networking problems probably perhaps multicast

[Jochen Müller]

Hi there,

we temporarily have networking problems and I could break it down to multicast suspectively causing them.

What we have:
- a ProCurve 5412 core switch
- some other switches (also older models, eg. 2810 or 2900)
- all servers connect to the core switch
- all HP hardware servers
- some physically installed but also some VMware vSphere servers, together about 70 Windows Servers
- one of them is a Office Communication Server 2007 (OCS)
- one is an Exchange 2007 Hub/Cas
- there is a NLB in multicast mode configured for the Hub/CAS server but he is the only member of this NLB
- some networking device (serial2LAN adapter for a serial GSM modem) which reacts very sensibly on networking timeouts.

The problems occur:
1) when vRanger backups the VMs or
2) when we copy a large amount of data from or to the Hub/CAS server over the NLB hostname/IP or
3) when we want to make conferencing calls over the OCS

The result:
Network ping times to some networking devices permanently rise up to high in the 3000ms, some pings fail. Especially the serial2LAN box resets the connection to the connected COM port modem due to the networking problems. Also, some other networking devices report connection problems, e.g. wimo devices need to get restarted so that the wireless site connection comes up again.
Speech and/or video is not satisfying on conferencing calls over the OCS.

I suspect "Multicast filtering (IGMP)" to have activated on all VLANs on all Switches that these problems disappear. This feature is currently turned off at all VLANs, also there is no Spanning Tree configured (yet).

Am I right?

And if so: what will it cause in the moment I turn on IGMP? Will there be some significant lag or downtime or something like that? Will servers have to have reconnected or rebooted?

Any hints would be appreciated...

Best regards, Jochen

[Oleg Sukharev]

Jochen,

It seems that multicast traffic is flooded to all ports of 5412 and other switches.

You can try to enable IGMP for all VLANS. STP is not needed to be activated. Also the switch that is central in multicast flows can be made IGMP querier (some kind of multicast router).

Commands look like this:

vlan ip igmp
vlan ip igmp querier

To define sources of high taffic you can connect to switch web server using web browser. The page will show dynamic port usage via Java applet.

HTH,
Oleg

[Jochen Müller]

Hi Oleg,

thank you for your reply.

You wrote I can try to enable IGMP for all VLANS. So do you think that would solve the problems I described above? And what will happen in the moment I activate it? Will there be some lag or loss? Should I plan some downtime for that or do it while inspection windows?

I know that STP is not needed to be activated but we plan to activate RSTP or MSTP since we want to create backup paths in future - but that has nothing to do with this problem.

You wrote: Also the switch that is central in multicast flows can be made IGMP querier (some kind of multicast router).
By default there is none, neither a multicast router nor a IGMP querier. Am I right? Should I then either create one of them, in this case preferably the IGMP querier on the 5412?

I have connected to the web server java applet to see high peaks of port usages but none of the ports gone over 50% while I simulated and tested network overload via multicast. What level of port usage would be suspective?

I also made some wireshark captures if anyone is interested... :o)

Regards, Jochen

[Gerhard Roets]

Hi Jochen

I hope you are well. What software version is the 5412 running. I assume it does the routing in your network.

Did you configure the Microsoft NLB with or without IGMP support on the server ?

Do not turn the above one on just yet. Lets see what you have first :).

Gerhard

[Jochen Müller]

Hi Gerhard,

we have ProCurve Switch 5412zl (J8698A) Version: K.12.02, ROM K.11.03

What exactly do you mean with "IGMP support on the server"?

Thank you and best regards,

Jochen

[Gerhard Roets]

Hi Jochen

There is a tickbox when you enable NLB with Multicast for IGMP. You can see the server virtual MAC changing from something like 03-BF-... to 01-00-5E-...

K.14 routes NLB with Multicast in hardware which is propably what you want(gut feel).

Now if you want to use IGMP mode see below.

My suspicion is in the long run you might need to upgrade to K.14.65

You might want to look at upgrading the switch to K.14.65. There is a feature there which will help.

The software can be found here http://h10144.www1.hp.com/customercare/support/software/summarypages/k-j8692-c.htm

You will be looking for the following in the release notes released in K.14.55
â Enhancement (PR_0000018427) â  Multicast ARP support enhancement.

ProCurve(config)# ip arp-mcast-replies

This should help you just have a look at the release notes.

KR
Gerhard