Persist and Keepalive Timers - PowerPoint PPT Presentation

1 / 12
About This Presentation
Title:

Persist and Keepalive Timers

Description:

TCP could stall. Solution: Send periodic probes to elicit ACKs from receiver ... Persist timer probes are sent at intervals based on the RTO ... – PowerPoint PPT presentation

Number of Views:484
Avg rating:3.0/5.0
Slides: 13
Provided by: cseOhi
Category:

less

Transcript and Presenter's Notes

Title: Persist and Keepalive Timers


1
Persist and Keepalive Timers
2
Window Advertisement
  • What happens when a window advertisement with
    recvwin gt 0 is lost?
  • ACKs are not retransmitted
  • TCP could stall
  • Solution
  • Send periodic probes to elicit ACKs from receiver

3
Persist Timer in Action
  • Why 1 byte is sent?
  • Persist timer probes are sent at intervals based
    on the RTO
  • But it is at least 5 sec and at most 60 sec
  • TCP never gives up sending persist timer probes
    unless
  • Window opens up
  • Application terminates

4
Silly Window Syndrome
  • Window based flow control schemes may suffer from
    silly window syndrome
  • Receiver side
  • Receiver advertises small window and sender sends
    small chunk of data
  • Sender side
  • Sender sends small amount of data instead of
    waiting for more data

5
Receiver Side Avoidance of SWS
  • Do not send a new window update unless it is
    larger than current window by
  • 1 MSS
  • Or, half of receivers buffer space, whichever is
    smaller (rare)

6
Receiver AvoidanceSWS
  • Notice persist timer in action
  • Although bytes are being received, win is still
    0. Why?

7
?
  • Why 1022 bytes are not advertised?
  • Why 509 bytes are advertised?

8
Sender Side Avoidance of SWS
  • Do not transmit data unless
  • A full-size segment can be sent
  • Or, we can send at least half of the window that
    the other side has ever advertised (older,
    primitive hosts)
  • Or, we can send everything we have and either we
    are not expecting an ACK or Nagles algorithm is
    disabled

9
Keepalive Timer
  • Send packets every 2 hours to check if client is
    up
  • If no activity from client, four possibilities
  • Client is up
  • Client is down
  • Client has crashed and rebooted
  • Client unreachable (similar to scenario 2)

10
Host has crashed
  • If no response, send 10 probe packets 75 seconds
    apart, before declaring the connection dead

11
Host has crashed and rebooted
  • TCP reset is useful

12
Host unreachable
  • ICMP host unreachable messages are ignored by TCP
  • TCP returns error of No route to host to the
    application
Write a Comment
User Comments (0)
About PowerShow.com