Want to throttle bandwidth on specific tcp ports using a 2811

[keithsauer507]

Hello,

I am working on vmware vsphere replicaiton and this does not have any bandwidth management in it at all.  However between our production site and DR site we have a Cisco 2811 router and the WAN is 40 meg for the lan and 16 meg for the voice / voip system dr.

If I do not throttle the vsphere replication, it will use all of the 40 meg constantly as I have monitored using cacti bandwidth graphs of this interface.  I would like it to use as much bandwidth that is available, however I want other types of traffic to take priority such as the end users at this branch office file and print, ssh, web surfing, https, voip, microsoft lync, etc...

vSphere replication uses the following tcp ports:

• Port 31031 for initial replication
• Port 44046 for on-going replication

Is there an easy way to create a policy-map or something to match on those port numbers (either direction) to give it a lower priority whereas it will use whatever bandwidth it can, but set it aside if other ports / traffic are more in demand?

This is what we have right now as far as class and policy maps.  The WAN is tied to FIFTYMEG but limited to 40 because any more than that and Verizon squashes it and we have bandwidth all over the place.  Verizon insists nothing is wrong, but it works at 40 meg so that explains that.  The TENMEG is because we have a small site of a few people that have a 10 meg wan connection with Verizon.

class-map match-any RT
 match  dscp ef
 match  dscp af41
 match  dscp cs3
!
!
policy-map QOS50
 class RT
  priority 10000
  set cos 2
policy-map FIFTYMEG
 class class-default
  shape average 40000000
  set cos 0
  service-policy QOS50
policy-map QOS10
 class RT
  priority 2000
  set cos 2
policy-map TENMEG
 class class-default
  shape average 10000000
  set cos 0
  service-policy QOS10
policy-map QOS16
 class RT
  priority 10000
  set cos 2
policy-map SIXTEENMEG
 class class-default
  shape average 16000000
  set cos 0
  service-policy QOS16

Here is the sub interface that connects to the site that I want to replicate to:

interface FastEthernet0/1
 description VZ FE CONNECTION
 no ip address
 duplex full
 speed 100
!
interface FastEthernet0/1.296
 description 50MEG TO DR
 encapsulation dot1Q 296
 ip address 192.168.100.1 255.255.255.0
 service-policy output FIFTYMEG
!

Any help would be appreciated.

[Joseph W. Doherty]

Disclaimer

The Author of this posting offers the information contained within this posting without consideration and with the reader's understanding that there's no implied or expressed suitability or fitness for any purpose. Information provided is for informational purposes only and should not be construed as rendering professional advice of any kind. Usage of this posting's information is solely at reader's own risk.

Liability Disclaimer

In no event shall Author be liable for any damages whatsoever (including, without limitation, damages for loss of use, data or profit) arising out of the use or inability to use the posting's information even if Author has been advised of the possibility of such damage.

Posting

You're on the right path.  You want to shape for available bandwidth and then use a subordinate policy to treat any shaped congestion.

Your subordinate policies already have a class for RT, using LLQ, so you just need match your replication traffic and place it into one class, with a low bandwidth allocation (perhaps even minimum possible) and also I would suggest explicitly defining class-default so you can explicitly define bandwidth for it or use FQ.  (NB: if you use FQ, in pre-HQF QoS, you won't be able to define bandwidth for class-default, so be careful how much bandwidth you allocate to the replication class.)

If the replication traffic can be identified by the ports you've noted, you write an ACL to match against those and then have your class-map use that ACL.

PS:

BTW, regarding the Verizon providing 50 Mbps, but you seem to need 40 Mbps, that's because (I believe) many shapers don't account for L2 overhead, but Verizon does.  Often I've found you need to shape at least 10 to 15% slower than the nominal bandwidth.  Your 20% is in the ballpark.  Unfortunately, L2 overhead, as a percentage, varies per packet size.  To insure you're okay with something like VoIP, you need to shape more toward worst case overhead.  (You might want to make also reduce your shape values for your 10 and 16 Mbps caps.)

(NB: I would guess you see this issue on your 50 Mbps link because your replication routinely will fill to capacity.)

Also BTW, what's the CPU loading of your 2811 look like?  My experience has been they top out at about 40 Mbps aggregate.

[michael kalinowski]

Hi Keith,

I am looking at trying to do the same thing on a 20 meg link.....did you come up with a good solution?

Thx!