Architecting network with 10 Mbps half duplex phone system

As Wim noted already, the half-duplex doesn't matter. The switches are capable of running each port at different speeds and duplex.

The easiest way to evaluate it is to think of the path from component to component, and taking the weakest link. All your communication between building #2 and headquarters happens via a 3Mbps/300kpbs line shared with other internet duties; whether you have a 10Mbps or 100Mbps connection back at headquarters won't matter because the VPN link will be the dominant factor in determining the interconnection bandwidth.

Looking at your diagram, the material change I see in your proposal is introducing a 1Gbps link between the two HP 2524 switches at headquarters. One one switch you have a bunch of servers, each limited to 100Mbps, and on the other you have a bunch of client workstations, also limited to 100Mbps or 54Mpbs on wireless. No one machine here is going to be capable of exhausting the data link between the two switches, but with heavy traffic on multiple machines between the clients and the servers, you'll appreciate the 1Gbps link.

Are you trying to solve a theoretical design problem or are you having an actual VoIP call quality problem?

Any of the switches should be able to handle a mix of port speeds (10/100/1000) and duplexes (half/full). That in itself shouldn't be a problem.

I would let the RV082 be only the router, with a single LAN cable going to your HP switch. Let the router be a router, and the switch a switch...

It is a bit silly that the Talkswitch only does 10 HD. But still, an uncompressed ulaw/alaw VoIP call takes max 100 kbps so you can run lots of concurrent calls no problem.

You could/should explore the QoS features of your router and switches a bit more.

This might help: http://www.hp.com/rnd/pdf_html/traffic_profiles.htm#environment2 but you may have to hunt a bit more for correct doc for your switch.

I've run specialized, critical telephony systems for stock traders at 10 Mbit, half-duplex, and I haven't seen issues with that aspect of the system. (The IPC turrets operated only at 10/half.)

Have you tested call quality with an IP telephone directly connected to the Procurves at the HQ? That should eliminate the switching gear as a possible culprit.

I'm sensing also that you haven't got any sort of monitoring system in place to evaluate bandwidth usage on those RV082's. If there isn't an easy way to check bandwidth usage using the web admin console for those routers, consider implementing a performance monitoring system. Another quick Google shows that those routers support SNMP. If you can find a spare computer upon which to throw Cacti or PRTG, that should go a long way towards determining the saturation level of the Internet link. (Throw Nagios on there too, and you've got an availability monitor for your network.)

Without hard data, I suspect it's the Internet link at Building #3 that is constraining your bandwidth. So it's worth talking to your provider to understand your options and costs for upgrading. But before buying an upgrade, I would advise confirming the problem through performance monitoring. Understand how much more you need to buy before buying it.

Also, what codec are those IP phones using? I'm not familiar with Talkswitch, but a quick Google shows they support G.711 or G.729. If they are using G.711, which uses 80 kbits bandwidth, you'll be able to run at most 3 calls across that Internet link at the remote site. G.729 will cut telephony bandwidth usage by an order of magnitude. The call quality will suffer, so make sure your management is on board before you make that change. But it might be helpful to do that in the short term if evaluation of bandwidth usage ends up taking longer.

HTH!