Go to primary content
Oracle® Communications EAGLE SIGTRAN User's Guide
Release 46.6
E97352 Revision 1
Go To Table Of Contents
Contents

Previous
Previous
Next
Next

SCTP Timers

Oracle provides two retransmission modes: RFC and Linear. The SCTP retransmission control feature allows the tailoring of retransmissions to detect a network fault in a timely fashion through these configuration parameters:

RFC Timer Setting

With an exponential timer setting, the RTO value is doubled for each retransmit attempt. When transmitting a packet, the RTO has to expire before attempting to retransmit. With the second attempt, the last RTO value is doubled (RTO * 2) before retransmitting; with the third attempt, the last RTO value is doubled again (RTO * 4); and so on. This method significantly increases the time to determine that a link is lost.

For example, if data is being transmitted for five retransmits, the time to determine a lost link is:

RTO.min * Path.Max.Retransmits (or 1 + 2 + 4 + 8 + 16 + 32) = 63 sec

Table 5-19 shows RFC timers and their RFC and Oracle-recommended default values.

Table 5-19 SCTP Configuration Data Descriptions for Oracle EAGLE

RFC Name Description RFC Recommended Default Value Oracle Default Value Oracle Configurable? Oracle Ranges
RTO.initial

Initial

RTO Value

3 seconds 120 ms

Yes

Assoc RMIN parameter

1-1000 ms
RTO.max Upper limit of RTO 60 seconds 800 ms

Yes

Assoc RMAX parameter

1-1000 ms
RTO.min Lower limit of RTO 1 second 120 ms

Yes

Assoc RMIN parameter

1-1000 ms
Max.Init. Retransmits Maximum Initial Retransmit Attempts 8 attempts 10 attempts

Yes

Assoc RTIMES parameter.Not configurable independently of Assoc.max. retrans

1-12
Association. max. retrans Maximum Association Data Retransmit Attempts 10 attempts 10 attempts

Yes

Assoc RTIMES parameter

1-12 ms
Path.max. retrans Maximum Data Retransmit attempts per Destination (used for multi-homing only) 5 attempts 5 attempts Indirectly ½ of the assoc RTIMES parameter 1-6 ms
Acknowledge- ment timer SACK Transmit User Configurable not to exceed 500 ms ½ RTO or 200 ms, whichever is less

Indirectly

RTO is bound by the assoc RMIN and RMAX parameters

5-200 ms
T3-rtx Timer Data Retransmit

RTO

(see RTO.initial for initial value

)
RTO

(see RTO.initial for initial value)

Yes

RTO is bounded by the assoc RMIN and RMAX parameters

10-1000 ms
T1-init Timer Init retransmit timer

Initially 3 seconds

RTO thereafter

Initially 1 second, RTO thereafter

No for initial value

Indirectly thereafter via RMIN/RMAX bounding of RTO

10-1000 ms
HB.Interval Heart Beat Interval 30 seconds

RTO+500 ms

Yes

RTO + Assoc HBTIMER parameter

RTO+500 ms to

RTO+10000 ms

Shutdown timer Shutdown timer t2 RTO RTO

Indirectly

RTO is bound by the assoc RMIN and RMAX parameters

10-1000 ms
Cookie Timer Cookie-t1 – Cookie Echo retransmit timer

Initially 3 seconds

RTO thereafter

Initially 1 second

RTO thereafter

No for initial value

Indirectly thereafter via RMIN/RMAX bounding of RTO

10-1000 ms
Cookie life Cookie Life 60 seconds 5 seconds No 5 seconds

Note:

SCTP Heart Beats (HBs) are sent after every HB interval on an idle path of the SCTP association. If more than one idle path exists, then HBs are sent alternately on each idle path, after each HB interval.

LIN Timer Setting

Oracle has implemented a more aggressive timer method called Linear (LIN), in which the RTO between attempts is constant. Oracle recommends this setting to detect a failure more quickly than the RFC method.

With the LIN timer setting, the time to declare the association down is at least

RMIN * RTIMES

For very high-throughput associations, RTIMES (and if possible, RMIN) should be lowered and CWMIN increased. CWMIN is the parameter that sets the minimum size of the congestion window, which determines the number of packets that can be sent without having received the corresponding ACK packets.

On the far end, the LIN mode can coexist with RFC mode, but in contrast to the Signaling Gateway, the far-end may experience congestion in the ASP-to-SGP direction because of network impairments.

Jitter Effects

Since the RTO is a moving average of network RTT samples, as the jitter range increases, bounding the lower limit of the RTO at or near the average will cause the amount of unnecessary retransmissions to increase, since for each transmission that takes longer than the current RTO to acknowledge a retransmission will occur, wasting bandwidth..

If the lower limit of the RTO is bounded to the upper end of the jitter range to minimize retransmits, then connection failure detection time is similarly increased.

So, minimizing jitter in the network translates into a small range for network RTT, and the RTO can be bounded to minimize retransmissions while being able to detect a loss of connection in a timely fashion.