help needed: switching traffic from 10 gig to 1 gig; speed issues?

Unanswered Question
Dec 16th, 2009
User Badges:

Hi Guys,

Can someone please help in the below scenario?


We are looking to pass some traffic from one site (C) to other (A) transiting through another site (B). The three site are going to be connected in the following way:
  Site C ==>> Site B (cisco 3560 - SFP + X2 module) == >> Site A (cisco 6509 - XENPAK module)
(4 x 1GE)                 (4 x 1GE + 1 x 10GE)                                              (1 x 10GE)

Cisco 3560 at Site B will have four 1 GE links from Site C plugged to it as well as one 10 GE link to Site A. About 2-3 Gbps of traffic will be coming from Site C and switched over to Site A by this switch at site B and there is no routing involved. I will just configure access vlans for each 1 GE link and trunk them across through the 10 GE link.

Now, the traffic on 10 GE link will be traveling faster as compared to 1 GE links and my concern here is some traffic will be dropped as a result. I believe traffic will be buffered when traveling from 10 GE to 1 GE but any excess traffic will just be dropped as it can't travel at the same rate.


1. Can someone please advise if this is going to be an issue with the amount of traffic we are talking about?  If yes, then what measures can we take to avoid this situation?
2. Can we possibly do some shaping/rate-limiting and slow down the traffic coming from 10 GE and how?

Any help will be much appreciated.


  • 1
  • 2
  • 3
  • 4
  • 5
Overall Rating: 0 (0 ratings)
Giuseppe Larosa Thu, 12/17/2009 - 08:21
User Badges:
  • Super Silver, 17500 points or more
  • Hall of Fame,

    Founding Member

Hello Ankit,

my first note is that I would use the 4 Gbps links as an etherchannel bundle for providing redundancy.

you need to prepare single links to be configured exactly in the same way:

int gi 0/x


switchport trunk protocol dot1q

switchport mode trunk

switchport trunk allowed vlan x,y,z

then you add commands to create the bundle

int gi0/x

channel-group n mode active

I recommend LACP for this instead of mode on that is unconditional bundle.

about your concerns:  if most of traffic is TCP it will be autoadapt to the limit of 4 Gbps.

shaping is not an option even on WS-6704-10GE you could police but it is not recommended

As you can see here:

Hope to help


ankitagrawal Thu, 12/17/2009 - 15:20
User Badges:

Thanks Giuseppe....

We can't do etherchannel because we don't have any control at Site C where the 4 x 1 Gbps links originate.

At Site B, we will have four VLANs per say, one each for 1 GE link which will then be trunked across to Site A through 10 GE. On three links we will have roughly around 500 Mbps of traffic in one direction and fourth link will have around 500 Mbps of traffic in other direction, so in total about 1.5 Gbps IN and 0.5 Gbps OUT through the 10 GE link.

Most of our traffic is UDP and we want no packet loss, like nothing at is critical traffic and any packet loss will impact us very badly. So, I am not sure how UDP will behave when it is coming from 10 GE and then switching through 1 GE links.

Could you possibly advise as to how packets might get dropped because of the difference in speed between the two links? Could bursts or micro bursts play a part in this?
If we were to go ahead with such a design what kind of packet loss should we expect and is this workable at all?
How could we avoid it? Or the solution is to only use 1 GE links all the way and just not bother about using 10 GE?
paolo bevilacqua Sun, 12/20/2009 - 17:58
User Badges:
  • Super Gold, 25000 points or more
  • Hall of Fame,

    Founding Member

As long the traffic offered does not exceed link(s) capacity, no drops will occour, and the speed difference is not a concern.

This is a general rule always true in networking.

Jonathan Marchand Sun, 12/20/2009 - 18:43
User Badges:

Hello and thanks for your reply!

Agreed, if the throughput doesn't go over the treshold, it should be fine.

However, I've had issues in the past where a link was switched at 100mb and then would switched at 10mb further down the path. The link was carrying PPP traffic.

The link utilization was well below 30% of usage.

As the traffic went from 100mb to 10mb "microbursts" would go above the CIR for a very short period of time, thus resulting in packet loss.

See the links below (about microbursts):

"I principally work on high end switching and the 4 main ways we see it
- Speed Mismatch (10G into 1G, 10G into 10M) More extreme the speed
mismatch the more dramatic the issue."

http:[email protected]/msg20663.html

Wouldn't this apply to this scenario?


Marvin Rhoads Sun, 12/20/2009 - 20:49
User Badges:
  • Super Silver, 17500 points or more
  • Cisco Designated VIP,

    2017 Firewalling, Network Management, VPN

Yes, UDP traffic in excess of the downstream link speed may be dropped if it exceeeds the intermediate device's ability to buffer it. Given that it is UDP traffic that's an unfortunate fact. Even policiing would just drop it in a bit bucket closer to the origin. If there was some traffic that you cared about more than others, you could apply QoS marking to it in order to give it a better shot at not being adversely afffected in scenarios of potential packet loss.

Jonathan Marchand Sun, 12/20/2009 - 21:27
User Badges:

Okay - if we were to use different hardware on site A and B, any feature/hardware we could use to ensure no packets are dropped? I guess shaping would be the best way, so hardware that supports this feature on gig and 10gig ports?

Is this even available on 10gig?



This Discussion