Ultimate Network Performance for Cluster Computing Application

Last Optimization for Network Performance

Prev Question Next Question

Question

You need the absolute highest possible network performance for a cluster computing application.

You already selected homogeneous instance types supporting 10 gigabit enhanced networking, made sure that your workload was network bound, and put the instances in a placement group.

What is the last optimization you can make?

Answers

Explanations

Click on the arrows to vote for the correct answer

A. B. C. D.

Answer - A.

Jumbo frames allow more than 1500 bytes of data by increasing the payload size per packet, and thus increasing the percentage of the packet that is not packet overhead.

Fewer packets are needed to send the same amount of usable data.

However, outside of a given AWS region (EC2-Classic), a single VPC, or a VPC peering connection, you will experience a maximum path of 1500 MTU.

VPN connections and traffic sent over an Internet gateway are limited to 1500 MTU.

If packets are over 1500 bytes, they are fragmented or dropped if the Don't Fragment flag is set in the IP header.

For more information on Jumbo Frames, please visit the below URL:

http://docs.aws.amazon.com/AWSEC2/latest/UserGuide/network_mtu.html#jumbo_frame_instances

Option B is INCORRECT as it is not a recommended way to segregate the instances into different peered VPCs.

Option C is INCORRECT as there is no necessity for creating an AMI for the instances.

Option D is INCORRECT as turning off the SYN/ACK will not help optimize the throughput here.

In the given scenario, the highest possible network performance is required for a cluster computing application. The steps that have been already taken to optimize network performance include selecting homogeneous instance types with 10 gigabit enhanced networking, making sure that the workload is network-bound, and placing instances in a placement group.

The last optimization that can be made is to use Jumbo Frames with an MTU (Maximum Transmission Unit) of 9001 instead of the default 1500 MTU. Jumbo Frames are packets that have a larger payload than the standard Ethernet frame. By increasing the MTU to 9001, the packet body to packet overhead ratio is increased, which reduces the overhead of each packet and increases the overall network throughput.

Option B, which suggests segregating the instances into different peered VPCs, does not help in improving the network performance. It may even add additional overhead and network latency due to the need to route traffic between different VPCs.

Option C, which suggests baking an AMI for the instances and relaunching them, may help in ensuring that the instances are fresh and do not have noisy neighbors. However, it does not provide any direct optimization for network performance.

Option D, which suggests turning off SYN/ACK on the TCP stack or using UDP for higher throughput, is not a viable option. Turning off SYN/ACK can result in packet loss and data corruption, and using UDP does not guarantee reliable data transfer.

Therefore, the correct answer is A, to use 9001 MTU instead of 1500 for Jumbo Frames to raise packet body to packet overhead ratios.