Welcome to the Cisco Networking Professionals Ask the Expert conversation. This is an opportunity to discuss Deploying QoS For IP Networks with Cisco expert Santiago Alvarez. Santiago is a member of the Technical Marketing group within the Cisco IOS Technology Division. He focuses on MPLS and QoS technologies. Feel free to post any questions relating to Deploying QoS For IP Networks.
Santiago may not be able to answer each question due to the volume expected during this event. Our moderators will post many of the unanswered questions in other discussion forums shortly after the event. This event lasts through March 16. Visit this forum often to view responses to your questions and the questions of other community members.
If we configure advanced QoS/queuing mechanisms, are we likely to see performance problems with lower-end routers?
(For instance, a C2600 routing between ATM WAN and FastEthernet LAN, with LLQ, IP RTP priority etc. configured. - How will the router perform?)
It's important to understand the two tasks involved when we're queuing
packets. First, we have the decision of how a packet should be queued
(packet classification). This corresponds to packets entering the
queue. Second, we have the scheduling decision of what packet should be
serviced next out of all non-empty queues. This corresponds to packets
leaving the queues. The scheduling portion has a low impact on
performance. It is packet classification that can impact router
performance in some cases. Cisco IOS has support for a large number of
classification mechanisms (e.g. standard/extented ACLs, input interface,
NBAR, DSCP, etc) that provide maximum flexibility for defining both very
simple and very complex classification criteria. There is not easy
answer for performance impact on a particular platform. Final
performance is a function of the number of classes, the complexity of
traffic classification and the impact of other services enabled on the
router (e.g. ACL, NAT, PBR, H.323 Gateway, etc). New router platforms
provide specialized hardware to accelerate services such as packet
classification minimizing the impact on the router switching
Weve been using combination of Multilink/FairQueue on serial two-way connection before. It works perfect, providing efficient data transferring and 4 good quality voice trunks even on 64K satellite link.
Now we are thinking of deploying QoS for asymmetrical satellite IP Network one way serial uplink and broadband DVB type downlink. Due to different interfaces for uplink (serial) and downlink (Ethernet) what kind of technique could be applied to provide voice traffic priority?
Sakhinfo Technical Manager.
Your design doesn't need to change too much. The major difference is
that you don't need to have any LFI (Link Fragmentation and
Interleaving) mechanism in the downstream connection due to the higher
bit rate. You still need to use this mechanism in the upstream
direction (using MLPPP) to control the latency that large data packets
can introduce for voice packets if your upstream speed is lower than
786kbps. You don't mention the details of your current QoS solution,
but in general, LLQ is the queuing mechanism suggested for carrying
traffic that demands very low latency such as voice. This traffic
should be marked with a DSCP of 46 (EF) or an IP precendence of 5. LFI
is required in links slower.
Your network obviously is introducing lots of delays. Adding the serialization delay, propagation dealy, plus the 360ms x 2 (total 640ms) satellite delay, how is the quality of the voice affected? Every single article that deals with voice delays says that total delay should not exceed 200ms.
What kind of satelitte link do you use: K band or Cu band? Is it Frame Relay or PPP?
The challenging part of this problem is that it will be one way PPP (uplink)over Cisco serial port, relatively slow - 192K, which mean that we have to
use LFI (Multilink) and WFQ and RTP header compression. I don't know how one-way PPP will affect this - obviously we have to shut down keep alive. What else?
The downlink sounds like even bigger challenging. It's DVB stream over PAS8 in which we have 384K timeslot. Problem is that DVB receiver/transmitter is a separate unit with Ethernet interface. Actually it is point-to-point connection, but physical interface is LAN type (same as low speed DSL
bridge, but one way delay will be at least 320 ms)
Due to this:
1. How possibly we can use RTP on this link, may be some kind on tunnel or PPP over Ethernet technique?
2. Any ideas of using LFI ?
3. May be there is a way to combine this two connection into one virtual PPP, and than apply regular technique?
Is there any white papers, sample configurations or Cisco guidelines for implementing Voice over IP over ATM?
We are going to transmit Voice traffic (and other delay sensitive traffic too) with normal data over ATM.
There is some general ATM QoS/CoS sample configurations in CCO, but I would like to see some case studies focusing on ATM, VoIP and QoS.
Per-PVC LLQ for ATM was introduced in 12.0(7)T (dLLQ for the 7500 was
released on 12.1(5)T.) This is a key feature to enable VoIP over ATM.
From a QoS point of view, there's not much difference between this
solution and solutions such as VoIP over Frame Relay. An important
feature, it's the possibility of taking advantage of ATM CoS to support
IP QoS. In that case, multiple ATM PVCs with different classes of
service can be configured between two sites and map IP precendence
values to PVCs that support the QoS expected for that marking. These
are some URLs that you may find useful regarding this topic:
I'll bring your request for a VoIP-over-ATM case study to the attention
of the owners of the TAC page on CCO
I've read that material through and from my point of view the PVC Bundle/Multiple service classes seems to be more scalable solution. Least if we need to classify other delay sensitive traffic besides VoIP.
Is there any known Caveats with configuring PVC Bundles - some trouble with CAR + PVC Bundling? How many members a Bundle can have? How about other PVCs that are not Bundle members?
I have a 64k link with 4 E&M and mail traffic. The quality isn't so good. I've tried all configurations i think best and some that i found and still the same. Is there any chance of better voice quality without increasing the link?
Let's assume that you are using the default codec (G.729) that requires
8kbps plus overhead. The IP+UDP+RTP headers will consume 16kbps alone
for a total of 24kbps per call. With this configuration, you're only
able to accomodate two calls. Enabling RTP header compression takes
down the required bandwidth per call to around 10kbps. At this point,
you would conclude that it's safe to accomodate four calls on a 64k
link. However, there's more overhead that needs to be taken into
consideration. You have layer-two headers (PPP/Frame Relay), layer-2
keepalives, other background protocols (e.g. CDP). Additionally, you
HAVE to enable fragmentation and interleaving to reduce the delay that
large data packets introduce to voice packets. There's additional
header information that needs to be added to individual fragments. On
top of this, you're trying to pass mail traffic.
I'd suggest that you review your configuration and make sure you're
using LLQ, a low-rate codec, fragmentation and RTP header compression.
There's still some room for optimization by using a very low-rate codec
and modifying the size of your voice payload. If you don't find the
voice quality acceptable after these optimization, you'd just have to
accept that no QoS technique can create bandwidth for you.
Thanks for the advice but all that is configurated already. The solution should be increase the bandwidht after all. Thanks again
I'm surprised that 64K is not enough to provide good voice quality. We are currently using 64K satellite link for 4 E&M trunks between two Nortel PBX and none of customers could recognize the voice quality difference between inbound and outbound calls. Ip data transfers also goes very smooth. The only problem was call setup time before we upgrade IOS to support H323v2. On satellite link with ping delay 700 msec setup time were more than 10 seconds - now it's 2-3 second, which is OK.
So, I guess there is something wrong in your configuration.
Can you give one configuration that i can try? In 64k i pass 4 voice channels, email traffic, 4 or 5 users of AS/400.