DOC PREVIEW
Rose-Hulman CSSE 432 - Congestion control

This preview shows page 1-2-3-4 out of 11 pages.

Save
View full document
View full document
Premium Document
Do you want full access? Go Premium and unlock all 11 pages.
Access to all documents
Download any document
Ad free experience
View full document
Premium Document
Do you want full access? Go Premium and unlock all 11 pages.
Access to all documents
Download any document
Ad free experience
View full document
Premium Document
Do you want full access? Go Premium and unlock all 11 pages.
Access to all documents
Download any document
Ad free experience
View full document
Premium Document
Do you want full access? Go Premium and unlock all 11 pages.
Access to all documents
Download any document
Ad free experience
Premium Document
Do you want full access? Go Premium and unlock all 11 pages.
Access to all documents
Download any document
Ad free experience

Unformatted text preview:

09-Congestion controlTransport Layer3-1Approaches towards congestion controlpp gTwo broad approaches towards congestion control:End-to-end congestion control:Network-assisted congestion control: no explicit feedback from network congestion inferred from  routers provide feedback to end systems bits indicating gffmend-system observed loss, delay approach taken by TCPgcongestion  explicit rate sender should send atpp yTransport Layer3-2Case study: ATM ABR congestion controlygABR: available bit rate: RM (resource management)  “elastic service”  if sender’s path “underloaded”: (g)cells: sent by sender, interspersed with data cellsunderloaded :  sender should use available bandwidthif sender’s path with data cells bits in RM cell set by switches (“network-assisted”) RM cells returned to sender by if sender s path congested:  sender throttled to minimum guaranteed RM cells returned to sender by receiver, with bits intactminimum guaranteed rateTransport Layer3-3Case study: ATM ABR congestion control EFCI (explicit forward congestion indication) bit in data cells: set to 1 in congested switch if data cell preceding RM cell has EFCI set, sender sets CI bit in returned RM cellCI( ti iditi) d NI( i ) bit i RM llCI(congestion indication) and NI(no increase) bits in RM cells Switch sets NI to 1 (mild congestions) CI bit to 1(severe congestion)twobyte ER (explicit rate) field in RM cell3-4two-byte ER (explicit rate) field in RM cell congested switch may lower ER value in cell sender’s send rate thus minimum supportable rate on pathTCP Congestion Control end-end control (no network TCP Congest on ControlHow does sender (assistance) perceive congestion? loss event = timeout or3 d li t ksHow does sender limit the rate?CongWin(congestion window)3 duplicate acksThree mechanisms to CongWin (congestion window)is dynamic, function of perceived network congestionThree mechanisms to adapt CongWin: AIMDl  LastByteSent-LastByteAcked≤ CongWin slow start conservative after timeout eventsTransport Layer3-5TCP AIMDmultiplicative decrease:cut CongWin in half ft l tadditive increase:increase CongWin by 1 MSS RTT i congestionwindowafter loss event1 MSS every RTT in the absence of loss events24 Kbyteswindow8 Kbytes16 Kbytes8bytestimeTransport Layer3-6Long-lived TCP connectionTCP Slow StartTCP Slow Start When connection begins, When connection begins, g,CongWin = 1 MSS available bandwidth may b MSS/RTTWg,increase rate exponentially fast until first loss eventbe >> MSS/RTT desirable to quickly ramp up to respectable ratefirst loss event For every successful ACK, send two packets.Transport Layer3-7TCP Slow Start (more)TCP Slow Start (more) When connection Host AHost Bbegins, increase rate exponentially until first loss event:Host ARTTHost Bfirst loss event: double CongWin every RTTd b i ti done by incrementing CongWin for every ACK receivediiil  Summary:initial rate is slow but ramps up exponentially fasttimeTransport Layer3-8pyRefinementRef nement After 3 dup ACKs: (AIMD)•3 dup ACKs indicates Philosophy: CongWin is cut in half window then grows li l•3 dup ACKs indicates network capable of delivering some segments bf d linearly But after timeout event:CongWininstead set to •timeout before 3 dup ACKs is “more alarming”CongWininstead set to 1 MSS;  window then grows exponentially to a threshold, then grows linearlyTransport Layer3-9linearlyRefinement (more)Q: When should the exponential i ith t 1214size increase switch to linear? A: When CongWint t 1/2 f it 681012tion window ssegments)thresholdgets to 1/2 of its value before timeout.024123456789101112131415congest(sImplementation: Variable Threshold At l t Th h ld i 123456789101112131415Transmission roundAt loss event, Threshold is set to 1/2 of CongWin just before loss eventTransport Layer3-10S TCP Cti CtlSummary: TCP Congestion ControlWhen CongWinis below Threshold sender in When CongWinis below Threshold, sender in slow-start phase, window grows exponentially.When CongWinis above Threshold sender is in When CongWinis above Threshold, sender is in congestion-avoidance phase, window grows linearly.When a triple duplicate ACKoccurs ThresholdWhen a triple duplicate ACKoccurs, Thresholdset to CongWin/2 and CongWin set to Threshold. When timeout occurs, Threshold set to CongWin/2 and CongWin is set to 1 MSS.Transport


View Full Document

Rose-Hulman CSSE 432 - Congestion control

Download Congestion control
Our administrator received your request to download this document. We will send you the file to your email shortly.
Loading Unlocking...
Login

Join to view Congestion control and access 3M+ class-specific study document.

or
We will never post anything without your permission.
Don't have an account?
Sign Up

Join to view Congestion control 2 2 and access 3M+ class-specific study document.

or

By creating an account you agree to our Privacy Policy and Terms Of Use

Already a member?