[cs615asa] Green Team Link Week 6

[Chengzhi Yang]

Hi Everyone,

This week, Green Team focused on the performance improvement of TCP/UDP network.

1. TCP BBR

After our group disscussion, I just recall this magic technology that had helped me a lot for access foreign network before.

BBR, ("Bottleneck Bandwidth and Round-trip propagation time") is a new congestion control algorithm developed at Google. (https://cloud.google.com/blog/products/gcp/tcp-bbr-congestion-control-comes-to-gcp-your-internet-just-got-faster)

This technology can dramatically improvement TCP performance in the environment with packet loss. The biggest improvement case is international transmission usually have more packet loss rate than usual. And it also could suitable for the case which you need to connect with the different data center.

It is one of important technology for international proxy service.

This is the article shows how to enable BBR in Linux(https://www.linuxbabe.com/ubuntu/enable-google-tcp-bbr-ubuntu). It's use Linode VPS as example, but the AWS EC2 should be very similar.


2. A review on TCP/UDP performance for delivering video on an adhoc network.

TCP and UDP deal with lots of problems in the wireless network. TCP retransmission may occur frequently, while UDP may experience tremendous loss. Transport layer protocols experience even more difficulties in an unstructured wireless network such as ad hoc network. The challenge is even bigger when the traffic is a real-time video. This paper reviews TCP, TCP Vegas, and UDP performances in delivering real-time video traffics over an ad hoc network.

https://iopscience.iop.org/article/10.1088/1757-899X/237/1/012018/pdf


3. Performance improvement of TCP via Jumbo Frames

In this short page(http://web.cecs.pdx.edu/~jsnow/wireless_performance/tcp_udp.html), We find UDP could have 50% more throughput than TCP. And TCP has 3 possible methods to improve performance.
a. Fast preambles and headers (Optional part of 802.11b specification)
b. Ack filtering / reconstruction (see Hari Balakrishnan's TCP asymetry paper)
c. Combine multiple packets into one frame

And we focus on the c: Combine multiple packets into one frame. So we got Jumbo Frame (https://en.wikipedia.org/wiki/Jumbo_frame).
Jumbo Frames are the packets which can have more than 1500 bytes of payload, the limit set by the IEEE 802.3 standard.

In this article(https://blog.codinghorror.com/the-promise-and-peril-of-jumbo-frames/), you can have a direct look about network performance by Jumbo Frames in-home network. And the author gives an idea why the Jumbo Frames never enabled by default: "Some networking equipment can't deal with the non-standard frame sizes. “

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.stevens.edu/pipermail/cs615asa/attachments/20190304/36d56d43/attachment.html>