QoS Design

[dcanady55]

Hello,

I understand this is a very broad complex topic but was hoping someone could answer the following for me or point me in the direction for current info.

These questions are based off a video I watched presented by a member of Cisco that works solely on QoS but this was from 2013 and maybe the design has changed significantly.

I understand that our Mitel phones mark the packet with DSCP 46 as I set this up using DHCP and I get how this number gets treated through out the network. In the video he talked about using 5 class model to start off with then move to 8 or 12 if need be. How does other applications get put in the Best Effort which is DF or Transnational Data AF2 according to his slides (I have attached this slide as well)? Do I setup ACLs on the switch and define it this way? I have attached a picture of a slide showing recommend percentages and I am wondering if this is for the up-link to the router that should adhere to this or every interface?

This slide shows 4 categories so within each category like best effort would you or could your prioritize traffic that would get preferred treatment in the best effort category?

Basically, I would like to setup QoS in the following matter.

Voice first

Then we have a number of applications that would be considered core applications and i want them treated better

basic internet traffic can be best effort along with non production applications

scavenger/bulk like windows updates or backups would have very little of the pipe

we don't use video yet but that is coming down the road and would be slotted in after voice from what I've seen so far.

thanks for any help.

[Joseph W. Doherty]

I understand that our Mitel phones mark the packet with DSCP 46 as I set this up using DHCP and I get how this number gets treated through out the network.

Are you sure you do?  Packets having an DSCP EF marking doesn't guarantee special treatment for those packets.

How does other applications get put in the Best Effort which is DF or Transnational Data AF2 according to his slides (I have attached this slide as well)?

I'm unsure of your usage of "get put in".  Are you asking how packets actually get marked or how they get treated by their DSCP marking?  The 5-class model is just that, a logical model, that indicates how different kinds of traffic are classified into different classes and, in this instance, also how packets in those classes use specific DSCP markings to indicate they belong to that class of traffic.

Do I setup ACLs on the switch and define it this way?

You might, might not too.  Packet tagging can be done at the source, as you noted for your Mitel phones.

I have attached a picture of a slide showing recommend percentages and I am wondering if this is for the up-link to the router that should adhere to this or every interface?

Another you might, or might not.  Your QoS policy should serve your QoS service needs.  Often you might be able to use a single logical QoS policy.  If you have different platform hardware, you often need different physical QoS policies.

This slide shows 4 categories so within each category like best effort would you or could your prioritize traffic that would get preferred treatment in the best effort category?

Same class traffic generally shares the same forwarding priority however same class traffic may have different drop priorities.

[dcanady55]

Joseph,

I was asking how packets get marked in the first place? Besides the phones getting statically assigned this number via DHCP I am not sure how applications would get marked. Then you bring up another good point but once traffic gets marked with a certain DSCP how does that marking affect its way through the network.  Clearly, I have a lot to learn on this subject so if you have any suggestions on reading material I'm all ears.

[Joseph W. Doherty]

I was asking how packets get marked in the first place?

They can be marked by the application that generates them and/or then can be marked (or remarked) by network devices along the path the packet takes.  (BTW, understand packets don't need ToS markings to be provided QoS treatment, conversely ToS markings don't guarantee any QoS treatment.)

Then you bring up another good point but once traffic gets marked with a certain DSCP how does that marking affect its way through the network. 

That's up to the devices the packet transits.  (Again, network device might not provide any QoS treatment.  Further, if it does provide QoS treatment, it may differ from other devices along the path.)

. . . if you have any suggestions on reading material I'm all ears.

Unfortunately I don't have any suggested QoS reading material, as I haven't come across QoS material that explains QoS as I believe it should be explained.

I can suggest you read as much as possible, but take what you read with a very large grain of salt.  If you read much on the subject of QoS, you will come across good information, unfortunately it's difficult to identify.  You'll also find, as is often the case, the "text book" overlooks real world problem issues.

For example, when reading about QoS you might find something like telnet should be treated good (because of its low bandwidth usage and latency sensitivity), FTP should be treated bad (because of its high bandwidth usage, not being latency sensitive), but what about traffic from a telnet session showing the whole Internet route table (w/o pause) vs. FTP of a file containing a sentence or two?  Or, if telnet is good and FTP is bad, is SSH also good and SCP also bad, and if so, how do you tell them apart?  How do you classify HTTP/HTTPS traffic or Microsoft's NetBEUI traffic as so many different applications might use either for transport.

[dperezoquendo]

Hello,

Depending on how your network works and how traffic flows, I believe you would generally mark traffic either coming in or coming out with a policy-map. I think the attachments you've provided are a good base to start from. For example:

ip access-list extended NETWORK_MGMT

 permit ip x.x.x.x x.x.x.x

class-map NETWORK_MGMT

 match access-group name NETWORK_MGMT

policy-map MARK-IN-POLICY

 class NETWORK_MGMT

   set dscp cs6

int g0/1

 service-policy input MARK-IN-POLICY

For the actual policing or shaping, I think that's the tricky part that is quite broad. I expect much testing and planning to get the best results for your environment. There is no magic number or configuration in my opinion as each environment and their needs differ from one another.

So it's also important you know where to place these policy-maps as well. I would recommend diagraming you're network out if one doesn't exist. This can help best determine the placements for your marking/classification and then policing/shaping.