WRED Configuration

WRED Configuration

Because WRED manages drops based on chain depth, WRED charge be configured alongside a

particular queue. However, best queuing mechanisms do not abutment WRED; as a result, WRED

can be configured alone in the afterward locations:

■ On a concrete interface (with FIFO queuing)

■ For a non-LLQ chic central a CBWFQ activity map

■ For an ATM VC

To use WRED anon on a concrete interface, IOS absolutely disables all added queuing mechanisms

and creates a distinct FIFO queue. WRED again manages the chain with attention to drops. For

CBWFQ, WRED is configured in a chic central a activity map, in the aforementioned area as the

bandwidth and antecedence commands discussed beforehand in this chapter.

The random-detect command enables WRED, either beneath a concrete interface or beneath a class

in a activity map. This command enables WRED to use IPP, and not DSCP. The random-detect

dscp-based command both enables WRED and tells it to use DSCP for free the traffic

profile for a packet.

To change WRED agreement from the absence WRED contour for a accurate IPP or DSCP, use

the afterward commands, in the aforementioned area as the added random-detect command:

random-detect antecedence antecedence min-threshold max-threshold [mark-probdenominator]

random-detect dscp dscpvalue min-threshold max-threshold [mark-probabilitydenominator]

Finally, abacus of the rolling boilerplate chain abyss can be afflicted through configuring a

parameter alleged the exponential weighting constant. A low exponential weighting connected means

that the old boilerplate is a baby allotment of the calculation, consistent in a added bound alteration average.

The ambience can be afflicted with the afterward command, although alteration it is not

recommended:

random-detect exponential-weighting-constant exponent

Note that earlier, Archetype 13-4 showed basal WRED agreement central some classes of a

CBWFQ configuration.

Modified Arrears Round-Robin

MDRR is a queuing affection implemented alone in the Cisco 12000 alternation router family. Because

the 12000 alternation does not abutment CBWFQ and LLQ, MDRR serves in abode of these features. Its

main claims to acclaim are bigger candor than bequest queuing methods such as antecedence queuing and

custom queuing, and that it supports a antecedence chain (like LLQ). For the CCIE Routing and

Switching condoning exam, you charge to accept how MDRR works at the conceptual level, but

you don’t charge to apperceive how to configure it.

MDRR allows classifying cartage into seven round-robin queues (0–6), with one added priority

queue. When no packets are placed into the antecedence queue, MDRR commonly casework its queues

in a round-robin approach, cycling through anniversary chain already per cycle. With packets in the priority

queue, MDRR has two options for how to accommodate the antecedence chain in the chain service

algorithm:

■ Austere antecedence mode

■ Alternating mode

Strict antecedence approach serves the antecedence chain whenever cartage is present in that queue. The benefit

is, of course, that this cartage gets the aboriginal account behindhand of what is activity on in the added queues.

The downside is that it may advance to chain starvation in added queues if there is consistently cartage in the

priority queue. In this mode, the antecedence chain additionally can get added than the configured bandwidth

percentage, because this chain is served added than already per cycle.

By contrast, alternating approach serves the antecedence chain in amid confined anniversary of the added queues.

Let’s say that bristles queues are configured: 0, 1, 2, 3, and the antecedence chain (P). Assuming that there

is consistently cartage in anniversary queue, actuality is how it would be processed: 0, P, 1, P, 2, P, 3, P, and so on.

The aftereffect is that chain starvation in non-priority queues does not occur, because anniversary chain is

being served. The check of this approach is that it can account jitter and added cessation for the

traffic in the antecedence queue, compared to austere antecedence mode.

456 Affiliate 13: Congestion Management and Avoidance

Two agreement in MDRR, different to this queuing method, advice to differentiate MDRR from other

queuing tools:

■ Breakthrough amount (QV)

■ Deficit

MDRR supports two types of scheduling, one of which uses the aforementioned accepted algorithm as the

legacy CQ affection in Cisco IOS routers (other than the 12000 series). MDRR removes packets

from a chain until the breakthrough amount (QV) for that chain has been removed. The QV quantifies

a cardinal of bytes and is acclimated abundant like the byte calculation is acclimated by the CQ scheduler. MDRR

repeats the activity for every queue, in adjustment from 0 through 7, and again repeats this round-robin

process. The end aftereffect is that anniversary chain gets some allotment bandwidth of the link.

MDRR deals with the CQ scheduler’s botheration by alleviative any “extra” bytes beatific during a cycle

as a “deficit.” If too abounding bytes were taken from a queue, abutting time about through the queues,

the cardinal of added bytes beatific by MDRR is subtracted from the QV. In effect, if added than the QV

is beatific from a chain in one pass, that abounding beneath bytes are taken in the abutting pass. As a result,

averaged over abounding passes through the cycle, the MDRR scheduler provides an exact bandwidth

reservation.

Figure 13-7 shows an archetype of how MDRR works. In this case, MDRR is application alone two

queues, with QVs of 1500 and 3000, respectively, and with all packets at 1000 bytes in length.

Figure 13-7 MDRR: Making Up Deficits

Queue1

P6 P5 P4 P3 P2 P1

Queue2

P12 P11 P10 P9 P8 P7

Beginning Q1 Arrears Calculation = 1500

Ending Q1 Arrears Calculation = -500

Beginning Q1 Arrears Calculation = 3000

Ending Q1 Arrears Calculation = 0

Note: All Packets are 1000 bytes long!

1st MDRR Canyon Through the Queues

Queue1

Queue2

P12 P11 P10

Beginning Q1 Arrears Calculation = -500 + 1500 = 1000

Ending Q1 Arrears Calculation = 0

Beginning Q1 Arrears Calculation = 0 + 3000 = 3000

Ending Q1 Arrears Calculation = 0

P6 P5 P4 P3

2nd MDRR Canyon Through the Queues

LAN Switch Congestion Management and Avoidance 457

Some altercation of how to adapt Amount 13-7 may advice you abstract what is activity on. The figure

shows the activity during the aboriginal round-robin canyon in the top bisected of the figure, and the activity during

the added canyon in the lower bisected of the figure. The archetype begins with six packets (labeled P1

through P6) in chain 1, and six packets (labeled P7 through P12) in chain 2. Anniversary arrowed line

to the appropriate abandon of the queues, pointing to the right, represents the best by MDRR to accelerate a

single packet.

When a chain aboriginal fills, the queue’s arrears adverse (DC) is set to the QV for that queue, which is

1500 for chain 1 and 3000 for chain 2. In Amount 13-7, MDRR begins by demography one packet from

queue 1, decrementing the DC to 500, and chief that the DC is still greater than 0. Therefore,

MDRR takes a added packet from chain 1, decrementing the DC to –500. MDRR again moves on

to chain 2, demography three packets, afterwards which the arrears adverse (DC) for chain 2 has decremented

to 0.

That concludes the aboriginal round-robin canyon through the queues. MDRR has taken 2000 bytes from

queue 1 and 3000 bytes from chain 2, giving the queues 40 percent and 60 percent of link

bandwidth, respectively.

In the added round-robin pass, apparent in the lower bisected of Amount 13-7, the activity begins by

MDRR abacus the QV for anniversary chain to the DC for anniversary queue. Chain 1’s DC becomes 1500 +

(–500), or 1000, to activate the added pass. During this pass, MDRR takes P3 from chain 1,

decrements DC to 0, and again moves on to chain 2. Afterwards demography three added packets from

queue 3, decrementing chain 2’s DC to 0, MDRR completes the added pass. Over these two

round-robin passes, MDRR has taken 3000 bytes from chain 1 and 6000 bytes from chain 2—

which is the aforementioned arrangement as the arrangement amid the QVs. In added words, MDRR has absolutely achieved

the configured bandwidth arrangement amid the two queues.

The arrears affection of MDRR provides a agency that, over time, gives anniversary chain a guaranteed

bandwidth based on the afterward formula:

For added examples of the operation of the MDRR arrears feature, accredit to http://

www.cisco.com/warp/public/63/toc_18841.html. Alternatively, you can go to Cisco.com and

search for “Understanding and Configuring MDRR and WRED on the Cisco 12000 Alternation Internet

Router.”

LAN Switch Congestion Management and Avoidance

The final area of this affiliate looks at the admission and departure queuing, SRR, WTD, and WRED

features on Cisco 3550 and 3560 switches.

QV for Chain X

Sum of All QVs