PROGRESS project: Quality of Service in In-Home Digital Networks System Architecture and Networking...
-
Upload
clifton-curtis -
Category
Documents
-
view
213 -
download
0
Transcript of PROGRESS project: Quality of Service in In-Home Digital Networks System Architecture and Networking...
PROGRESS project: Quality of Service in In-Home Digital Networks
System Architecture and Networking
4. TCP-MM – extension to TCP-RTM
Affiliation1) Eindhoven University of Technology
Department of Mathematics and Computer Science
HG 6.57, P.O. Box 513, NL-5600 MB, Eindhoven, The Netherlands2) Philips Research NatLab
Prof. Holstlaan 4, 5656AA Eindhoven, The Netherlands
TCP-MM – a real-time transport protocol for multimedia in in-home networks
Author: S. Kozlov 1. Co-authors: Peter v.d. Stok 1,2, Johan Lukkien 1
About the AuthorSergei Kozlov received his M.Sc. in Informatics from the Department of Applied Mathematics of Belarussian State University in 1998.
In 2003 Sergei started a Ph.D. project within the SAN group of Computer Science Department, TU/e in collaboration with Philips Research NatLab
Single packet losses
Bursty losses
Bursty loss size, packets
Minimum pbd to cover losses, mS
1 70
2 260
3 220
4 700
5… -*
2. Evaluation of TCP-RTM [1]
Narrow congestion window problem:
Step 1. Congestion window size (snd_cwnd) prevents delivery of packets after the “hole”. Hence, receiver cannot send immediate acknowledgement.
Step 2. The application blocks at the moment it requests the packets that were lost. The connection freezes untill a retransmission time-out occurs.
Step 3. The application receives a burst of packets, which are out-of-date.
Time-outs grow exponentially with every consecutively lost packet:
frame_size=1024K, frames sent every 20mS
* for loss sizes >= 5 - sending buffer gets overflowed
5. Conclusions- TCP-RTM is not suitable for real-time streaming over networks where bursty losses take place regularly
- “Free congestion window” (FCW) modification helps avoiding long “recovery” times after a bursty loss has occurred
- Concept of FCW was the key extension of TCP-RTM, which lead to the creation of the proposed TCP-MM (multi-media) protocol
- Play-back delay needed to resent the lost packets with TCP-MM depends linearly on the burst size
1. Introduction Digital networking is a fundamental building block of the Ambient Intelligence environment. A crucial point here is the effective use of network resources. As part of the task to enable multimedia streaming over the network we investigate real-time transport protocols and their applicability for wireless networks.
TCP-RTM vs TCP-MM
0
100
200
300
400
500
600
700
800
1 2 3 4 5 6 7 8 9 10
Play-back delay, mS
Burst size, packets
Receiving application
?
Receiving application
Receiving application
re-request
step1
step2
step3
? ?
!
resent
packet consumed by application packet received but not consumed by application
skipped packet empty buffer
recovered packet
Reference: [1] TCP-RTM: Using TCP for Real Time Multimedia Applications. Sam Liang, David Chariton. Distributed Systems Group, Stanford University. 2003
The figure shows that in case of TCP-MM recovery play-back delay is linearly (versus quadratic for TCP-RTM) bounded to the burst size.
3. Loss properties of wireless networks
Observations and conclusions:
- Bursty losses of size up to 80 packets occur regularly (e.g. because of “hand-overs”)
- An adaptation of TCP-RTM should be made to allow multimedia streams
Measurements set-up: infrastructure wireless networks by NatLab, Philips Research Eindhoven
Conclusions
- TCP-RTM performs well on networks with non-bursty losses…
- … but bursty losses make TCP-RTM hardly applicable
“Skip-over-hole” technique:
Step 1. While there are packets for the application to read, “holes” can still be filled up.
Step 2. When the application meets a “hole”, TCP-RTM accomplishes a “skip-over-hole” – it delivers the next arrived packet from out-of-order queue. In meanwhile, it acknowledges the lost packet to the receiver.
Step 3. The application has to handle the packet loss on application level
Assumptions on the network conditions:
- Only occasional single packets are lost
More than 90% of the holes get filled when the assumptions are met** with: pbd=250mS, rto<60mS, loss rate <10%
empty buffer
Receiving application
?
Receiving application (blocked)
Receiving application
re-request on rto
step1
step2
step3
? ?
snd_cwnd
?? ?
out-of-date
packet consumed by application
received out-of-date packet
packet received but not yet consumed by application
“Free congestion window” (FCW) concept
Step 1. Number of packets-on-air is not limited with snd_cwnd, hence the packets after the hole are not prevented from being sent. It allows receiver acknowledge immediately.
Step 2. To the moment when the application requests lost packets, those can be (partially) resent.
Step 3. The application continues receiving timely arrived packets.
Measurements conditions- Controlled packet losses are introduced in the kernel of the sending machine
-Clocks at the machines are synchronized at the moment when the connection is established
Receiving application
?
Receiving application
!
Receiving application
Too late to resend - acknowledge
re-request
step1
step2
step3
packet consumed by application packet received but not yet consumed by application
skipped packet empty buffer
snd_cwnd