Seeking Optimal AWS Networking Solution for Fastest Transatlantic EC2 Communication
Hello AWS Community,
I am seeking advice on determining the best possible networking solution to facilitate the fastest communication between two EC2 instances in different VPCs, specifically for transatlantic communication, using both TCP and Multicast.
My current setup involves two EC2 instances each located in different VPCs and different regions (North America, Virginia and Europe London). I am currently using VPC peering to establish connectivity between these instances. However, I am experiencing significant latency and jitter issues when transmitting data across the Atlantic using TCP, which is impacting my application's performance.
From my observations using Wireshark, I've noticed that packets are being buffered for 2-5 milliseconds from the time they arrive to the time (packet in NIC) they are consumed (application timestamping).
I am also exploring a multicast solution using a third-party provider, but I'm finding this to be challenging between AWS regions.
I am looking for solutions that will not only reduce the overall transmission latency but also the jitter in packet arrival times. From my initial research, I've learned about the AWS Global Accelerator, AWS Direct Connect, Amazon CloudFront, and Inter-Region VPC Peering. While these seem like promising solutions, I am uncertain about which would be the most effective for my use case, or if there is a better solution I haven't considered yet.
My requirement is simple: I need the fastest, most consistent data transfer between my EC2 instances across the Atlantic. Any advice on the pros, cons, and considerations for these solutions (or others) would be much appreciated. Also, if anyone has benchmarks or real-world experiences with these solutions for transatlantic data transfer using TCP or Multicast, it would be extremely helpful.
Thank you for your help and looking forward to your insights!
Best Regards, Yoni
- Newest
- Most votes
- Most comments
It would be (quite!) difficult to avoid using the Amazon backbone network when communicating between regions.
VPC peering is definitely the lowest latency connection type that you can use; the additional latency you're experiencing is probably due to AWS encrypting all traffic between regions.
If you moved to using public IP connectivity (by putting an Internet Gateway in each VPC and communicating using Elastic IPs in each region) the traffic would still use the AWS backbone (this is called out in VPC FAQ). It's the same network with the same performance characteristics.
Given that the communication you're doing is between two EC2 instances (in separate regions), services such as CloudFront and Global Accelerator are not appropriate here as they are more for internet-based clients connecting to your applications in AWS. Direct Connect is also not a useful service as that is for private connectivity from on premises networks to AWS.
If you are experiencing packet loss and latency above what you'd expect, I strongly encourage you to create a support case so that the network team can troubleshoot this with you.
Relevant content
- asked 3 months ago
AWS OFFICIALUpdated 8 months ago- AWS OFFICIALUpdated a year ago
- AWS OFFICIALUpdated a year ago
thanks a lot for your response, i am trying to mitigate the jitter cause indeed unsuccessfully It does not seem like decryption latency because it is not a constant latency rather then a jitter as you can see in the plot i added to the question
i also tried to see if the jitter is a cause of packet loss but thats also does not seem the cause we are now working to be able to open a support case thanks!