Oversubscribing Interface Bandwidth
The term oversubscribing interface bandwidth means configuring shaping rates (peak information rates [PIRs]) so that their sum exceeds the interface bandwidth.
On Channelized IQ PICs, Gigabit Ethernet IQ PICs, and FRF.15 and FRF.16 link services IQ (LSQ) interfaces on Services PICs, Multiservices PICs, and Multiservices DPCs, you can oversubscribe interface bandwidth. This means that the logical interfaces (and DLCIs within an FRF.15 or FRF.16 bundle) can be oversubscribed when there is leftover bandwidth. In the case of FRF.16 bundle interfaces, the physical interface can be oversubscribed. The oversubscription is capped to the configured PIR. Any unused bandwidth is distributed equally among oversubscribed logical interfaces or data-link connection identifiers (DLCIs), or physical interfaces.
For networks that are not likely to experience congestion, oversubscribing interface bandwidth improves network utilization, thereby allowing more customers to be provisioned on a single interface. If the actual data traffic does not exceed the interface bandwidth, oversubscription allows you to sell more bandwidth than the interface can support.
We recommend avoiding oversubscription in networks that are likely to experience congestion. Be cautious not to oversubscribe a service by too much, because this can cause degradation in the performance of the routing platform during congestion. When you configure oversubscription, starvation of some output queues can occur if the actual data traffic exceeds the physical interface bandwidth. You can prevent degradation by using statistical multiplexing to ensure that the actual data traffic does not exceed the interface bandwidth.
You cannot oversubscribe interface bandwidth when you configure traffic shaping using the method described in Applying Scheduler Maps and Shaping Rate to DLCIs and VLANs.
When configuring oversubscription for FRF.16 bundle interfaces, you can assign traffic control profiles that apply on a physical interface basis. When you apply traffic control profiles to FRF.16 bundles at the logical interface level, member link interface bandwidth is underutilized when there is a small proportion of traffic or no traffic at all on an individual DLCI. Support for traffic control features on the FRF.16 bundle physical interface level addresses this limitation.
To configure oversubscription of the interface, perform the following steps:
Include the
shaping-ratestatement at the[edit class-of-service traffic-control-profiles profile-name]hierarchy level:[edit class-of-service traffic-control-profiles profile-name] shaping-rate (percent percentage | rate);
Note:When configuring oversubscription for FRF.16 bundle interfaces on a physical interface basis, you must specify
shaping-rateas a percentage.On LSQ interfaces, you can configure the shaping rate as a percentage from 1 through 100.
On IQ and IQ2 interfaces, you can configure the shaping rate as an absolute rate from 1000 through 6,400,000,000,000 bps.
For all MX Series router and EX Series switch interfaces, the shaping rate can be from 65,535 through 6,400,000,000,000 bps.
Alternatively, you can configure a shaping rate for a logical interface and oversubscribe the physical interface by including the
shaping-ratestatement at the[edit class-of-service interfaces interface-name unit logical-unit-number]hierarchy level. However, with this configuration approach, you cannot independently control the delay-buffer rate, as described in Step 2.Note:For channelized and Gigabit Ethernet IQ interfaces, the
shaping-rateandguaranteed-ratestatements are mutually exclusive. You cannot configure some logical interfaces to use a shaping rate and others to use a guaranteed rate. This means there are no service guarantees when you configure a PIR. For these interfaces, you can configure either a PIR or a committed information rate (CIR), but not both.This restriction does not apply to Gigabit Ethernet IQ2 PICs or LSQ interfaces on Multiservices and Services PICs. For LSQ and Gigabit Ethernet IQ2 interfaces, you can configure both a PIR and a CIR on an interface. For more information about CIRs, see Providing a Guaranteed Minimum Rate.
For more information about Gigabit Ethernet IQ2 PICs, see CoS on Enhanced IQ2 PICs Overview.
Optionally, you can base the delay-buffer calculation on a delay-buffer rate. To do this, include the
delay-buffer-ratestatement at the[edit class-of-service traffic-control-profiles profile-name]hierarchy level:Note:When configuring oversubscription for FRF.16 bundle interfaces on a physical interface basis, you must specify
delay-buffer-rateas a percentage.[edit class-of-service traffic-control-profiles profile-name] delay-buffer-rate (percent percentage | rate);
The delay-buffer rate overrides the shaping rate as the basis for the delay-buffer calculation. In other words, the shaping rate or scaled shaping rate is used for delay-buffer calculations only when the delay-buffer rate is not configured.
For LSQ interfaces, if you do not configure a delay-buffer rate, the guaranteed rate (CIR) is used to assign buffers. If you do not configure a guaranteed rate, the shaping rate (PIR) is used in the undersubscribed case, and the scaled shaping rate is used in the oversubscribed case.
On LSQ interfaces, you can configure the delay-buffer rate as a percentage from 1 through 100.
On IQ and IQ2 interfaces, you can configure the delay-buffer rate as an absolute rate from 1000 through 6,400,000,000,000 bps.
The actual delay buffer is based on the calculations described in Managing Congestion on the Egress Interface by Configuring the Scheduler Buffer Size. For an example showing how the delay-buffer rates are applied, see Examples: Oversubscribing Interface Bandwidth.
Configuring large buffers on relatively slow-speed links can cause packet aging. To help prevent this problem, the software requires that the sum of the delay-buffer rates be less than or equal to the port speed.
This restriction does not eliminate the possibility of packet aging, so you should be cautious when using the
delay-buffer-ratestatement. Though some amount of extra buffering might be desirable for burst absorption, delay-buffer rates should not far exceed the service rate of the logical interface.If you configure delay-buffer rates so that the sum exceeds the port speed, the configured delay-buffer rate is not implemented for the last logical interface that you configure. Instead, that logical interface receives a delay-buffer rate of zero, and a warning message is displayed in the CLI. If bandwidth becomes available (because another logical interface is deleted or deactivated, or the port speed is increased), the configured delay-buffer-rate is reevaluated and implemented if possible.
If you do not configure a delay-buffer rate or a guaranteed rate, the logical interface receives a delay-buffer rate in proportion to the shaping rate and the remaining delay-buffer rate available. In other words, the delay-buffer rate for each logical interface with no configured delay-buffer rate is equal to:
(remaining delay-buffer rate * shaping rate) / (sum of shaping rates)
where the remaining delay-buffer rate is equal to:
(interface speed) - (sum of configured delay-buffer rates)
To assign a scheduler map to the logical interface, include the
scheduler-mapstatement at the[edit class-of-service traffic-control-profiles profile-name]hierarchy level:[edit class-of-service traffic-control-profiles profile-name] scheduler-map map-name;
For information about configuring schedulers and scheduler maps, see Configuring Schedulers and Configuring Scheduler Maps.
Optionally, you can enable large buffer sizes to be configured. To do this, include the
q-pic-large-bufferstatement at the[edit chassis fpc slot-number pic pic-number]hierarchy level:[edit chassis fpc slot-number pic pic-number] q-pic-large-buffer;
If you do not include this statement, the delay-buffer size is more restricted. We recommend restricted buffers for delay-sensitive traffic, such as voice traffic. For more information, see Managing Congestion on the Egress Interface by Configuring the Scheduler Buffer Size.
To enable scheduling on logical interfaces, include the
per-unit-schedulerstatement at the[edit interfaces interface-name]hierarchy level:[edit interfaces interface-name] per-unit-scheduler;
When you include this statement, the maximum number of VLANs supported is 768 on a single-port Gigabit Ethernet IQ PIC. On a dual-port Gigabit Ethernet IQ PIC, the maximum number is 384.
To enable scheduling for FRF.16 bundles physical interfaces, include the
no-per-unit-schedulerstatement at the[edit interfaces interface-name]hierarchy level:[edit interfaces interface-name] no-per-unit-scheduler;
To apply the traffic-scheduling profile , include the output-traffic-control-profile statement at the
[edit class-of-service interfaces interface-name unit logical-unit-number]hierarchy level:[edit class-of-service interfaces interface-name unit logical-unit-number] output-traffic-control-profile profile-name;
You cannot include the
output-traffic-control-profilestatement in the configuration if either thescheduler-maporshaping-ratestatement is included in the logical interface configuration.
Table 1 shows how the bandwidth and delay buffer are allocated in various configurations.
Configuration Scenario |
Delay Buffer Allocation |
|---|---|
You do not oversubscribe the interface. You do not configure a guaranteed rate. You do not configure a shaping rate. You do not configure a delay-buffer rate. |
Logical interface receives the remaining bandwidth and receives a delay buffer in proportion to the remaining bandwidth. |
You do not oversubscribe the interface. You configure
a shaping rate at the |
For backward compatibility, the shaped logical interface
receives a delay buffer based on the shaping rate. The multiplicative
factor depends on whether you include the Unshaped logical interfaces receive the remaining bandwidth and a delay buffer in proportion to the remaining bandwidth. |
You oversubscribe the interface. You do not configure a guaranteed rate. You do not configure a shaping rate. You do not configure a delay-buffer rate. |
Logical interface receives minimal bandwidth with no guarantees and receives a minimal delay buffer equal to four MTU-sized packets. |
You oversubscribe the interface. You configure a shaping rate. You do not configure a guaranteed rate. You do not configure a delay-buffer rate. |
Logical interface receives a delay buffer based on the scaled shaping rate: scaled shaping rate = (shaping-rate * [physical interface bandwidth]) / SUM (shaping-rates of all logical interfaces on the physical interface) The logical interface receives variable bandwidth, depending on how much oversubscription and statistical multiplexing is present. If the amount of oversubscription is low enough that statistical multiplexing does not make all logical interfaces active at the same time and the physical interface bandwidth is not exceeded, the logical interface receives bandwidth equal to the shaping rate. Otherwise, the logical interface receives a smaller amount of bandwidth. In either case, the logical interface bandwidth does not exceed the shaping rate. |
You oversubscribe the interface. You configure a shaping rate. You configure a delay-buffer rate. |
Logical interface receives a delay buffer based on the delay-buffer rate. For example, on IQ and IQ2 interfaces: delay-buffer-rate <= 10 Mbps: 400-millisecond (ms) delay buffer delay-buffer-rate <= 20 Mbps: 300-ms delay buffer delay-buffer-rate <= 30 Mbps: 200-ms delay buffer delay-buffer-rate <= 40 Mbps: 150-ms delay buffer delay-buffer-rate > 40 Mbps: 100-ms delay buffer On LSQ DLCIs, if delay-buffer-rate = 1 second On LSQ DLCIs, if delay-buffer-rate = 200 ms The multiplicative factor depends on whether you include
the The logical interface receives variable bandwidth, depending on how much oversubscription and statistical multiplexing is present. If the amount of oversubscription is low enough that statistical multiplexing does not make all logical interfaces active at the same time and the physical interface bandwidth is not exceeded, the logical interface receives bandwidth equal to the shaping rate. Otherwise, the logical interface receives a smaller amount of bandwidth. In either case, the logical interface bandwidth does not exceed the shaping rate. |
You oversubscribe the interface. You do not configure a shaping rate. You configure a guaranteed rate. You configure a delay-buffer rate. |
Logical interface receives a delay buffer based on the delay-buffer rate. |
You oversubscribe the interface. You do not configure a shaping rate. You do not configure a guaranteed rate. You configure a delay-buffer rate. |
This scenario is not allowed. If you configure a delay-buffer rate, the traffic-control profile must also include either a shaping rate or a guaranteed rate. |
You oversubscribe the interface. You configure a shaping rate. You configure a guaranteed rate. You do not configure a delay-buffer rate. |
Logical interface receives a delay buffer based on the guaranteed rate. This configuration is valid on LSQ interfaces and Gigabit Ethernet IQ2 interfaces only. On channelized interfaces, you cannot configure both a shaping rate (PIR) and a guaranteed rate (CIR). |
In Junos OS Release 13.3, IP packets with DLCI 0 or 1023 are identified as part of control traffic and routed to the high-priority queue. This oversubscribes the high-priority queue, which is reserved for frame relay control traffic. Oversubscribing the high-priority queue causes the frame relay Local Management Interface (LMI) packets to be dropped.
Verifying Configuration of Bandwidth Oversubscription
To verify your configuration, you can issue this following operational mode commands:
show class-of-service interfacesshow class-of-service traffic-control-profile profile-name
Examples: Oversubscribing Interface Bandwidth
This section provides two examples: oversubscription of a channelized interface and oversubscription of an LSQ interface.
Oversubscribing a Channelized Interface
Two logical interface units, 0 and 1,
are shaped to rates 2 Mbps and 3 Mbps, respectively. The
delay-buffer rates are 750 Kbps and 500 Kbps, respectively.
The actual delay buffers allocated to each logical interface
are 1 second of 750 Kbps and 2 seconds of 500 Kbps,
respectively. The 1-second and 2-second values are based on the following
calculations:
delay-buffer-rate < [16 x 64 Kbps]): 1 second of delay-buffer-rate delay-buffer-rate < [8 x 64 Kbps]): 2 seconds of delay-buffer-rate
For more information about these calculations, see Managing Congestion on the Egress Interface by Configuring the Scheduler Buffer Size.
chassis {
fpc 3 {
pic 0 {
q-pic-large-buffer;
}
}
}
interfaces {
t1-3/0/0 {
per-unit-scheduler;
}
}
class-of-service {
traffic-control-profiles {
tc-profile1 {
shaping-rate 2m;
delay-buffer-rate 750k; # 750 Kbps is less than 16 x 64 Kbps
scheduler-map sched-map1;
}
tc-profile2 {
shaping-rate 3m;
delay-buffer-rate 500k; # 500 Kbps is less than 8 x 64 Kbps
scheduler-map sched-map2;
}
}
interfaces {
t1-3/0/0 {
unit 0 {
output-traffic-control-profile tc-profile1;
}
unit 1 {
output-traffic-control-profile tc-profile2;
}
}
}
}
Oversubscribing an LSQ Interface with Scheduling Based on the Logical Interface
Apply a traffic-control profile to a logical interface representing a DLCI on an FRF.16 bundle:
interfaces {
lsq-1/3/0:0 {
per-unit-scheduler;
unit 0 {
dlci 100;
}
unit 1 {
dlci 200;
}
}
}
class-of-service {
traffic-control-profiles {
tc_0 {
shaping-rate percent 100;
guaranteed-rate percent 60;
delay-buffer-rate percent 80;
}
tc_1 {
shaping-rate percent 80;
guaranteed-rate percent 40;
}
}
interfaces {
lsq-1/3/0 {
unit 0 {
output-traffic-control-profile tc_0;
}
unit 1 {
output-traffic-control-profile tc_1;
}
}
}
}
Oversubscribing an LSQ Interface with Scheduling Based on the Physical Interface
Apply a traffic-control profile to the physical interface representing an FRF.16 bundle:
interfaces {
lsq-0/2/0:0 {
no-per-unit-scheduler;
encapsulation multilink-frame-relay-uni-nni;
unit 0 {
dlci 100;
family inet {
address 10.18.18.2/24;
}
}
}
class-of-service {
traffic-control-profiles {
rlsq_tc {
scheduler-map rlsq;
shaping-rate percent 60;
delay-buffer-rate percent 10;
}
}
interfaces {
lsq-0/2/0:0 {
output-traffic-control-profile rlsq_tc;
}
}
}
scheduler-maps {
rlsq {
forwarding-class best-effort scheduler rlsq_scheduler;
forwarding-class expedited-forwarding scheduler rlsq_scheduler1;
}
}
schedulers {
rlsq_scheduler {
transmit-rate percent 20;
priority low;
}
rlsq_scheduler1 {
transmit-rate percent 40;
priority high;
}
}
On an FRF.15 bundle, apply the following configuration:
class-of-service {
traffic-control-profiles {
rlsq {
scheduler-map sched_0;
shaping-rate percent 40;
delay-buffer-rate percent 50;
}
}
interfaces lsq-2/0/0 {
unit 0 {
output-traffic-control-profile rlsq;
}
}
}
interfaces lsq-2/0/0 {
per-unit-scheduler;
unit 0 {
encapsulation multilink-frame-relay-end-to-end;
family inet {
address 10.1.1.2/32;
}
}
}