Multicast Issues Between 2 Switches

[silvio]

There is no interface for 192.168... at the querier switch (48p). So I think problems with sources 192.168... are related there. But which traffic are affected? What is the problem? I see a lot of successful MC-traffic.
regards
Silvio

[silvio]

You use 224.2.... MC-Adresses. This are reserved adresses für SDP/SAP (see network guide an IANA). So here I am also not sure about the behavior in the switches.
So I assume there are some ideas for better design (see last posts).
regards
Silvio

[mperezATX]

there are no missing sources and i just finished putting all sources on the 10.0 network with 255.0.0.0 subnet. there is now change in behavior. in looking at the port health page, im seeing some numbers and icons on both uplink ports that perhaps indicate what the issue might be? Please see attached. everything else is green on both switches.

thanks again for your help,

[mperezATX]

i changed all the sources to have a 239.#.#.# address and so far everything looks good. the icons in the health page are still yellow and red on the uplink ports but i can at least see the streams with no issues for now. the issue was that when i were to pull up a mc source from 48 port switch off the 24 port switch, there would be tons of noise, jitter, etc. Pretty much you couldn't see the source cleanly.

[mperezATX]

didn't last long. im seeing errors again and actually when i pull up sources from the 24 port switch on the 48 port switch, they stop broadcasting within 10 seconds. really bizarre

[silvio]

Hi,
screenshot means no direct problem. You uplinks are working at their capacity. The thresholds are are exceeded for some times (more than 80% of 1Gig).
Now I understand your problem: the mc-streams are not interupted, but problems with noise and jitter. I assume the issue is there if you have a lot of traffic at the same time.
So my next questions:
- what type of traffic is the MC (video, special application)? How many kbps for any stream?
- And is there other traffic that made the buffers and queues full (not MC-traffic)?
- How are the MC-clients are connected - 1Gig or 100Mbps?

For improvement I have some ideas:

1. instead of 1Gig (I suggest that is your uplink now) you can upgrade to 10Gig (with PERF-License and new SFP+).

2. Or at least double the speed with a linkagg between both switches. But don't forget to activate the loadbalancing also for MC with:

Code: Select all

hash-control load-balance non-ucast enable

3. QoS: If you have a lot of other (not time critical) traffic between the switches you can priorise your more important MC traffic. F.e:

Code: Select all

policy condition MC destination ip 239.0.0.0/8
policy action PRIO4 dscp 38 priority 4
policy rule MC condition MC action PRIO4
qos apply

4. 48Port -> 2x 24Port: there is a known limitation in the buffers of the 48port models. I am not sure if this is your problem. I have seen this behavior with f.e. only one MC-Source to a lot of Clients connected with 100Mbps. So same MC-stream to a lot of clients. In your case I mostly see a 1to1 connection (server-client) - correct? Maybe you have the chance to exchange the 48port with 2x 24port for testing....

regards
Silvio

[mperezATX]

i'm not in the office today but i will give it a shot with the LAG when i come in tomorrow. this is for a test lab so we likely don't want to spend money on the 10gb setup for now. that totally makes sense though; the sources that used to have the 192.168... each has an output of around 82 mb and the rest range from 1-80 as well so we are most likely overdriving the uplink ports. All the mc traffic is video with audio and the only other traffic is low bitrate information streams from 6 other devices. essentially all our devices on the network have 2 ethernet ports, one is for management (separated on a separate switch) and the other is exclusively for mc traffic and other data such as electronic program guide info, conditional access info (think cable packages for you tv cable at home) set top box upgrades, and cable tv network information that goes out to the cable boxes. The MC clients all use 1gb connections and if i had to guess, the total bitrate of all the non mc traffic may be around 40-60 mb.

thanks again

[mperezATX]

when i enter the hash-control command, i get the following:

-> hash-control load-balance non-ucast enable
^
ERROR: Invalid entry: "load-balance"

[silvio]

yeah, this was the command for R8 switches sorry.

For 6450 the following is important (a problem from the knowledgebase):
When 5 consecutive ip addresses are used for the multicast groups, index is calculated using XOR (hash-control mode) in multicast source table which results in same hash value. When more than 4 copies of a single hash are present, collisions occur in the forwarding database. Hence we are unable to see the 5th multicast source.

Issue solved after changing the hash-control mode from XOR to CRC

-> hash-control mode fdb crc



With the following command you can check the outgoing counters of both linkagg-ports:
show interfaces 1/25 counters -> OutMcastPkts has to be similar values (after clearing to 0).

[mperezATX]

i was finally able to do this today and after a few hours everything looks good. hopefully it will remain that way from now on i'll let you know if it breaks again.

thanks for all your help,