instances

Count was taken from investigation when we had to run our test cluster on different instance types. 

Type instance 

count nodes

m5.xlarge

5

r5.large

7

r5a.xlarge

3

r6i.xlarge

3

Table below show the total cost that we will spend for 3 years using different EC2 payment options for run test application on folio-tmp cluster.

 

 

Standard RI

Save plan

Type instance 

On Demand

no upfront

Partial upfront

all upfront

no upfront

Partial upfront

all upfront

m5.xlarge

24883,2

10906

10038,2

9485

12745,8

11826

11563,2

r5.large

22861,44

9934

9233

8715

12141,36

11222,82

11037,6

r5a.xlarge

17573,76

7726

7111,8

6699

9303,12

8593,32

8435,88

r6i.xlarge