1 / 13

Improving Slow Start

Improving Slow Start. InKwan Yu. Slow Start?. Slow Start Grows the congestion window size exponentially. Terminates when ssthresh (64K) is reached or a packet is lost Three cases of Slow Start Initial : when a TCP connection is established Idle : packets are not sent for a few minutes

milly
Télécharger la présentation

Improving Slow Start

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Improving Slow Start InKwan Yu

  2. Slow Start? • Slow Start • Grows the congestion window size exponentially. • Terminates when ssthresh (64K) is reached or a packet is lost • Three cases of Slow Start • Initial : when a TCP connection is established • Idle : packets are not sent for a few minutes • Timeout : transmission timeout

  3. Related Work • Estimating Bandwidth • Hoe’s Approach • First packet pair to estimate ssthresh • RTT / Inter packet gap = cwd (initial congestion window size or ssthresh) Network Inter Packet Gap s t

  4. Related Work • Caching • Source has cache of ssthresh values of destinations • When a new connection is from a cached destination, use the ssthresh of the cache.

  5. Related Work • Hu’s Method • Congestion : input inter packet gap is smaller than output packet gap • No congestion : input inter packet gap is larger • Binary search using about 3 to 4 rounds

  6. Related Work • Hu’s Method

  7. Goals • Faster Start • In a large bandwidth*delay path, slow start is not fast enough due to TCP’s self clocking • Better ssthresh value • Overestimated ssthresh value will cause packet drops

  8. New Methods • History based • With fair TCP, the equal amount of bandwidth will be shared 1Mb s t 512Kb s t 512Kb

  9. New Methods • History based • When there are already k connections, a new connection will have B*1/(k+1) share from B total available bandwidth 1/k …… …… s t 1/k ?

  10. New Methods • History based • Extension to a subnet, inside of which delay is negligible and there is no congestion • There is a common path from the source to the subnet t Subnet t s Host Host t

  11. New Methods • Estimation based • Weakness of Hu’s method • The binary search method can be misleading with the first few errors • The cwd growth rate is to the 2nd power (slow) • The delayed acknowledgement • Modification • Regression based method • Confidence interval • Sequence numbers of packet pair are reversed

  12. New Methods • Estimation based y=x Arriving Inter Packet Gap (ACK gap) Departing Inter Packet Gap

  13. Reference • Ningning Hu, P. Steenkiste, Improving TCP startup performance using active measurements: algorithm and evaluation, ICNP 2003. • J. Hoe, Improving the start-up behavior of a congestion control scheme for TCP, SIGCOMM 1996.

More Related