All,
In troubleshooting an issue with output drops on one of our switches attached to a customer, one of our engineers used the information in the 'show interface gix/x' as a guide for the queue size. The section that i'm describing is this:
GigabitEthernet1/30 is up, line protocol is up (connected)
<output omitted>
Queueing strategy: fifo
Output queue: 0/40 (size/max)
The customer asked our engineer if he could increase that value of 40. I came across the ticket, and lent a hand and explained that the 4948 uses hardware queues, and we can see these here:
clt3-access-13#sh platform hardware interface g1/30 tx-queue
Switch Phyport Gi1/30 Tx-Queue Hardware State
Phyport TxQ High Shape Share Shape Share Curr
Priority Mant. Exp. Mant. Exp. NextTime NextTime Time
-------------------------------------------------------------------------------
Gi1/30 0 Disabled 0 0 45 16 0x11B1D00D4 0x11B1D00D4 0xAB97B6383
Gi1/30 1 - 0 0 45 16 0xAA3506F73 0xAA3507324 0xAB97BAAB8
Gi1/30 2 - 0 0 45 16 0x11B1F99F3 0x11B1F99F3 0xAB97BEF88
Gi1/30 3 - 0 0 45 16 0xAB97B7DCD 0xAB97B8B12 0xAB97C3327
Phyport TxQ Head Tail Full Not Num BaseAddr Size
Empty Packets
---------------------------------------------------------------
Gi1/30 0 0x0000 0x0000 False False 0 0x20800 2080
Gi1/30 1 0x03C5 0x03C5 False False 0 0x21020 2080
Gi1/30 2 0x0000 0x0000 False False 0 0x21840 2080
Gi1/30 3 0x0070 0x0070 False False 0 0x22060 2080
My question is two part really. First off, why does the output in the show interface have the output queue as 40 when it's obviously not being used(if I remember correctly, this had to do with router interfaces)?
My other question, is that the QoS documentation for the 4948 is here (apparently the 4500 docs are what is supposed to be used):
http://www.cisco.com/en/US/docs/switches/lan/catalyst4500/12.2/15.02SG/configuration/guide/qos.html#wp1463316
But I can find no refereence to the timeframe that these queues get serviced. This is important to me because when I look at the queue's for the interface:
clt3-access-13#sh qos interface g1/30
QoS is enabled globally
Port QoS is enabled
Administrative Port Trust State: 'untrusted'
Operational Port Trust State: 'untrusted'
Trust device: none
Default DSCP: 0 Default CoS: 0
Appliance trust: none
Tx-Queue Bandwidth ShapeRate Priority QueueSize
(bps) (bps) (packets)
1 250000000 disabled N/A 2080
2 250000000 disabled N/A 2080
3 250000000 disabled normal 2080
4 250000000 disabled N/A 2080
I'm thinking that that packet count can easily get overrun. If we do the math real quick, and say we use a 1500 byte packet, and use the formula (1500*2080)8, we end up getting 24960000, which is much lower than 250000000. So I'm a bit confused on the correlation between the queue size regarding packets, and the bandwidth. This is why I was asking about the timeframe that the above rates are referencing, because it would seem to me that you would have to have quite a bit more than 2080 (standard size packets) to accumulate 250 Mbps of traffic, not to mention that a queue limit of 2080 packets may seem like alot, but we have customer links that are showing pps well in excess of 4000 with no output drops, which again leads me to believe that the packet count shown above is for a timeframe that is shorter than a second.
Does this make sense?
Any help is appreciated,
James
