Re: XTRA TNSM review

[Michael Welzl < >]

Hi,

I have two ideas.

First: to get one more diagram, but on something that may not be very relevant, we could just re-introduce a diagram that we had in the SIGCOMM submission, which showed that TBTCP produces less latency (in one case) than Linux TCP. That was fig. 5 in the SIGCOMM paper. We stopped using this because we didn’t want to stress how good TBTCP is, as this was distracting from the main story and in a way even a disadvantage ("you had to make TCP so different so you could port it?”). But at this point, if we urgently need one more plot, using this one is a possibility.

Second, and I think that’s more useful: I remember reviewers (maybe for CoNext?) asking for throughput plots. Indeed, with FPGA we should be able to send faster than with a pure software implementation…  wouldn’t it make sense to make a raw throughput plot of how fast one TCP can go, on the different platforms? That would show how beneficial it can be to port TCP to hardware.

Maybe that’s also not a fantastic idea… but it’s the best I could come up with.

Cheers,

Michael

On Jul 11, 2019, at 2:17 PM, Marco Faltelli < " class=""> > wrote:

Dear all,

yesterday me and Angelo had a short talk about the XTRA review and we agreed that question 2.3 is the most critical one: other performance tests are required and right now we don't have any idea of how to solve this issue. We suggest a short meeting in the next days in order to discuss about it together. Michael, your suggestions are welcome if you have any!

Cheers

Marco

--

Dr. Marco Faltelli

Researcher @CNIT

University of Rome "Tor Vergata"

~ e-mail: " target="_blank" class="">

~ LinkedIn: https://www.linkedin.com/in/marcofaltelli/

~ Google Scholar: https://scholar.google.it/citations?user=joXyxfoAAAAJ&hl=it