Hi everyone,
In 1 month, I’m re-organising the network for a lan party with one of our non-profit .
I’m looking at the following design :
1 Forbidden Routeur (Either OPNSense VM or dedicated host) to a 10G feed (Yeah, happy Switzerland )
1 Arista 7050sx as Core, multiple VLANs to the users switches
5 user switches (Cisco 2960X/Juniper EX3300) with 10G uplink
Some other switches for alternative ressources (webcaster, stage, etc)
I’ll plan the following logic :
1 Management lan with the IPs of the switch and servers
1 public lan (wifi and user ports connected) with DHCP and DHCP helper on the user switches to avoid rogue DHCP servers (i plan to probes for that here and there)
IGMP Snooping, to avoid multicast madness
1 adguard/pihole for ad blocking / DNS caching
1 Steam/game cache with SSDs and 2x10G connection to the lan.
(Edit : 1 big broadcast domain for users, a /24 is enough, I’ll have 1 more broadcast domain for the management with another /24, nothing crazy, minimal routing between the 2 lans).
My questions :
is IGMP snooping in a broadcast domain still important/making sense or are some games broken by that ?
Same questions for the DNS filtering , any risks ?
Anything I might have missed since the last lan I organized 25+years ago with terrible un-managed switches and a 2Mbps uplink to the net ?
Don’t. False positives are way too high. Ad blocking is not your responsibility. DNS caching does not really matter nowadays.
Multicast? Where do you expect multicast? I don’t know much about webcasters and stuff, but gaming does not use multicast AFAIK.
Maybe think about some per VLAN and/or per client limits.
Or make a TCP pipe set to 9GB/s. That way there is always enough bandwidth for UDP gaming.
Not sure what is the current status. 10y ago, we registered a small internet cafè to get steam cafè cache server. After that, we got fiber. In recent years, the clients themselves act as cache servers. Problem is that you can’t really configure anything, last time I checked. So there might be two people fighting for the 1GBit/s bandwidth, because they both transfer from PC1, despite there being others that could offer the files.
Anyway, good luck and have fun!
Ohh, and get 25GBit/s from init7. They also have a steam cache server in their core
I sold my 25Gb optic ages ago (I was one of the 1st in Geneva, and I even migrated my own port when they changed hardware at the time ) . Because the service is going to be used by others after us, 10Gbps will be simpler for “ those who comes after “ .
I’ll try a cache just for the experiment then, we’ll see what it shows.
Overall a good build, not sure steam caching is possible anymore with https being the default.
Before going deeper, the main sizing questions are:
How many players/clients (wired + Wi-Fi)?
How many days, and what are the peak hours?
What’s the incoming Internet speed and expected external traffic (updates, streaming, webcaster)?
Having only one of anything is usually the weak point: one router/firewall, one DNS resolver, one cache server, one core switch, one uplink. For a non-profit event, at least have a cold spare or a documented fallback for the router, DNS, and core switch.
about 100. We don’t expect that much Wifi users, the orga will push them to cable themselves. We will plan to have a more APs than needed (4 UAP-HD-PRO or alike), that will be overkill by all means.
Friday evening to sunday late afternoon, I don’t have the peak planning yet.
10Gbps via Init7. I can reserve some bandwidth for casters and streaming. I suspect the peak traffic will be friday during initial connection, if no updates occurs in the week end. (maybe there will be linux iso downloading happening also, but users only have a 1G port).
Good point, I was wondering if I load everything (I don’t run the game servers) on one proxmox, so having it’s “ clone “ aside is a simple one. I was considering the lancache as a side machine for the experiment.
Regarding caching, I’ll check that unbound is configured correctly, otherwise I’ll setup a bind somewhere, that’s quite a one liner for me.
Regarding the core, I’ll take a spare, that’s also an easy one, and we will get spare users switches also.
Makes sense. For ~100 players, I’d still size the network around the Friday-evening update rush, not the steady-state load. Even with 10 Gbps from Init7, the first few hours can get eaten by Steam/launcher updates, Windows updates, and reinstalls. A LANCache box is definitely worth testing, Steam especially.
A few things I’d do:
Pre-warm the cache before doors open with the games/launchers likely to be popular.
Give the cache more than 1 Gbps if you can. If clients are on 1 Gbps ports, a 1 Gbps cache NIC becomes the bottleneck once a few people download at the same time. If you have a spare 10gb card use it here and connect to your core switch.
Use Unbound for LAN DNS, make sure DHCP points clients at it, and enable prefetching with a decent cache size.
Keep the LANCache DNS redirect separate from normal DNS. If the cache VM/container dies, clients should fall back to normal DNS instead of losing resolution.
Watch uplink usage, cache hit/miss, DNS load, AP client counts, and interface errors during the event. That’ll show whether Friday is cache-bound, DNS-bound, Wi-Fi-bound, or actually internet-bound. If you can setup something like grafana so you can see real time the data.
Reserve caster/streaming bandwidth with QoS or separate VLANs/shaping rather than assuming there’ll be enough.
For redundancy, Proxmox plus a cloned/cold config is fine for DNS, DHCP, monitoring, and the cache, as long as you’ve got documented restore steps and the spare core switch. I wouldn’t make the cache a hard dependency though; if it dies, just fall back to direct internet access.
With 10 Gbps upstream, spare core hardware, over-provisioned APs, and a tested cache/DNS setup, you should be in good shape for 100 people.
Yup, if I have enough hardware, I was planning to do 2x10G LACP
What do you think ? like my recursive bind points to the lancache dns ? or the other way around ? Or setup a LB with main on lancache and fallback to recursive bind ?
I’m a librenms person, and those old switches don’t do streaming telemetry well. But beside that the opnsense monitor is one thing and i plan to use ntop-ng.
Yeah I think a isolated vlan will be good enough, w/QOS
Yup, my thinking. Looking at the shape of my users traffic when I was serving as an ISP, (up to a year ago), I know what is really used, I know I’ll be higher in the lan party, but hope i can mitigate that with the cache.
Why? Seriously, this isn’t 2003 anymore.
Just use init7 DNS and don’t bother with it.
Even if you don’t cache a single DNS request locally, you would still be fine.
For regulation and control, split dns functions, and speeds. Local cache is still faster than distant lookups (if setup right). It’s not for the bandwidth saving anymore - so no, it’s not 2002 :).
I would be head of DNS with the infrastructure - diverting non-sane lookups and filtering is an easy task to ensure at least botless-ish network, added security. Having the upstream default to 1.1.1.2 for example.
split dns needs you to have control over the dns, so yes - I see a LAN as an networking experiment - so testing out different options and providing services (heck, throw in a proxy function as well!)
And faster is faster when it comes to cache!
While users of the network does their thing, you as a network admin gets experience that is still relevant in functions that in my opinion should be in place for insights on “easy layers”.