Clean Up Network Rack Help

ChuckMountain

How are you finding the Unifi USG Pro 4 and switch in terms of both use and noise?

I am just in the process of redoing some of my network and I need to add a couple more network devices in.  I need around 35 ports or so which would suggest that I need a 48port so was looking a the Unifi one as well as upgrading my router.

Originally I was going to segment everything onto multiple VLANs as I can do with my original equipment.  However stuff that is controlled by apps on my phone on the local network needs to be able to reached in some cases from the phone rather than just via the net.  I found it quite cumbersome in some cases to get this to work so reverted to two VLANs with 3 SSIDs, one core, one infra and one guest.

Should really post pictures up of mine but its in an in between state at the moment.

The other challenge I have is that just upgraded to Virgin Gig1 and whilst the service will apparently hit 1.1Gbps, in reality, it is not there yet and limited to 1Gbps per port which will be around 930Mbps line speed [Nice problem to have ] . So if I have it in modem only mode then I lose 15% or so so wonder if can dual wan it with regular hub mode to get the full bandwidth in my network.

mickevh

It might be worth checking whether you router has the grunt to route the required throughput and see if you are loosing it there. Most decent kit cites the routing throughput in the datasheet.

ChuckMountain

Thanks, the router an Linksys LRT224 has supposed to 900Mbps throughput which I know isn’t quite up there but should be ok.  Probably will find it is a lower number now . I disabled all the firewall and additional stuff but it made very little difference.

I did have a mess about last night and did find the following. (Tests with speedtest.net to TNP Manchester, not necessarily the best but trying to be consistent)

1) My cabling either CAT5e or CAT6 did not make a difference, which when working correctly would be expected however given a few recent posts and swapping cables I double check
2) My main managed switch is ok
3) The Linksys is definitely a bottleneck at maxes at around 700Mbps ish depending on the client
4) Main desktop client struggles when connected directly to Super Hub 4 to get past 750Mbps (tried jumbo frames, safe mode, firewall off). It is not CPU bound and sits at around 25% individual core usage (i7 5820k @ 4.4GHz).  The Intel Ethernet builtin NIC would seem to be the limiting factor but it can quite happily do iPerf at 930Mbps between that and main server
5) A new Lenovo P52 Laptop with an i7 maxes out at 550Mbps when corrected directly to SH4, this is supposed to be a dev machine but that was tried on battery though Energy Saver set to Max Power.  Need to try again with mains power.
6) A MacBook Pro via Thunderbolt into a dock will run at around 930Mbps when connected directly to the Hub

In conclusion it would seem that speed tests do max out some of my NICs quite easily and that requires further investigation but it would appear that the SH4 is working at around the advertised speed.

The question I have now is how I go and obtain that in my network as to make the most of it I need to have two gigabit ports connected to the SH4 hence wondering if the USG4 would have the right throughput to supply it when using link aggregation back to main switch.

mickevh

Link Aggregation normally won't be the magic wand: The rules of LA's mean that they must not introduce out of order packet delivery across the LA channel and almost invariable that means all packets for any given pair of peers goes down the same physical link in the LA. IE - it doesn't A/B the traffic across the available physical links to "boost" throughput. So for any given single pair of endstations, it's exactly the same as a none LA link. Where you get the performance boost with LA, is in servicing multiple pairs concurrently up to the number of physical links in the LA channel. (and some fault tolerance.)

One way you might check out the NIC's and switches locally would be to test them with iPerf or NETIO which takes the routing engine and ISP link out the equation. That'll tend to prove/disprove whether your router and/or ISP link is culpable and give you some "baseline" for the performance of all your local stuff (laptops, Macs, switches, etc.)

One shouldn't really use JF's unless your entire LAN can support them - it potentially gives issues with large broadcast packets that are unable to be read by non-JF clients and infrastructure. It could also potentially make you ISP link worse if the router has got to fragment the packets into smaller ones for onward transmission up the ISP link.

When sizing router routing performance, we tend to expect real world to yield less that the stated performance as the spec's cited are something or a "nominal" figure - there's a lot of real world variance due to packet size mix, traffic mixes, interface types, yada, yada. When sizing a router throughput, if say, we need throughput X we'd want to look for a router with a healthy margin over and above what's required to give us a bit of headroom.

ChuckMountain

Yes get it re LA however it isn't it a case of wanting a single faster overall connection but wanting to be able to use the 1.1Gbps that I get on my Internet connection across multiple clients.

I am trying to work out the best way of doing that whilst still being able to provision at least two VLANs and not have multiple smaller switches which then means LAN traffic goes through the Super Hub 4.

On testing Speedtest seems to max out something as iPerf is fine

JFs are enabled on my switches to allow amongst other Sky Q\Mini to work which is\was a known issue on managed switches.

Kristian

If you think those cabinets need tidying up then I hope you never visit any of my sites .  There's nothing wrong with your cabs.  Spend the time doing something more enjoyable...

mickevh

Is your SH4 VLAN capable..? If not I guess you'll need "something else" to route between the VLAN's (if you have need for an inter-VLAN traffic) and/or the ISP link.

It's years since I did it, but I never had any trouble running VLAN's over LA's. From memory the way one enacts it varies amongst vendors: Either one creates the LA from the physical links and gets a kind of "virtual" interface (swicth port) each end, then bind the VLAN's to such "virtual" LA endpoints, or one does it "the other way around" and binds the VLAN's to the physical ports ("trunk" ports) of the soon-to-be LA, then bind the physical interfaces into the LA. IIRC some kit essentially "ignores" the "secondary" ports when you bind up an LA and clones the VLAN participation (and trunk/hybrid/simple) state from the "first" port in the group however that may be defined. I guess some RTM is required.

On my trunked (VLAN carrying) links, LA or not, I prefer not to have any "untagged" traffic so all traffic across the trunk bears a VLAN tag, but I know some kit insists on there being an untagged VLAN bound to a trunk (such as PVID.) I used to create a "dummy" VLAN (a "black hole") VLAN for the untagged traffic so in effect it all got dropped and I only had to onward direct tagged traffic. Thusly I can always be 100% certain I never got any traffic on the "wrong" VLAN: at the ingress on any trunked port into any switch (or router) - if traffic is tagged, I know where it came from, if it's untagged, I shouldn't be receiving it in the first place so I just drop it.

I've never had my hands on one, but I rather like the concept in a lot of modern (enterprise) routers that they can support VLAN's so one no longer needs a separate physical interface for each VLAN one wants to route between. One just creates a whacking great trunked LA with sufficient physical links to avail redundancy and capacity, then in the "software" of the router, break out the physical link into multiple "virtual" interfaces, one for each subnet, then route/firewall/ACL etc between the vitrual interfaces (subnets) as one did traditionally.

I used to do something similar in a laptop for testing/diagnostics - a lot of OS's and endstation NIC's are now VLAN aware/capable. So my (Windows) laptop would have a load of virtual NIC's bound to the physical NIC - one for each subnet I might be interested in, (without any routing between them,) using static IP addresses so I don't have to "worry" about DHCP complicating things, then I used to create a "test/diagnostic" port on my switch carrying all the VLAN's I might be interested in. Whence fiddling with things or problem solving, I connect up my lappy to this "diagnostic" port, thence open up loads of CMD windows and continuous ping things on each subnet so I can keep an eye on it and make sure I've not killed the network whilst I'm fiddling with the plumbing (physical or config) elsewhere. Perhaps you might find that a useful tool.

ChuckMountain

It currently works with a couple of vlans and routing via the linksys.  That is now the bottleneck so does need an upgrade.  Don’t think the SH4 supports VLANs and certainly there is no way of specifying them.

The only current LA I have is between main switch and server at mo.

mickevh

Yeah - I guess no-one is surprised that's the case in a cheap ISP router. Some of the after market and pro-sumer kit from the likes of ASUS, Draytek, et al maybe does, but it's something one would be "shopping for" if one needed it rather that "just assuming" it'd be there.