Re: hold-queue parameter<Output queue 0/1024, 745838161 drops>

lawwm · ‎08-27-2008

Hi All,

Recently, I found out that it's something wrong with the parameter about hold-queue.

Below is the configuration. When I issued show interface 3/0/0, I can see the output queue is increasing a lot with 5 seconds.

I've compared other site of the router. When the hold-queue in and out are the same there is no out queue-drops.

Any idea?

Current configuration : 232 bytes

!

interface GigabitEthernet3/0/0

description *** Physical Interface to the Celcom SDH network ***

mtu 1600

no ip address

no ip directed-broadcast

load-interval 30

negotiation auto

hold-queue 2048 in

hold-queue 1024 out

end

RouterA#show interfaces gigabitEthernet 3/0/0

GigabitEthernet3/0/0 is up, line protocol is up

Hardware is GigabitEthernet, address is 0017.0e2b.4d00 (bia 0017.0e2b.4d00)

Description: *** Physical Interface to the Celcom SDH network ***

MTU 1600 bytes, BW 1000000 Kbit, DLY 10 usec, rely 255/255, load 126/255

Encapsulation ARPA, loopback not set

Keepalive set (10 sec)

Full Duplex, 1000Mbps, link type is autonegotiation, media type is LX

output flow-control is on, input flow-control is on

ARP type: ARPA, ARP Timeout 04:00:00

Last input 00:00:00, output 00:00:00, output hang never

Last clearing of "show interface" counters 4w4d

Queueing strategy: fifo

Output queue 0/1024, 745838161 drops; input queue 0/2048, 0 drops

RouterA#sh int gigabitEthernet 3/0/0

GigabitEthernet3/0/0 is up, line protocol is up

Hardware is GigabitEthernet, address is 0017.0e2b.4d00 (bia 0017.0e2b.4d00)

Description: *** Physical Interface to the Celcom SDH network ***

MTU 1600 bytes, BW 1000000 Kbit, DLY 10 usec, rely 255/255, load 130/255

Encapsulation ARPA, loopback not set

Keepalive set (10 sec)

Full Duplex, 1000Mbps, link type is autonegotiation, media type is LX

output flow-control is on, input flow-control is on

ARP type: ARPA, ARP Timeout 04:00:00

Last input 00:00:00, output 00:00:00, output hang never

Last clearing of "show interface" counters 4w4d

Queueing strategy: fifo

Output queue 0/1024, 748766847 drops; input queue 0/2048, 0 drops

Here is the config of Router B with the same parameters

Current configuration : 266 bytes

!

interface GigabitEthernet3/1/3

description *** Physical Interface to the Celcom SDH network ***

mtu 1600

no ip address

no ip redirects

no ip directed-broadcast

no ip proxy-arp

load-interval 30

negotiation auto

hold-queue 1024 in

hold-queue 1024 out

RouterB#show int gigabitEthernet 3/1/3

GigabitEthernet3/1/3 is up, line protocol is up

Hardware is GigabitEthernet, address is 0019.e72e.7613 (bia 0019.e72e.7613)

Description: *** Physical Interface to the Celcom SDH network ***

MTU 1600 bytes, BW 1000000 Kbit, DLY 10 usec, rely 255/255, load 11/255

Encapsulation ARPA, loopback not set

Keepalive set (10 sec)

Full Duplex, 1000Mbps, link type is autonegotiation, media type is LX

output flow-control is on, input flow-control is on

ARP type: ARPA, ARP Timeout 04:00:00

Last input 00:00:00, output 00:00:00, output hang never

Last clearing of "show interface" counters 15w5d

Queueing strategy: fifo

Output queue 0/1024, 0 drops; input queue 0/1024, 0 drops<<<no drop

Available Bandwidth 911496 kilobits/sec

Joseph W. Doherty · ‎08-27-2008

At least with the information you've posted, the important difference is likely the output load on RouterA vs. RouterB, load 130/255 vs. load 11/255.

Input queues and output queues don't need to balance, and input queues are often not congested since the input link sets a cap on the inbound rate.

A 2,048 deep output queue is rather large. It's counter to what you might expect, but a too deep queue can actually contribute to drops and poor performance.

What type of devices and IOS are Routers A and B? What's the nature of the traffic?

lawwm · ‎08-27-2008

Cisco Internetwork Operating System Software

IOS (tm) GS Software (C12KPRP-K4P-M), Version 12.0(32)S3, RELEASE SOFTWARE (fc1)

Technical Support: http://www.cisco.com/techsupport

Compiled Fri 09-Jun-06 22:13 by leccese

Image text-base: 0x00010000, data-base: 0x05159000

ROM: System Bootstrap, Version 12.0(20040128:214555) [assafb-PRP1P_20040101 1.8dev(2.83)] DEVELOPMENT SOFTWARE

BOOTLDR: GS Software (C12KPRP-K4P-M), Version 12.0(32)S3, RELEASE SOFTWARE (fc1)

uptime is 2 years, 3 days, 11 hours, 40 minutes

Uptime for this control processor is 2 years, 3 days, 11 hours, 40 minutes

System returned to ROM by Stateful Switchover at 12:52:19 MY Mon Aug 14 2006

System restarted at 19:37:36 MY Fri Aug 25 2006

System image file is "disk0:c12kprp-k4p-mz.120-32.S3.bin"

cisco 12416/PRP (MPC7457) processor (revision 0x00) with 1048576K bytes of memory.

MPC7457 CPU at 1263Mhz, Rev 1.1, 512KB L2, 2048KB L3 Cache

Last reset from power-on

Channelized E1, Version 1.0.

Traffic will Data and Voice.

I've changed to 1024. But no luck. I also tried 2048 for in and out. Still the same. the queues counter still increasing.

You are right, the Load both router do not equal. Do you have any idea?

Appreciate your input

Joseph W. Doherty · ‎08-28-2008

I've not worked with a 12,000 series router. Looking over the documentation, you might be able to implement QoS features beyond using a simple FIFO queue. You could experiment with WRED and/or FQ.

Assuming TCP is the bulk of your traffic, if the output link is oversubscribed, it's normal to see drops, especially as flows attempt to use all bandwidth. However, what you want to obtain is the minimal amount of drops at peak usage. A large FIFO queue might allow multiple flows to all ramp up and then has them all fall into slow start with (global) synchronous tail drops. Both WRED and/or FQ might avoid the synchronous tail drops, both improving overall throughput and, surprisingly, often reducing overall drops.

Unsure whether the following can be configured on your interface, but try this:

policy-map yourname

class class-default

fair-queue

random-detect

(RouterA)

interface gigabitEthernet3/0/0

service-policy output yourname