Load balancer Performance Issue

[Priyam Vaidya]

Hello Everyone,

I am doing load testing on one of our service hosted on GCP. The services runs on compute engine instance behind a load balancer. I have used a similar setup on AWS to perform load testing and compare results.

The issue is the performance on AWS is significantly faster than one on GCP. I am using similar setup for both the platforms. I wanted to know what might be cause this much performance difference.

Specification
GCP: e2-standard-2 (General Purpose, 2vCPUs, 8 GB memory)

Region:North Virginia us-east4

Number of Instance:50

Average Response Time per call(57000 calls) : 1.75 seconds

AWS:ec2 (m5.large, general purpose, 2 vCPUs, 8 GB memory)

Region:North Virginiaus-east-1

Number of Instance:50

Average Response Time per call (57000 calls): 0.496 seconds

PS: I am using custom script to generate calls for load testing. This custom script runs on an GCP compute engine instance in the same region. (This instance is used to load test on both GCP and AWS )

Thanks in advance.

[Naman Parekh]

Google Cloud accounts for bandwidth per virtual machine (VM) instance, not per network interface (NIC) or IP address. A VM's machine type defines its maximum possible egress rate; however, you can only achieve that maximum possible egress rate in specific situations.

[1] https://cloud.google.com/compute/docs/machine-types#general_purpose

Please verify the same using higher machine type in GCP end same like AWS side.

[Priyam Vaidya]

Hello,


Thanks Naman for the information.  I was using e2-standard-2 type instance previously. I changed it to n2-standard-2 and it improved the results. But the results on GCP are still slower than one on AWS.  The average response time on GCP takes around 50% more time than the one on AWS.
I want to what might be causing the issue. Is there anything I am missing here?

I can share more information regarding the loatest and results if required.

Thank you.

Regards,
Priyam Vaidya.

[Md (Google Cloud Support)]

Hello, we may require to understand the exact end to end traffic flow path involving all the components in between. You can compare the exact traffic path for both cases to confirm if both are following the same route. Which will eventually help to find out where it's taking time. For this you may run mtr from source to GCP and mtr from source to AWS.

You can also check the same load test without LB in between and they still see higher latency difference,