Let's drop everything

My main goal for this year is:

1.) Modify SCTP protocol to use other switching policies.
2.) Change all graph into gnuplot graphs.
3.) Create scripts to automate graph creation.

Nahagilap na random vs congestive loss techniques

1.) TCP Cerl (2010)
2.) TCP Westwood Family (2002-2004?)
3.) TCP Veno
4.) TCP-NCL (2010)

Start of CS 300 Thesis

SCTP Thesis!

Objective:
1. Create a host-to-host delay based scheme for you know.
2. Based on the result, implement the best switching policy.

Need to address:

1. Popular techniques on distinguishing congestive vs random loss.
2. Study GNU Plot
3. How to compute the throughput

Dalawamputisa

Thesis proposal is almost finished.

TODO:

1. My primary problem is the Proposed scheme (chap 4.1)

2. Figures.

3. Insert citation

4. Proof read

STOP

tigil blog muna. next sem nalang ulit. :)

Ikalabingsiyam

packet probing for bandwidth estimation:
packet-pair?
active probing?
meron bang triple probe?

Ikalabingwalo

A reliable byte stream is a common service paradigm in computer networking; it refers to a byte stream in which the bytes which emerge from the communication channel at the recipient are exactly the same, and in the exact same order, as they were when the sender inserted them into the channel.

A reliable stream is not the only reliable service paradigm which computer network communication protocols provide, however; other protocols (e.g. SCTP) provide a reliable message stream, i.e. the data is divided up into distinct units, which are provided to the consumer of the data as discrete objects.

------------
* Larry L. Peterson and Bruce S. Davie, Computer networks: a systems approach, 3rd edition, Morgan Kaufmann Publishers, 1996, Section 6.2.

* Steve Steinke, Network Tutorial, Elsevier, 2000, page 163.

Ikalabingpito

my proposed research topic:
'Effective Multihoming in Mobile Networks'
-Efficient rules on switching
-Minimize false degradation of congestion window due to fading, blackout, etc.
-Minimize spurious retransmissions and timeouts ?

read 'SCTP Applications' (2003)
this is OK

Multihoming for availability and fail-over

Congestion and switching in multihomed connections is still for further research, specially on mobile networks.

We can extend or combine different congestion control approaches from TCP and use them on multihomed SCTP

As of now, multihoming is used for availability and fail-over, not CMT

Why use multihoming?
Why not mSCTP?
Why not CMT? -> bandwidth aggregation (link aggregation)
Why switch from TCP to SCTP?
What are the advantages of using SCTP on mobile networks?

From RFC4960 (2007)
"The limited scope of TCP socets complicates the task of providing highly-available data transfer capability using multi-homed hosts"

"When an endpoints peer is multihomed, the endpoint will calculate a separate RTO for each different destination transport address of its peer endpoint"

"When its peer is multihomed, an endpoint SHOULD try to retransmit a chunk that timed out (not 3 or 4 dupacks) to an active destination transport that is different to the last destination address to which the DATA chunk was sent"

"When its peer is multihomed, the SCTP endpoint should always try to send the SACK to the same destination address from which the last DATA chunk was received"

"SCTP sender MUST keep a set of these control variables cwnd, ssthresh and partial_bytes_acked for EACH destination address of its peer (when multihomed). Only one rwnd is kept for the whole association (no matter if the peer is multihomed or has a single address)."

PS:
Eto sana ung techniques pag topic is TCP:

X How slow-start ends?
Change error threshold to n (eg. n=3) before going to congestion avoidance stage.

X Decrease of cwnd & sshthresh - di dapat ganun (eg. Reno, SACK). Must use probing to check "blackout", "handover".

X For fading, interference, RTT computations must be done.

X Probes should be greater than 2. (eg. NOT packet-pair algorithm)

Ikalabinganim

read 'Congestion and Flow Control in the Context of the Message-Oriented Protocol SCTP' (2009)
medyo di kelangan pero ok na rin

read 'Performance Improvement of SCTP for Heterogeneous Ubiquitous Environment' (2007)
some good implementation discussions are shown here.

need to find more articles on congestion control on multihomed connections.

Ikalabinglima

Read stuff about Reno, New Reno, and SACK

don't need MobileIP. Its used to maintain a permanent ip address on a mobile node moving to another network.

still have a dilemma on thesis topic :(

about use of sctp multihoming:
http://www.ietf.org/mail-archive/web/tsvwg/current/msg05084.html

latest SCTP RFC (2007):
http://tools.ietf.org/html/rfc4960

Ikalabingapat

Uses of multihoming:
mobility, availability, load balancing

Ikalabingtatlo

read 'SCTP Congestion Control: Initial Simulation Studies' (2003)
-describes SCTP congestion control
-differentiates congestion control between SCTP and TCP

read 'SHOP: An Integrated Scheme for SCTP Handover Optimization in Multihomed Environments' (2008)
-An SCTP endpoint maintains one receiver window (rwnd) for the entire association, while other variables, such as congestion window (cwnd), slow-start threshold (ssthresh) and RTT, are kept on a per destination address.

-handover ="a procedure in which an endpoint replaces the primary address with a secondary one, as well as redirects ongoing flows from the original path (towards old primary address) to the new one (towards new primary address)."

read 'SCTP Switchover Performance Issues in WLAN Environments' (2008)
-"Switchover" is the same as "Handover" in the previous paper

a useful wiki page:
http://en.wikipedia.org/wiki/SCTP_packet_structure



(My new thesis topic is something like "Enhancing througput of multihomed SCTP connections on mobile networks")

Ikalabingdalawa

My adviser's advice on SCTP, why SCTP? why Multihoming? check these thesis in library:
Ibabao, Joana EEE
Villacorta, Rachel EEE
Balagtas, Florence CS
Tiglao, Nestor EEE

read 'Experimental studies of SCTP multi-homing' (2001).
-it explains some rules specified in RFC 2960.
-ns2 simulation using 2<->2 IP. single-homed, multihomed, w/ w/o packet losses.

read 'Retransmission policies for multihomed transport protocols' (2006)
-AllRtxAlt and AllRtxSame
"Without a priori knowledge about the available paths, a sender cannot have a static policy that decides where to retransmit lost data and expect to guarantee the best
performance. Through simulation, we have measured and demonstrated the tradeoffs of three policies in non-failure and failure conditions"

Ikalabingisa

Decided to work on Congestion control of SCTP's multihoming in mobile networks.

On 'Host-to-Host Congestion Control for TCP' (2010), the authors suggested researching on this.

Ikasampu

Got to present my ideas on panel. Panel suggests:
1. mobile IP or not? -> medyo di ko gets.
2. "Markings" of packets before hand-off/etc. (like ECN?)
3. Cross Layer communications
4. Try "patch" to different TCP standards.

i tried to differentiate my technique to TCP-Probing. Here are my findings. The advantage of my protocol from TCP-Probing is:
1. Simple
2. Better througput (goodput)
3. Sender side modification only.
4. No need to modify TCP header. Don't use other TCP options (eg. Timestamp, SACK)
5. The algorithms can be "patched" to other TCP standards.


ps: Need to change my mindset about "probing". Mine is not really probing. It just sends packets to "check" why packets are lost.

Ikasiyam

On 'Host-to-Host Congestion Control for TCP' (2010)
will read on related topics:
1. packet reordering as congestion control
-eifel*
-Door
-PR
-RR
-TD-FR
2. UCLA research on wireless
-TCP Westwood
-Westwood+
-ABSE
-CRB
-BBE
-BR

Ikawalo

Googled TCP Westwood and TCP Vegas.

Sent an email about changing the topic to sir cedric. :)

read 'The Eifel Algorithm: Making TCP Robust Against Spurious Retransmissions' (2000)
the technique uses timestamp option in the tcp header to eliminate "spurious" retransmissions.

"The sender always stores the timestamp of the first retransmission independent of whether that was triggered by an expiration of the retransmission timer or by the receipt of three consecutive DUPACKs5. In our implementation, we call that timestamp ts_first_rexmit. Then, when the first ACK that acknowledges the retransmission arrives, the sender compares the timestamp of that ACK with ts_first_rexmit. If it is smaller than ts_first_rexmit, this indicates that the retransmission was spurious."

Ikapito

created a ppt presentation about TCP. https://docs.google.com/present/edit?id=0AecHgxabb_zmZGc5Y3doYnFfMzF4YmtmNXJjcA&hl=en
still not talking to my adviser about changing of thesis topic.

started reading High Performance TCP/IP Networking: Concepts, Issues, and Solutions. Mahbub Hassan, Raj Jain (2003)
there are 3 interesting mathematical models in chapter 5 that i can use in my thesis.
Periodic Model -> a simple model that predicts throughput (packet/s ?)
Detailed Packet Loss Model -> a more accurate periodic model.
Stochastic Model with General Loss Process -> a more complex model that uses a new loss process to calculate throughput.

i need to read the math over and over again to understand how the model is built. i think they are relatively easy mathematical models.

Ikaanim

read 'Behavior of TCP-Probing with Hand-offs' (2001). Hand-offs occur (specially in GSM networks)when one mobile node connected on a fix node changes its location where its signal is not covered by the fixed node. The fix node will then "hand-off" its connection to the nearest fixed node that covers the mobile node.

read 'Improving Reliable Transport and Handoff Performance in Cellular Wireless Networks' (1995) -> di kelangan. BOO

read 'Improving TCP/IP Performance over Third Generation Wireless Networks' (2008)
di rin kelangan bano.

read 'Transport Layer Design in Mobile Wireless Networks' (2005)
SCTP is thoroughly discussed here.
in TCP , one notable flavor here is TCP-Peach, yet another probing technique. TCP-peach is for Satellites. So there's no need to read on this.

another important figure is Table 4.2 Comparison of different TCP enhancement schemes.

Ikalima

read 'TCP-Probing: Towards an Error Control Schema with Energy and Throughput Performance Gains' (2000)
this is where i'll start. My protocol and algo is almost like the paper's technique in tcp-probing. i will make a new or modified type of this probing technique. it will be more complicated and my findings must be superior from the old one. i must research extensively on tahoe, reno, new reno. Probing must be an add-on to many tcp flavors.

will integrate randomization and probabilities to determine the nature of error.

will i consider energy efficiency? will i consider tcp-probing on receiver's side?

will i test on many users? will i check fairness, convergence, etc?

must research on testing.

Ikaapat

read 'A Comparison of Mechanisms for Improving TCP Performance over Wireless Links' (1997)
the techniques are divided in 3 types:
1. Link-layer mod
2. Split-connection protocols -> (eg. snoop)no
3. End to end
We will focus only on #3
SACK
SMART -> improved SACK
ELN (explicit loss notification) -> flag on TCP. used for knowing type of error.
ELN with retransmit on first dupack

Ikatlo

Continue reading 'Open Issues on TCP for Mobile Computing'

Error recovery:
Freeze-TCP -> avoids timeouts by sending a ZWA from receiver.
TCP probing :)
TCP Santa Cruz
TCP-Real
DAKS delayed acks
fast retransmission
TCP New Reno
TCP-SACK selective

No interest in proxy-based approaches:
Indirect-TCP
MTCP
Explicit Bad State Notification
WTCP
Snoop

Protocol Strategy
PROBING!
high goodput, energy-efficient

END

Read 'An End-to-End Transport Protocol for Extreme
Wireless Network Environments' (2006)

it extends TCP-sacks technique by incorporating:
ECN -> so the routers/base stations must be ECN enabled
uses FEC (Forward error correction) -> add redundant data into packet.
changing MSS (Maximum segment size) -> cwnd

Ikalawa

1. read 'Analysis of reliable transport layer protocols in wireless networks' (2006)
this is just some report of a random student but its references are OK

keywords:
I-TCP
Snoop-TCP
NS2

basic issues of TCP over wireless networks are mentioned here. I-TCP and Snoop-TCP are 'middleware' apps on TCP, thus not end to end. They buff the packets to increase throughput.

2. read 'Open Issues on TCP for Mobile Computing' (2002/2003)
broad info on problems of TCP over mobile wireless networks

TCP tahoe: slow start, congestion avoidance and fast retramsmit
TCP reno: fast recovery w/ fast retransmit
TCP vegas: good rtt measurements.

Research issues:
detect nature of error -> fading channel, congestion, handoff, burst errors
must be energy-saving due to constrained devices
ECN -> flag in IP
RED -> random early detection.
End-to-end argument -> where to place desired function

Error Detection:
TCP-probing
WTCP -> base station buffs data received
levels of congestion->
"The efficiency of such measurements on the congestion level has two dimensions: how to measure congestion, and where to measure it (i.e., at the sender or the receiver). In the next section we extend the discussion on the latter, since it crucial for the error recovery tactics of TCP. As for the former, whether an estimation of congestion should be based on RTT measurements, on the packet-pair approach, on the wave pattern, or on single probes is an open issue of further research and experimentation."
TCP-Real

Error Recovery:

Una

Time to pick a thesis topic.
i'm interested in topics which are something about:
reliable transfer on mobile ad-hoc networks
reliable transfer on wireless networks
reliable multicast
end to end reliability in wireless networks

the title is something like improving/enhancing end to end reliability in mobile networks.

in wireless mobile networks, packet loss can occur due to weak connection signal and interference. TCP only recognizes packet loss as a result of congestion.

Mobile:
SMS
GPRS
HSDPA
wifi
etc.

=> will research on these.

How to test:
Omnet++
java mobile -> emulator

How to prove that my protocol is correct.
-math!
-tests

As of now, my knowledge is limited to:
tcp stuff
convergence
fairness
rtt
who will initiate retransmits?
ordering? -> bat sa sctp optional

research pa sa tcp variants.
other end to end techniques for reliable transfer
checksum, hash, etc.

Papakialaman pa ba ang underlying networks?
multicast
manets
mesh
routing
gsm