Network Credits
- Instance families such as R4 and C5 use a network I/O credit mechanism
- Most application do not consistently need a high network performance
- These instances perform well above baseline n/w performance during peak requirement
- Make sure that you consider the accumulated n/w credits before doing performance benchmark for instances supporting network I/O credits mechanism
Use cases for high performance networks
- High Performance Computing (HPC) workloads
- Using cluster placement groups with HPC enables access to a low-latency, high-bandwidth network for tightly coupled, IO-intensive, and storage-intensive workloads.
- For faster Amazon EBS IO, it is recommend using Amazon EBS-optimized instances and Provisioned IOPS volumes for high performance.
- Real Time Media
- For applications like VoIP, Media streaming using RTP and RTMP
- Enhanced networking provides smoother packet delivery with less packet loss and jitter
- More use cases: Data Processing, Backup, On-prem Data Transfer, etc.
Summary - Network IO Credits
- For high network bandwidth and throughput, consider using Jumbo Frames, EC2 Enhanced Networking, Placement groups, EBS optimized instances, DPDK
• Instance level network optimization with
• Enhanced Networking (SR-IOV, ENA, Intel VF 82599)
• Placement Groups
• EBS-Optimized Instances
• Operating system level network optimization with DPDK
• EFA is ENA with additional OS-bypass capability which provides further improved network performance for HPC workloads
- Within VPC the MTU size can be 9001 bytes with Jumbo Frames
• For the following cases the maximum MTU is 1500 bytes:
• Traffic over an internet gateway
• Traffic over an inter-region VPC peering connection
• Traffic over VPN connections
• When PPS is bottleneck, increased MTU provides more throughput
• While Enhanced Networking and placement groups lowers the latency between EC2 and hypervisor, DPDK improves the packet processing at OS level
- EC2 network bandwidth depends on many factors including Instance family, size, enhance networking support etc.
• Bandwidth for aggregate multi-flow traffic available to an instance depends on the destination of the traffic
• Within the Region
• Full bandwidth of the EC2 instance
• Within the AWS region you can get up to 100 Gbps bandwidth between EC2 instances or between EC2 and S3 by using multiple-flows
• Outside of the Region, Internet or Direct Connect
• 50% of the available network bandwidth for current generation instances with minimum 32vcpu
• 5 Gbps for the instances with less than 32 vcpu
• Bandwidth for single-flow (5-tuple) traffic is limited to 5 Gbps and can get up to 10 Gbps if the instances are within the Cluster placement group