DiffServ-Aware Traffic Engineering (DS-TE)
So far, you've seen that DiffServ service with MPLS packets is basically the same thing as with IP packets, with a few differences in configuration. Overall, MPLS packets with a given EXP setting are treated just like IP packets with a given IP Precedence setting.
There's more to MPLS TE and DiffServ than just applying IP mechanisms to MPLS packets, though. The rest of this book spends a great deal of time showing you that there's benefit in making a headend resource-aware, because the headend can then intelligently pick paths through the network for its traffic to take. However, after you add QoS, you're almost back at square one. Things are better than they were—you can steer IP traffic away from the IGP shortest path in a resource-aware fashion. What you can't do is steer traffic per QoS. If you have traffic destined for a particular router, all that traffic follows the same path, regardless of the DSCP/EXP settings on that packet.
To expand on this point further, suppose you are a service provider offering four classes of service: Low latency, Gold, Silver, and Bronze. Low latency being self-explanatory, suppose Gold is defined as guaranteed delivery. Edge QoS is applied to mark EXP bits to differentiate your low-latency traffic from Gold and other traffic. Now, if the forwarding is based on IGP best path alone, all your traffic, regardless of what class it belongs to, is forwarded to a downstream neighbor dictated by your routing table. Even if you use MPLS TE and mpls traffic-eng autoroute announce, you are still limited by the routing table and how it decides to forward traffic.
The problem with this implementation is that when you have a congested link at a downstream node along the forwarding path, even though most low-latency traffic might get through, some of the Gold traffic might get dropped. This congestion knowledge is localized at the downstream node and is not propagated back to the edge devices that send traffic down that path. As a result, your edges continue to send traffic to the same downstream router that continues to drop some of your Gold traffic. What is needed to fix this situation is per-class call admission control—and this is exactly what you get when you combine DiffServ and TE. Surprisingly enough, this combination is called DS-TE.
This might lead you to believe that all you need to do is mark the EXP bits according to your QoS policy and let them ride over TE tunnels, conquering all the problems.
Not quite. The problem with TE, as it's been discussed so far, is that it doesn't do admission control on a per-QoS class basis.
TE certainly offers call admission control in addition to the PHB offered by DiffServ. This takes care of the first problem—sending more traffic down a certain path than there is available bandwidth—while queuing your higher-priority traffic ahead of your low-priority traffic.
A second problem is that there might be contention between different high-priority traffic streams. For example, if you sold two voice pipes to two customers, both with a low-latency requirement, if you forward both streams down the same path which is experiencing congestion, both streams might be affected. Remember, for voice, it is better to drop a call than to get degraded service. So how does DS-TE solve this problem?
DS-TE allows you to advertise more than one pool of available resources for a given link—a global pool and things called subpools. A subpool is a subset of link bandwidth that is available for a specific purpose.
The current DS-TE implementation allows you to advertise one subpool. Think of this as a pool with which you can advertise resources for a separate queue. The recommended use of the DS-TE subpool is to advertise bandwidth in your low-latency queue on a link, but you can do whatever you like with this subpool. The rest of this chapter assumes that you are advertising LLQ space using the subpool.
It's important to understand that DS-TE and its subpools are, like the rest of MPLS TE, control-plane mechanisms only. If you reserve 10 Mbps of global bandwidth, it's probably not a good idea to send 100 Mbps down that LSP. DS-TE behaves the same way. Building a separate control-plane reservation for a subpool doesn't mean that any special queuing policy is enforced as a result of this reservation. The actual queuing behavior at every hop is still controlled by regular DiffServ mechanisms such as LLQ; what DS-TE buys you is purely the ability to reserve queue bandwidth, rather than just link bandwidth, in the control plane. This ability lets you build TE-LSPs that specifically reserve subpool bandwidth and carry only LLQ traffic, and in effect build a second network on top of the one you already have. You have a network of physical interfaces and global pools and a network of subpools. This lets you get better resource utilization out of your network.
Going back to the example of two voice streams that might contend for the same low-latency bandwidth along the same path, this would not happen with DS-TE. When you try to build the second LSP requesting low-latency bandwidth and that bandwidth is not available along a certain path, the LSP is not signaled. If you used dynamic path options to configure your DS-TE tunnels, using CSPF on the router might find you an alternative path that meets the subpool reservation.
But how do you configure all this?
Configuring DS-TE
The best way to understand DS-TE is to see it in action. DS-TE is easy to configure. The configuration pieces provided in this section are for a 7200 series router with an OC-3 interface. The OC-3 interface is advertising 150 MB of global RSVP bandwidth. 45 Mbps (30 percent) of this is subpool bandwidth. The OC-3 also has 45 Mbps configured in a low-latency queue. Other configurations vary by platform and according to the goal you're trying to achieve.
There are five pieces to configuring DS-TE:
Per-link subpool bandwidth availability
Per-link scheduling
Headend subpool bandwidth requirements
Headend tunnel admission control
Tunnel preemption
Per-Link Subpool Bandwidth Availability
On each hop that you want to advertise a subpool, you configure the following command:
ip rsvp bandwidth interface-kbps sub-pool kbps
This is merely an extension of the ip rsvp bandwidth command discussed in earlier chapters. The interface-kbps parameter is the amount of bandwidth (in Kbps) on the interface to be reserved. The range is 1 to 10,000,000. sub-pool kbps is the amount of bandwidth (in Kbps) on the interface to be reserved as a portion of the total. The range is from 1 to the value of interface-kbps.
For this specific example of an OC-3 with 150 Mbps in the global pool and 45 Mbps in the subpool, the necessary configuration is as follows:
ip rsvp bandwidth 150000 sub-pool 45000
Per-Link Scheduling
Advertising a subpool in DS-TE doesn't change any packet queuing behavior on an interface. You need to configure the forwarding-plane LLQ mechanisms in addition to the control-plane subpool mechanisms. The subpool advertisement and LLQ capacity are usually set to the same value, but you can set them differently if you want to do fancy things such as oversubscription or undersubscription.
Per-link scheduling is just whatever LLQ mechanism exists in your platform. With MQC, the priority keyword builds a low-latency queue. The MQC LLQ configuration for this example is as follows:
class-map match-all voice
match mpls experimental 5
policy-map llq
class voice
priority percent 30
interface POS3/0
service-policy output llq
Headend Subpool Bandwidth Requirements
On the tunnel headend, you use the following command:
tunnel mpls traffic-eng bandwidth sub-pool kbps
This is the same thing as the tunnel mpls traffic-eng bandwidth kbps command that was covered earlier in this book, except that you are telling the headend to do its path calculation and bandwidth reservation based on the advertised subpool bandwidth.
You are allowed to have only one type of reservation per tunnel. If you try to configure a tunnel with the command tunnel mpls traffic-eng bandwidth followed by the command tunnel mpls traffic-eng bandwidth sub-pool, the sub-pool command overwrites the global pool command.
Headend Tunnel Admission Control
The next piece is controlling what traffic goes down the tunnel. There are three steps to this:
|
Step 1 |
Make sure no more traffic enters your network than you have sold. |
|
Step 2 |
Make sure that the only traffic to enter the DS-TE tunnel is traffic that belongs there. |
|
Step 3 |
Make sure your TE tunnel reservation accurately matches your traffic requirements. |
The first step is a generic DiffServ architecture component; it has nothing to do with MPLS TE. You can't have any kind of QoS guarantees if you provision your network for a certain amount of traffic and then send to that network far more traffic than you have provisioned for.
The second step is where things start getting TE-specific. The general idea is that if you've decided that your DS-TE subpool is for EXP 5 traffic, you send only EXP 5 traffic down the DS-TE tunnel. Why? Because if you send nonsubpool traffic down a DS-TE tunnel, this traffic interferes with other traffic at every hop. Let's say you have a tunnel for 10 Mbps of subpool bandwidth that you're mapping to a low-latency queue for EXP 5 traffic. If you send 100 Mbps of EXP 0 traffic down this DS-TE tunnel, that EXP 0 traffic is put into the same queue as other EXP 0 traffic, but it is not reserved from the network. You'll read more about this in the section "Forwarding DS-TE Traffic Down a Tunnel."
The reason for the third step should be obvious: If you don't reserve bandwidth in accordance with what you're actually sending, and your reservations are designed to track LLQ capacity at every hop, you run the risk of overloading the LLQ and providing poor service. One way to make this adjustment is to size your DS-TE tunnels based on the amount of DS-TE tunnel bandwidth you have provisioned for. The other way is to use something such as auto-bandwidth (see Chapter 5, "Forwarding Traffic Down Tunnels") to adjust tunnel size based not on the provisioned load, but on the real-life traffic load.
Tunnel Preemption
The last thing you have to do is make sure that the subpool tunnel can preempt nonsubpool tunnels. This is done with the command tunnel mpls traffic-eng priority, which was covered in Chapter 3, "Information Distribution." Why do you have to allow DS-TE tunnels to preempt non-DS-TE tunnels? Because of the way subpool bandwidth is advertised. The subpool is, as the name implies, a subset of the global bandwidth pool on an interface, not a separate bandwidth pool. If you don't allow the DS-TE LSP to preempt a global LSP, the global LSP can reserve bandwidth that would then no longer be available for the subpool.
If the subpool bandwidth were advertised as a separate bandwidth pool, rather than as a subset of existing bandwidth, you could easily end up with bandwidth fragmentation, as the following example shows.
If you have a link configured with ip rsvp bandwidth 150000 sub-pool 45000, this means that you are advertising that you have 150 Mbps (150,000 Kbps) of global reservable bandwidth on the link, and 45 Mbps of that 150 Mbps is subpool bandwidth. At this point, the link is advertising the output shown in Example 6-7.
Example 6-7 Available Bandwidth Before Any Reservations Have Been Made
vxr12#show mpls traffic-eng topology 192.168.1.12
IGP Id: 0168.0001.0012.00, MPLS TE Id:192.168.1.12 Router Node id 2
link[0 ]:Nbr IGP Id: 0168.0001.0001.00, nbr_node_id:3, gen:1
frag_id 0, Intf Address:2.3.4.12, Nbr Intf Address:2.3.4.1
TE metric:10, IGP metric:10, attribute_flags:0x0
physical_bw: 155000 (kbps), max_reservable_bw_global: 150000 (kbps)
max_reservable_bw_sub: 45000 (kbps)
Global Pool Sub Pool
Total Allocated Reservable Reservable
BW (kbps) BW (kbps) BW (kbps)
--------------- ----------- ----------
bw[0]: 0 150000 45000
bw[1]: 0 150000 45000
bw[2]: 0 150000 45000
bw[3]: 0 150000 45000
bw[4]: 0 150000 45000
bw[5]: 0 150000 45000
bw[6]: 0 150000 45000
bw[7]: 0 150000 45000
This shows a link with no reservations across it, a global pool of 150 Mbps, and a subpool of 45 Mbps.
Let's send three reservations across this link: a reservation for 60 Mbps from the global pool, 20 Mbps from the subpool, and 50 Mbps for the global pool, in that order. After the first reservation, the router advertises the bandwidth displayed in the output of show mpls traffic-eng topology, as shown in Example 6-8. All LSPs are set up with the default priority of 7.
Example 6-8 Available Bandwidth with 60 Mbps Reserved
vxr12#show mpls traffic-eng topology 192.168.1.12
IGP Id: 0168.0001.0012.00, MPLS TE Id:192.168.1.12 Router Node id 2
link[0 ]:Nbr IGP Id: 0168.0001.0001.00, nbr_node_id:3, gen:3
frag_id 0, Intf Address:2.3.4.12, Nbr Intf Address:2.3.4.1
TE metric:10, IGP metric:10, attribute_flags:0x0
physical_bw: 155000 (kbps), max_reservable_bw_global: 150000 (kbps)
max_reservable_bw_sub: 45000 (kbps)
Global Pool Sub Pool
Total Allocated Reservable Reservable
BW (kbps) BW (kbps) BW (kbps)
--------------- ----------- ----------
bw[0]: 0 150000 45000
bw[1]: 0 150000 45000
bw[2]: 0 150000 45000
bw[3]: 0 150000 45000
bw[4]: 0 150000 45000
bw[5]: 0 150000 45000
bw[6]: 0 150000 45000
bw[7]: 60000 90000 45000
This is because 60 Mbps was reserved from the global pool (at the default setup and holding priorities of 7/7), leaving 90 Mbps available on the link. Of that 90 Mbps, 45 Mbps can be reserved as subpool bandwidth should anybody want it.
The next reservation to come across takes 20 Mbps from the subpool. This means that the available bandwidth on the interface is now as advertised in Example 6-9.
Example 6-9 Available Bandwidth with an Additional 20 Mbps of Subpool Bandwidth Reserved
vxr12#show mpls traffic-eng topology 192.168.1.12
IGP Id: 0168.0001.0012.00, MPLS TE Id:192.168.1.12 Router Node id 2
link[0 ]:Nbr IGP Id: 0168.0001.0001.00, nbr_node_id:3, gen:4
frag_id 0, Intf Address:2.3.4.12, Nbr Intf Address:2.3.4.1
TE metric:10, IGP metric:10, attribute_flags:0x0
physical_bw: 155000 (kbps), max_reservable_bw_global: 150000 (kbps)
max_reservable_bw_sub: 45000 (kbps)
Global Pool Sub Pool
Total Allocated Reservable Reservable
BW (kbps) BW (kbps) BW (kbps)
--------------- ----------- ----------
bw[0]: 0 150000 45000
bw[1]: 0 150000 45000
bw[2]: 0 150000 45000
bw[3]: 0 150000 45000
bw[4]: 0 150000 45000
bw[5]: 0 150000 45000
bw[6]: 0 150000 45000
bw[7]: 80000 70000 25000
The available bandwidth on the interface is an important point to understand. 20 Mbps of additional bandwidth is reserved from the available link bandwidth. This brings the total allocated bandwidth from 60 Mbps to 80 Mbps and decreases the total available link bandwidth from 90 Mbps to 70 Mbps. It just so happens that the 20 Mbps reserved was taken from the subpool, which means that the available subpool bandwidth is now 25 Mbps. This is because the subpool is considered a subset of the global pool, rather than a whole separate pool.
The third reservation for 50 Mbps from the global pool brings the available link bandwidth down to that advertised in Example 6-10.
Example 6-10 Available Bandwidth with an Additional 50 Mbps of Subpool Bandwidth Reserved
vxr12#show mpls traffic-eng topology 192.168.1.12
IGP Id: 0168.0001.0012.00, MPLS TE Id:192.168.1.12 Router Node id 2
link[0 ]:Nbr IGP Id: 0168.0001.0001.00, nbr_node_id:3, gen:5
frag_id 0, Intf Address:2.3.4.12, Nbr Intf Address:2.3.4.1
TE metric:10, IGP metric:10, attribute_flags:0x0
physical_bw: 155000 (kbps), max_reservable_bw_global: 150000 (kbps)
max_reservable_bw_sub: 45000 (kbps)
Global Pool Sub Pool
Total Allocated Reservable Reservable
BW (kbps) BW (kbps) BW (kbps)
--------------- ----------- ----------
bw[0]: 0 150000 45000
bw[1]: 0 150000 45000
bw[2]: 0 150000 45000
bw[3]: 0 150000 45000
bw[4]: 0 150000 45000
bw[5]: 0 150000 45000
bw[6]: 0 150000 45000
bw[7]: 130000 20000 20000
The total allocated bandwidth is 130 Mbps (60 Mbps + 20 Mbps + 50 Mbps). 20 Mbps of that 130 Mbps is allocated from the subpool, and 110 Mbps (60 Mbps + 50 Mbps) is allocated from the rest of the global pool.
If you ponder this for a while, making the subpool a subset of the global pool rather than a whole separate pool makes sense. If the global pool and the subpool bandwidth were segregated, instead of advertising a pool of 150 Mbps available bandwidth (45 Mbps of which is in a subpool), the router would advertise a 105 Mbps pool and a 45 Mbps pool.
If you then have the same reservations, in the same order as before, the bandwidth reservations and availability look like what is shown in Table 6-8.
Table 6-8 What If You Advertised Global and Subpool as Separate Pools?
|
Reservation |
Global Available Bandwidth |
Subpool Available Bandwidth |
|
No reservations |
105 Mbps |
45 Mbps |
|
60 Mbps global |
45 Mbps |
45 Mbps |
|
20 Mbps subpool |
45 Mbps |
25 Mbps |
|
50 Mbps global |
????? |
?????? |
But what about the third reservation for 50 Mbps from the global bandwidth pool? It fails to get through, even though a total of 70 Mbps of bandwidth (45 + 25) is available.
Advertising DS-TE bandwidth as a subpool rather than as a separate pool means that you can reserve up to 45 Mbps for the subpool, but if it's not in use, other nonsubpool tunnels can use it.
If you want subpool tunnels to be able to take up to 45 Mbps of bandwidth no matter what else is reserved on the link, the subpool tunnels need to be able to preempt other tunnels on the link. However, see the section in Chapter 9, "Network Design with MPLS TE," titled, "The Packing Problem," before you start messing with tunnel priorities.