Router Multicast Inefficiency

[petergunn]

Seems like a 200-300 multicast packet/sec on the LAN will cause serious lockups on budget routers. Ive tried both a TRENDNet TEW-633GR and a Linksys WRT54G v4 flashed with DD-WRT. Enabling/disabling multicast in the config makes no difference.

I don't actually want LAN->WAN or LAN->wireless routing just LAN->LAN but I needs wireless->LAN TCP/IP access to manage my headless servers. I actually bought a Linksys SRW2008 to act as a LAN switch as the advertised iGMP snooping feature should have prevented multicast LAN->router traffic. Unfortunately the SRW2008 has badly broken firmware and iGMP snooping doesn't work so it floods all the ports including the one the router is attached to.

It does however have decent vlan support, but no mechanism for coloring the multicast data with a different vlan tag from other traffic. As a work around I am using Linux's vconfig command to create vlan tagged interfaces to let the application tag the data as its published. This works but complicates the system and application config.

Why does a few hundred irrelevant multicast LAN packets/sec have such an impact on these routers? Is there a way to get them to ignore this data?

Alternatively, can anyone recommend a decent gigE iGMP snooping switch that supports 64+ multicast groups?

Would it be possible to measure multicast performance as part of the SmallNetBuilder router/switch tests?

PG.

 

[thiggins]

Would it be possible to measure multicast performance as part of the SmallNetBuilder router/switch tests?

Interesting. This is the first time I have heard from anyone with interest in multicasting. What is your application?

 

[petergunn]

I'm utilizing synthetic raw market data feeds. These are usually a group of multicast streams (typically 9-24) of pricing information from stock exchanges like the NYSE/NASDAQ and other Automated Trading Centers (ATCs).

With a iGMP snooping switch (with a big enough multicast table) you can just add any old wireless router and it will be protected by the switch's multicast filtering. Enterprise environments would typically use a high end CISCO 6509s (with all the bells and whistles like sup720+DFCs) that retail for a 6 figure sum. In testing my D-Link DGS-2208 8-port gigE switched I was surprised to find it able to handle near wirespeed multicast throughput with zero packet loss and the lowest latency I've ever seen on any switch.

I'm probably a tiny niche - but from reading the Linksys forums the problem seems to be similar to people that have many multicast IP video streams (i.e. security cameras) and other niche groups that want to add WWAN access to their multicast enabled wired network.

As far as I can tell there are no metrics for routers/switches anywhere as to:

1) If iGMP snooping works, if so how many groups
2) If there is a 'filter multicast' option or some way to vlan tag multicast
3) If mutlicast is replicated in hardware or max. packets per sec
4) At what rate the router/switch becomes unresponsive

From a security/resilience perspective some people may be interested to know that without a capable iGMP implementation their gigE router/switch can likely be rendered unresponsive by a few hundred kilobits of multicast.

 

[thiggins]

Thanks for the info and the perspective on current uses for multicast. It sure ain't for web-based video!

If you can suggest a simple test, I'll see if I can implement it.

 

[petergunn]

If you can suggest a simple test, I'll see if I can implement it.

A very simple but perhaps meaningful test would be to try multicasting some video with VideoLan .

File -> Wizard -> Stream to Network... -> Next ... choose a video file

Then run another couple of Videolans elsewhere to display the multicast stream.

If you see pictures then just run a tcpdump on a separate port to verify that its not flooding and you're good to go.

If you have a IO::Socket::Multicast perl implementation then its not too hard to create a few stand alone benchmark scripts.

 

[thiggins]

Thanks. The test would be from WAN to LAN, correct?

 

[petergunn]

LAN->WAN would be a good test - especially if you fire up a few different publishers on the LAN. You'll probably find that the router has difficulty streaming hit bit rate videos to the WAN.

LAN->LAN with a WWAN router attached would be a good test as well - will let you know if your WWAN router is useable when people are streaming multicast on the LAN.

Fire up multiple publishers and see where things break

 

[shirsch]

And here I thought I was the only one...

Seems like a 200-300 multicast packet/sec on the LAN will cause serious lockups on budget routers. Ive tried both a TRENDNet TEW-633GR and a Linksys WRT54G v4 flashed with DD-WRT. Enabling/disabling multicast in the config makes no difference.

Wow. And I thought I was the only person in the world wrestling with this issue. Googled for a week solid before tripping over this post!

My ISP is a municipally-owned fiber based system providing 20Mb/s symmetrical internet service and IPTV delivery using multicast. Packet flooding from the latter brings a lot of routers to their knees.

A Linksys WRT54G v2 with stock firmware degrades by about 55% with a single multicast stream active. With two active, it falls on its face and cannot even manage 30Kb/s! Tomato firmware fared a bit better, with 5% and 40% degradation respectively. Turning on the mulicast filter does not mitigate the problem because the packet filtering is done in software (unlike some high-end PC adapters that have hardware-level filtering on the board). Whether or not it passes the packets through, they are still being seen at the interface and causing hardware interrupts at a pretty good clip.

In my case, the answer seems to have been moving to a more modern design with a higher-horsepower CPU. I've only been using it for a day, but a Netgear WGR614L has shown no degradation at all with two streams active. The processor runs at 200Mhz. (same as the Linksys), but I suspect the MIPS core used in the Broadcom chipset gets much more work done per clock cycle.

There must be about three people in the US with IPTV multicast delivery, based on the complete and utter lack of discussion.

 

[thiggins]

LAN->WAN would be a good test - especially if you fire up a few different publishers on the LAN. You'll probably find that the router has difficulty streaming hit bit rate videos to the WAN.

How many people multicast from their LAN, though?

 

[thiggins]

There must be about three people in the US with IPTV multicast delivery, based on the complete and utter lack of discussion.

Seems like it. I think multicast is more prevalent in Europe where IPTV is more common.

Doug Reid did a three-part series on Multicast almost 2 years ago and had trouble finding multicast sources to test. Perhaps that's changing.

 

[shirsch]

How many people multicast from their LAN, though?

That's not quite the point. I'm not multicasting from or within my LAN, nor do I want multicast traffic from outside getting into it. The issue here is that the WAN port on my router sees all the multicast packets sent by my ISP in the form of IPTV streams. They come directly off the optical interface to a dumb switch. From there, two ports feed the IPTV set-top boxes and a third goes to the WAN port of my router. The way multicast works, it "floods" out all ports of the switch.

In a perfect world, the ISP would install a smarter switch that could do multicast filtering/direction using IGMP. However, that's not the case and the choice of router has a huge impact on the bandwidth available for conventional IP traffic passing through it.

 

[thiggins]

Or you could install an inexpensive smart switch like the NETGEAR GS108T and enable the IGMP snooping and Block Unknown Multicast Address features?

 

[shirsch]

Or you could install an inexpensive smart switch like the NETGEAR GS108T and enable the IGMP snooping and Block Unknown Multicast Address features?

That looks like a nice unit! I'll keep that in mind if the WGR614L fails to work out. Otherwise, not looking for a reason to spend another $100+ at the moment

It wouldn't be the worst idea to find a repeatable and reasonable test for sensitivity to WAN multicast traffic. Doesn't FIOS distribute TV using multicast, or are they doing something else?

 

[thiggins]

Doesn't FIOS distribute TV using multicast, or are they doing something else?

Dunno. They might multicast to the FIOS boxes on the side of the houses. But don't know why they would need to multicast past that.

I'll see if Ixia has any ideas for multicast testing using IxChariot.

 

[britchie911]

HDHomerun Prime needs Multicasting IGMP snooping

Yes..very limited info in this topic. I have inquiry into Dlink (DGS-1228 Switch...supposedly supports IGMP snooping...although not sure of version i.e. IGMP v2 or v3) and NETGEAR WNDR3700v2 & v1 and a few GS108PE.

I'll post responses if anyone is interested in hearing what MFG has to say