response time

[josephschung]

Sir,

I have a network of three tiers.

1) Core : 2 x 6504E (VSS)

2) Dist : 2 x 3750 (Stack)

3) Access : 3 x 3560 (config as layer 2)

4) Layer 3 MCE is built between Core and Dist.

5) VLANs are created on Dist.

6) Etherchannel between Dist and Access.

7) eigrp is used for layer 3 ports.

I connect a XP machine to the access switch and ping the Core, Dist and Access respectively and got the following result.

1) Core response is <1ms

2) Dist response is around 2-4ms

3) Access response is around 2-4ms

My question is:

To reach the Core, PC have to go through Access and then Dist switch. But why does the response time from the core is shorter?

Any idea?

Thanks.

[Mahesh Gohil]

Hi Joseph,

When you ping destination the response time it not only depends on distance or no. of equipment. There are many other factors

like

> Serialization delay: It is time takes to serialize bits on physical media. (faster the link lower the response time)

> Propogation delay: depends upon distance between source and destination

> Queuing delay: More packet in a queue more is the response time

> Forwarding delay: It is time equipment take to process a packet. (depends upon architecture of equipment)

> Shaping delay (Qos stuff)

> Network delay: Understood

I hope you can conclude now what causes your core switch to respond faster than access/distribution

Regards

Mahesh

[josephschung]

Hi Mahesh,

Thanks for the information.

Though consider all your indicated factors, I cannot find a resonable explanation. The concept is simple. PC have to go through Access and Dist before reach Core. I cannot understand why the response from core is faster than Dist and Access.

I attached the network diagram for your information.

Thanks a lot.

[Adam Casella]

Hey Joe,

Along with what Manish has already covered. The difference in times would more than likely be due to the time it takes the CPU of the device you are sending the ICMP packet to generate a response.  In other words the diff would be a hardware switched packet vs a software switched packet.

Cisco switches forward traffic in hardware whenever possible and will forward traffic passing through the switch much quicker than traffic destined to it's self.  This is due to the fact that the CPU needs more time to process the packet and then generate a response in software.

On most cisco devices ICMP is given lower priortiy compared to other traffic (STP, OSPF, HSRP, etc), so if the CPU is busy processing something else it will take longer to repsond to ICMP traffic.

Overall, I would say in this situation the core switch probably has a faster CPU than your Dist/Access devices and thus responds faster.

Thanks,

Adam

[josephschung]

Hi Adam,

I have a quick check on the CPU usage and listed below.

Core 6504E VSS : 2%

Dist 3750X Stack : 19%

Access 3560X : 23%

Do you still think the delay difference is caused by CPU? Now, there is no traffic and we are using etherchannel and eigrp only.

Thanks.

[Adam Casella · Accepted solution]

Hey Joe,

Yup, still looks like CPU to me.  As you can see your Access and Dist CPU's are working harder than your 6500.   So it is taking longer for the 3560 and 3750 to generate a response because they are busier.

Thanks,

Adam