1 / 42

FAST TCP

FAST TCP. Speaker: Ray Veune: Room 1026 Date: 25 th October, 2003 Time: 10:00am. Motivation. Demand for ultrascale networking HENP (High Energy and Nuclear Physics) Data volumes of tens of Petabytes (10 15 ) to Exabytes (10 18 ) Require Terabit/sec (10 15 bit/sec or 1000Gbit/sec)

hieu
Télécharger la présentation

FAST TCP

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. FAST TCP Speaker: Ray Veune: Room 1026 Date: 25th October, 2003 Time: 10:00am

  2. Motivation • Demand for ultrascale networking • HENP (High Energy and Nuclear Physics) • Data volumes of tens of Petabytes (1015) to Exabytes (1018) • Require Terabit/sec (1015 bit/sec or 1000Gbit/sec) • Scalability problem of TCP • Losses must be extremely rare • TCP must induce loss • Underutilization and oscillation

  3. Scalability problem of TCP extremely loss packet loss possibility • Rate = 1.3 * MTU / (RTT * sqrt(Loss)) • MTU = 1500bytes, RTT = 10ms

  4. Scalability problem of TCP inevitable packet loss • TCP needs to create losses • Single bit network feedback signal

  5. Scalability problem of TCP Underutilization and oscillation • AIMD (1, 0.5) • At large window size (in excess of 10,000 pkts): • Halving window on loss event is too drastic • Increasing window by one packet per RTT is too conservative

  6. FAST TCPAchievement • CERN (European Organization for Nuclear Research) sent 1.1 Terabytes of data at 5.44 Gbps • Full-length DVD film in 7 seconds !!

  7. FAST TCP • Flow based vs Packet based • Network delay vs Packet loss • TCP-Vegas vs TCP-Reno • Stabilized Vegas

  8. TCP VegasTechniques • New Retransmission Mechanism • Congestion Avoidance Mechanism • Modified Slow-Start Mechanism

  9. TCP VegasNew Retransmission Mechanism • Timeout • n duplicate ACKs

  10. TCP VegasNew Retransmission Mechanism • Check time record for the first duplicate packet • non-duplicate ACKs first or second after retransmission • Catch other segment lost previous to retransmission

  11. Source Router Dest. TCP VegasCongestion Avoidance Mechanism • Detect network delay by monitoring RTT • BaseRTT and ActualRTT

  12. Diff α β TCP VegasCongestion Avoidance Mechanism • Expected = WindowSize / BaseRTT • Diff = Expected - Actual • Diff >> 0, decrease sending rate • Diff = 0, increase sending rate • α< Diff < β

  13. TCP VegasCongestion Avoidance Mechanism • Extra buffers occupied • BaseRTT: 100ms, segment size: 1KB Expected = WindowSize / BaseRTT α = 30KB/s, β=60KB/s α=> 30KBps * 100ms / 1KB = 3 β=> 60KBps * 100ms / 1KB = 6

  14. TCP VegasCongestion Avoidance Mechanism • α=β • Diff <α • increase one segment per RTT • Diff =α • no change in windows size • Diff >α • decrease one segment per RTT

  15. TCP VegasSlow-Start Mechanism • TCP-reno • Send two segment for each ACK received • Exponential growth every RTT • TCP-Vegas • Exponential growth every alternative RTT • γthreshold • Diff >γ • Changes from slow-start mode to linear I/D mode

  16. Stability of TCP Vegas Network model • Set of L links with finite capacities c • c = (cl , l  L) • N sources indexed by r • Each source r uses a set of link defined by the LN routing matrix Rlr = { • if source r uses link l • 0 otherwise

  17. lr lr Stability of TCP VegasNetwork model • For each link l, the congestion measure pl(t) is call price • For each source r, it maintains a rate xr(t) in packets/sec • Equilibrium forward delay from source r to link l :  • Equilibrium backward delay from link l to source r : 

  18. lr lr Stability of TCP VegasNetwork model • Aggregate price source r observes in its path • qr(t) =  Rlr pl (t -  ) • Aggregate source rate link l observes • yl(t) =  Rlr xr(t -  ) x1(t) p1(t) p3(t) p4(t) x2(t) p2(t) l r

  19. lr lr Stability of TCP VegasNetwork model • Tr denote equilibrium RTT  +  = Tr,  l  L • Dynamical system of TCP Vegas pl (t) = ( yl ( t ) –cl ) / clif pl (t) > 0 ( yl ( t ) –cl ) / cl )+if pl (t) = 0 xr (t) = 1/Tr2(t) sgn( 1 –xr(t)qr(t) / rdr ) Tr (t) = dr + qr( t ) Where sgn(z) = 1 if z > 0, -1 if z < 0 and 0 if z = 0 (z)+ = max { 0 , z }

  20. Stability of TCP Vegas Approximate model • xr (t) = 1/Tr2(t) sgn( 1 –xr(t)qr(t) / rdr ) • sgn(z)  2/ tan-1 (z) •    • xr (t) = (2/Tr2(t))tan-1(1 –xr(t)qr(t) / rdr )

  21. Stability of TCP Vegas Approximate model • In equilibrium, the source rate xr* and aggregate price qr* satisfy xr* qr*= r dr

  22. Stability of TCP Vegas Theorem 1 • Suppose for all r, k0Tr maxrTr for some k0. • Let M be an upper bound on the number of links in the path of any source, M  maxrlRlr. • The Vegas model is locally asymptotically stable around the equilibrium point (xr* , yl* , pl* , qr* ) if maxr xr*Tr sinc  (ň / xr*Tr ) <  / Mk02 • ň = 2/ • Let (a) be the unique solution in ( 0, /2) of  tan = a as a strictly increasing function of a • sinc = sin /

  23. Stability of TCP Vegas Theorem 1 • maxr xr*Tr sinc  (ň / xr*Tr ) <  / Mk02 • () is strictly increasing • sinc() is strictly decreasing • LHS is strictly increasing in windows size xr* Tr • Theorem 1: Stability condition impose a limit on max windows size

  24. qr* /Tr minr > Mk02 ň . sinc  ( ) qr*  Tr Stability of TCP Vegas Corollary 2 • maxr xr*Tr sinc  (ň / xr*Tr ) <  / Mk02 • All source has the same target queue length, r dr = for all r • Corollary 2: LHS is strictly increasing in qr* / Tr , implying a lower bound on queueing delay

  25. Stability of TCP Vegas Corollary 3 • Since () <  / 2, sinc () > 2 / , k0  1 • Corollary 3: minrqr* / Tr > 2M /  • If M 2, then RHS bigger than 1, since Tr = dr + qr* • M = 1 • The stability condition cannot be satisfied if a source has more than one link • Sufficient in multilink case • Necessary and sufficient in single-link-homogeneous-source case

  26. Stability of TCP Vegas Single link with homogeneous source • A single link of capacity c, • Shared by N homogeneous source, • with round trip propagation delay d

  27. Stability of TCP Vegas Single link with homogeneous source • From corollary 3: qr* / Tr > 2 /  for all r • Tr / qr*< /2, since Tr = d + qr* • d / qr* < (/2 – 1) => d < (/2 – 1) qr* • Since qr* =  /xr* => ( N)/c • cd < (/2 – 1)  N • Conclusion: bandwidth delay product should be small

  28. Stabilized Vegas • xr (t) = (2/Tr2(t))tan-1(1 –xr(t)qr(t) / rdr ) • xr (t) = (w/Tr2(t))tan-1r(t)(1 –xr(t)qr(t)/rdr -r(t) qr(t)) • 1 –xr( t ) qr( t ) /r dr • 1 –xr( t ) qr( t ) /r dr -r( t ) qr( t ) • r( t ) = (1 / ) (Tr( t ) / qr( t ) ) • r( t ) = (  / w ) (xr( t ) Tr( t ) )

  29. Stabilized Vegas • The gain r( t ) serves as a normalization to qr( t ) • Additional differential term r( t ) qr( t ) anticipates the future of qr( t ) • Without: xr( t ) will increase if xr(t)qr(t)< rdr • Even xr(t)qr(t)/rdr is small, xr( t ) may decrease if prices are rapidly growing

  30.  2 + 2a2 cd < ( - 1 ) N   2 + a2 Stabilized Vegas • Stability condition for stabilized Vegas where  = tan-1( (2)/(1-) ) • Stabilized Vegas can choose a small ( a>0, (0,1) ) such that RHS can be larger for better stability of the original Vegas cd < (/2 – 1)  N

  31. Simulation Results One-on-One (300KB and 1MB) Transfers c = 200KB/s 50ms delay

  32. Simulation Results •  = 20 • N = 100 flows • Fixed packet size of 1KB • FIFO /w Droptail, queue capacity = 20000 • ( a ,  ) = ( 0.5 , 0.015 )

  33. Simulation Resultsc = 100 pkts/s and d = 10ms

  34. Simulation Resultsc = 1000 pkts/s and d = 10ms

  35. Simulation Resultsc = 100 pkts/s and d = 100ms

  36. Simulation Results

  37. Experimental Results • FAST TCP was first demonstrated publicly in during the SuperComputing Conference (SC2002) in Baltimore, MD, in November 16–22 2002 • Caltech-SLAC research team • CERN • DataTAG • StarLight • TeraGrid • Cisco • Level(3).

  38. Experimental Results

  39. Experimental ResultsThroughput and utilization • SC2002 FAST experimental result • Current TCP implementation in Linux v2.4.18 on Jan 27-28, 2003

  40. FAST TCPConclusion • Problem of current TCP • Equilibrium • Dynamic problem • FAST TCP • Equation-based control with queuing delay • TCP Vegas • Stabilized Vegas

  41. Q&A

  42. Thank you

More Related