You are on page 1of 15

Thesis Meeting (30/01/15)

Si Quoc Viet TRANG

Overview

On going work: real implementation of FLOWER


following Networking'15 paper. Objective: to test
over CESARS platform
Analysis of the interaction between FLOWER and
AQM. Objective: assess the effective deployment
of FLOWER (several questions from IETF)

Next step and thesis defense

Brief report about NR-SACKs tests

FLOWER module for Linux


kernel

OWD calculation

FLOWER/LEDBAT use TCP Timestamps


Option tsval (timestamp value) & tsecr
(timestamp echo reply) to estimate the OWD:
owd = tsval tsecr
if the timestamp clocks at both ends are
running at roughly the same rate.

Problem: How can the sender determine the


clock rate HZ of the receiver?

Linux terminologies

Clock rate HZ: The number of clock ticks per


second, is contained in the kernel variable HZ.
Jiffies: The jiffies variable holds the number of
clock ticks since the system booted.
The kernel increments jiffies value HZ times
every second
/include/net/tcp.h
#definetcp_time_stamp((__u32)(jiffies))

Estimation of HZ

Remote HZ:
rhz=HZ*(tsvallst_tsval)/(tsecr
lst_tsecr)

OWD:
owd=tsval*HZ/rhztsecr

FLOWER/LEDBAT reuse the TCP-LP module for the


estimation of HZ that updates continuously the estimated
value of HZ.
Problem: a slight change of estimated remote HZ can cause
an important error in the calculation of OWD

Interaction between LBE and


AQM

LBE with AQM (1)

The bufferbloat problem raises an interest in AQM


and LBE protocols

How LBE protocols (FLOWER/LEDBAT) behave in the


presence of AQM?
We believe AQM studies are of interest for satellite : what
is the impact of bufferbloat at the satellite gateway?

Dario Rossi et al. show that LEDBAT flows become


aggressives when RED is in use

More general : random losses (to be verified)

LBE with AQM (2)

Need for testing the behavior of FLOWER in


the presence of AQM

Preliminary testing

We've reproduced the previous results using the


scripts provided by D. Rossi team
We observe similar behavior but some
discrepancies in the cwnd evolution
FLOWER seems to behave a bit better but not
solving the problem
We are currently figuring out how to tackle this
point and drive other measurements when back
from holidays

Future work

Future work

Finalize the implementation of FLOWER in the Linux kernel

Test FLOWER over a real satellite testbed (CESARS)

Test FLOWER with AQM (in collaboration with Nicolas


Kuhn)

We believe there is an interesting part to dissect here

Propose an IETF-draft before March 2015 in order to


present it at the next IETF meeting in July 2015
Acceptance notification of Networking'15 expected in
March 2015

Thesis defense

End of the thesis expected in October 2015


Manuscript must be written, ready to submit in
June 2015
Sent to the reviewers (TBD) in September 2015
Expected defense between November and
December 2015

Annexe

Testing NR-SACKs (in collaboration with Fan


Yang and Paul Amer)

NR-SACKs

NR-SACKs reduce send buffer blocking and


improve the throughput
TCP with NR-SACKs outperforms normal TCP
especially when the send buffer size is small
and/or the delay is high
Currently test NR-SACKs at ISAE lab need
more tests before testing over the CESARS
platform of CNES

You might also like