Register a SA Forums Account here!
JOINING THE SA FORUMS WILL REMOVE THIS BIG AD, THE ANNOYING UNDERLINED ADS, AND STUPID INTERSTITIAL ADS!!!

You can: log in, read the tech support FAQ, or request your lost password. This dumb message (and those ads) will appear on every screen until you register! Get rid of this crap by registering your own SA Forums Account and joining roughly 150,000 Goons, for the one-time price of $9.95! We charge money because it costs us money per month for bills, and since we don't believe in showing ads to our users, we try to make the money back through forum registrations.
 
  • Post
  • Reply
BlankSystemDaemon
Mar 13, 2009



I wish I had the money to play around with TP-Links Omada series of routers, switches, and access points - I keep hearing good things about it, to the point that it seems to be what Ubiquiti used to be.

Case in point, it's centrally managed by a piece of software that can be hosted on any Unix-like (because it's Java :sigh:) and the hardware is cheap yet lets you do a ton of stuff with it.

Adbot
ADBOT LOVES YOU

BlankSystemDaemon
Mar 13, 2009



Counterpoint: If you have devices that benefit from anything more than 1000BaseT, you'll probably benefit more from doing 10GBaseSR using OM3 as it has lower latency, uses less power, and is not going to be impacted by other people setting up their own BSSIDs and loving with your airtime.

If I understand things right, Wifi 6+ (and LTE(A), and most other modern standards) have time division which is made to work in environments where different beacons/cells can talk together over the wired connection to negotiate airtime - to ensure maximum coverage based on how many clients each device handle while minimizing the hidden node problem.
If you've got a lot of neighbours with a similar technology but your networks aren't connected together, you don't benefit from this and with enough neighbours, there'll be no airtime.

BlankSystemDaemon
Mar 13, 2009



Bits or bytes per second, orders of magnitude thereof, don't matter when there's no airtime.

Video generally is measured in bits per second, not bytes - and that's true for any bandwidth measurement.

BlankSystemDaemon
Mar 13, 2009



Because I think people should have realistic expectations about what wireless speeds they can get, I'm gonna keep harping on about airtime being the determining factor for what bandwidth you get - because what the specs say has nothing to do with reality which is where base stations usually exist.

Also, DFS is very explicitly a passive state, in that it starts listening for certain signals like radar, which governments (who own and regulate the airwaves) have reserved for their primary (though no longer exclusive) use.
If a radar or similar device is active, it'll pick another channel where everyone else has moved onto, and there'll be even less airtime - because it always comes back to airtime, when you're dealing with time division multiplexing.

BlankSystemDaemon fucked around with this message at 18:20 on Sep 12, 2022

BlankSystemDaemon
Mar 13, 2009



Eletriarnation posted:

You are correct, but I think I am missing your point. What/who are you arguing against here? As far as I see posters have been clear that you can't ever expect to get close to the full rated speed of a base station on a single endpoint, and even getting the performance you should expect can depend on a number of assumptions about the environmment.

e: In case it wasn't clear, I'm not saying that everyone should use 160MHz channels even if they're in an apartment block and drat the consequences. I'm saying that if you can use 160MHz channels without interference from sources you don't control, you would expect to see a substantial performance gain vs. 80MHz.
There is talk about the speeds promised in advertising/specifications - that's what I'm trying to argue against.
The people who make the specifications, gear, and anything else haven't achieved those speeds and neither will anyone else, unless they do a stunt like setting up two uni-directional antennas pointing straight at each other with nothing in the Fresnel zone.

Unless you live far from anyone else, you realistically can't use a 160MHz channel.
The best wifi I ever set up was achieved from having one unit in each room, all configured to just have enough signal strength to not pass through the walls, and not having to deal with devices that can't roam properly.

BlankSystemDaemon fucked around with this message at 20:42 on Sep 12, 2022

BlankSystemDaemon
Mar 13, 2009



Wibla posted:

This is good advice.

As for fibre, I would go with SM, but I'm professionally challenged in this area (I maintain a large metro-area network and we have banned MM fibre in our entire network for maintenance reasons). fs.com is your friend for cheap optics.
Interesting - when I maintained a campus-area network, what'd most often break would be the Lucent connectors.
It got so bad, that we seriously talked about using standard connectors - because they're intended to be used by providers who need to move the connections around with some regularity and we found that Lucent connectors are only rated for around a thousand disconnects.

BlankSystemDaemon
Mar 13, 2009



The software, specifically their built-in 802.11 signal simulation, is also an entirely reasonable argument for buying it, if you've got a machine to self-host it on (it uses java and mongodb, so it can run basically anywhere including an RPI and comes packaged most places).

The signal simulation lets you use a building schematics, the measurements on it, and a built-in placement tool to input devices, walls (including material, density and thickness), windows, doors, and everything else.
This gives anyone a good hint about where it might be an idea to put another access point, when you adjust the one(s) you have to not send more than the devices can send, so that you instead make use of roaming and strong signal strength everywhere.

BlankSystemDaemon fucked around with this message at 18:57 on Oct 12, 2022

BlankSystemDaemon
Mar 13, 2009



Pihole is just a project that integrates a WebUI and combined DHCP daemon and recursive caching name server daemon called dnsmasq.
You either configure your devices to use it to look up things via resolv.conf (or the equivalent on other OS), or get clients up for you by relying on the DHCP functionality.

It looks to me like it's not got the right permissions in /etc/pihole/, so that'd be the first thing to fix.

BlankSystemDaemon
Mar 13, 2009



It's kinda impressive how many Octeon CPUs that Cavium managed to get onto the market, before they got acquired by MicroSemi, for there to still be stock left.

The ASIC they use to fastpath network traffic (unless you're doing DPI, if memory serves) is plenty fast for doing 1Mpps of 64 byte Ethernet packets when I used it.
If you're doing DPI, set up a divert port to a more powerful machine running suricata.

BlankSystemDaemon
Mar 13, 2009



Head Bee Guy posted:

What’s a good linux program for viewing/analyzing web traffic from CLI?
tshark

BlankSystemDaemon
Mar 13, 2009



Stuff that can test even up to cat6a is insanely cheap, yeah.

I think I bought the one I have now for $42 at some point, and I'd be surprised if they're that expensive now.

BlankSystemDaemon
Mar 13, 2009



CopperHound posted:

Lol what? You have a $42 analyzer that actually measures cable performance instead of a basic continuity/miswire check?

For most people a wiremap check is good enough, but 6a implies 10gig speeds which will not tolerate sketchy splices or wires pinched between the stud and sheetrock or whatever was happening under this defective cable that I stripped back to troubleshoot:

It does continuity, miswire, and some kind of flow test but the latter is one I've never been completely sure about.
As for 10G, I use SFP+ because anything that's 10G in my network is also going to benefit from the order of magnitude lower latency that SFP+ has over RJ45.

I still maintain that the only proper way to do a speedtest is using iperf (preferably v2, as v3 can't do multithreading properly).

BlankSystemDaemon
Mar 13, 2009



MadFriarAvelyn posted:

So my apartment has ethernet wired through the walls, but whoever setup the patch panel only wired one ethernet port in each room instead of both. This throws a wrench in my networking plans. I've been dealing with this for almost a year now and want to get my switch wired back up so I can get every device in my apartment that has an ethernet port plugged into wired ethernet.

How hard and what tools would be required for a complete novice to finish setting up the rest of the ports? Or is this a job better suited for a call to my ISP to get a contractor to come out?

[Edit] Or should I just be a degenerate and buy a switch for the networking closet and each room and deal with it that way?
A distribution switch with PoE-out on most of its ports and a couple of edge switches with PoE-in is absolutely a way to solve this.
It'll also give you a really easy way to deploy multiple access points to make use of roaming, which makes wireless function about as well as it can.

BlankSystemDaemon
Mar 13, 2009



KS posted:

I sure wish this would die because no it will not benefit. That latency is measured in microseconds and it's a really really dumb reason to go with fiber at home in a small network. Weigh cost, power efficiency, and a need for mgig or poe support.
If you're doing iSCSI over 10G SFP+ to NVMe backed storage, the latency makes a difference.

Also, used X520 NICs can be had for insanely cheap in most places.

BlankSystemDaemon
Mar 13, 2009



KS posted:

It does not make a difference. The difference in 10gbase-T SFP+ vs fiber SFP+ is ~2µs. Here are links in that chain that dwarf switch latency:

The NVME drives themselves. 20µs for bleeding edge enterprise, 100µs+ for the very best consumer.
Kernel networking. 40µs typical.
NIC choice. Chelsio cards hold double-digit µs leads over those x520s.

Framing recommendations in this thread around home lab stuff seems pretty questionable in any case.
I'm talking about 10GBaseSR via SFP+ vs 10GBaseT via RJ45/8P8C not whatever "10gbase-T SFP+ vs fiber SFP+" means to you (because I'm not sure what it means).

SFP+ has a latency of about 0.3μs whereas RJ45 with 8P8C has a latency of about 2.5μs.
If the SCSI READ command of the iSCSI initiator returns something that was cached by the system that acts as the iSCSI target, it doesn't matter how fast the underlying storage is, since memory accesss happens on the order of 100ns.

I don't know what kernel you think takes 40μs to handle any networking request, but it certainly isn't FreeBSD which is what I'm using.

zpool iostat -w and -l showed a clear difference when I was testing by booting and using my buildserver off an iSCSI target.
Unfortunately I don't have the numbers anymore, but here's an example of the latency distribution from my T480s running FreeBSD 14-CURRENT as of a few days ago:
pre:
zroot        total_wait     disk_wait    syncq_wait    asyncq_wait
latency      read  write   read  write   read  write   read  write  scrub   trim  rebuild
----------  -----  -----  -----  -----  -----  -----  -----  -----  -----  -----  -----
1ns             0      0      0      0      0      0      0      0      0      0      0
3ns             0      0      0      0      0      0      0      0      0      0      0
7ns             0      0      0      0      0      0      0      0      0      0      0
15ns            0      0      0      0      0      0      0      0      0      0      0
31ns            0      0      0      0      0      0      0      0      0      0      0
63ns            0      0      0      0      0      0      0      0      0      0      0
127ns           0      0      0      0      0      0      0      0      0      0      0
255ns           0      0      0      0  2.54K  2.95K    469     59      1      2      0
511ns           0      0      0      0   415K  68.0K   186K  31.1K      7    875      0
1us             0      0      0      0   876K   100K   172K  89.7K      2  6.32K      0
2us             0      0      0      0  1.09M   240K  39.2K   166K      0  25.4K      0
4us             0      0      0      0   269K   138K  13.0K   350K      1  48.2K      0
8us             0      0      0      0  9.51K  59.7K  1.70K   129K      1  81.4K      0
16us            0      0      0      0  1.17K    870  1.60K  45.3K      2    688      0
32us        14.4K      0  15.3K      0    692    548  2.39K  88.8K      7    230      0
65us        37.6K  67.5K  39.1K   537K    235    195  3.72K   190K     30     11      0
131us       1.78M   367K  1.85M  1.28M    260    313  8.07K   341K     34      1      0
262us        901K   903K   858K  2.70M    301    676  19.5K   665K     23      3      0
524us        310K  1.42M   334K  1.56M    122  2.71K  22.3K  1019K     17      2      0
1ms         63.5K  1.35M  34.5K   168K     10  7.09K  15.5K   987K     13      2      0
2ms         26.7K  1.23M  11.0K   125K    196  1.57K  6.97K  1.07M     16  28.4K      0
4ms         6.98K   816K  4.36K   248K     13    456  1.45K   516K      1  67.1K      0
8ms         1.91K   319K  1.00K  28.0K     18    297  1.19K   265K      0   135K      0
16ms        1.01K   148K     10    441      1    174    940   125K      0   354K      0
33ms          237  51.3K      4     57      0    140    193  43.8K      0     52      0
67ms           97  9.95K      2     26      0    251     69  8.90K      0      8      0
134ms           0  1.59K      0      6      0    305      0  1.20K      0      0      0
268ms           0    122      0      1      0      0      0    119      0      0      0
536ms           0      0      0      0      0      0      0      0      0      0      0
1s              0      0      0      0      0      0      0      0      0      0      0
2s              0      0      0      0      0      0      0      0      0      0      0
4s              0      0      0      0      0      0      0      0      0      0      0
8s              0      0      0      0      0      0      0      0      0      0      0
17s             0      0      0      0      0      0      0      0      0      0      0
34s             0      0      0      0      0      0      0      0      0      0      0
68s             0      0      0      0      0      0      0      0      0      0      0
137s            0      0      0      0      0      0      0      0      0      0      0
---------------------------------------------------------------------------------------
Even with local NVMe storage, there are absolutely disk commands that happen in the time resolution we're talking about - specifically, the TRIM commands.

BlankSystemDaemon fucked around with this message at 21:21 on Nov 24, 2022

BlankSystemDaemon
Mar 13, 2009



Inept posted:

I think the omada routers also don't have stateful firewalls
Wait what? How in the gently caress do they get away with that?

BlankSystemDaemon
Mar 13, 2009



Cyks posted:

It’s not true.
That makes more sense - because I'm not sure how it'd even be possible.

BlankSystemDaemon
Mar 13, 2009



I had to do a triple-take.

BlankSystemDaemon
Mar 13, 2009



Rexxed posted:

I was able to setup a guest network with a ubiquiti cloud key for one of my clients. I don't know if it requires the server software to be running all of the time on something but you could consider a cloud key or putting it on a VM or something if you needed it.
The controller, which is really just a Java applet and a MongoDB instance packaged together, and can be found in many Unix-likes third-party software repositories, is only needed for things that require AAA/RADIUS such as WPA2 Enterprise, Guest hotspot with TOTP, and things like that.

BlankSystemDaemon
Mar 13, 2009



Yeah, I laughed at that too.

BlankSystemDaemon
Mar 13, 2009



You can also use headscale to host your own orchestration software that tailscale connects to.

BlankSystemDaemon
Mar 13, 2009



SwissArmyDruid posted:

My Edgerouters seem to finally be dying, so they're getting replaced.

Didn't PFSense say they were migrating to a Linux kernel at some point or was it a different router software?
The company behind pfSense also produce TNSR.
TNSR is a DPDK-based appliance, so it's not really using the Linux kernel for the networking part and thereby doesn't have to suffer its inferior performance when it comes to networking.

e.pilot posted:

pfsense did some shady poo poo with wireguard, I’d go opnsense
For what it's worth, WireGuard is back in FreeBSDs base system again, and in addition to being a downstream product from Jasons repo, it can take advantage of crypto(9).

BlankSystemDaemon
Mar 13, 2009



pfSense/OPNsense is an appliance - it's just an appliance OS.

The alternative to an appliance OS, in this case, would be FreeBSD with pf configured - but that also gives you the option of using ipfw, which pfSense/OPNsense doesn't.
As for hardware that'll run FreeBSD, a AMD64/x86_64 or Aarch64 (not on a SBC, though) with an Intel/Chelsio/Mellanox NIC is probably the best choice.

BlankSystemDaemon
Mar 13, 2009



devmd01 posted:

This is the way, one per floor isn’t overkill at all.


If money is no object, the best wireless experience is one AP per room, from a manufacturer with a good implementation of 802.11r, tuned to ~just enough~ transmit power.

MarcusSA posted:

So if these are the networks around me is there any reason not to use auto for the channel selection?


You'll want to use a tool like kismet to see what channel they're on. Then you can attempt to find a non-overlapping channel that isn't being used.

BlankSystemDaemon
Mar 13, 2009



Well, kismet uses either a TUI or webUI - so if someone's semi-comfortable on a Unix-like, it can be made to work on Windows.

For macOS, the fork is called KisMac.

BlankSystemDaemon
Mar 13, 2009



cr0y posted:

Fiber doesn't really have "modems", the box is called an ONT and it converts fiber to coax or ethernet, from there you plug it into your router or a router that the ISP gives you.

To be clear the fiber will come into your house and all of the boxes will be inside somewhere.
There's a big difference between whether you have *PON fiber, or whether it's just Ethernet 10GBase-(L|P|E|Z)R with OS2+, in so far as how much work the CPE at the DMARC is going to be doing and how easy it is to terminate on your own.

Although with the recent spat of *PON-capable SFP(+) modules, it's become a lot easier.

BlankSystemDaemon fucked around with this message at 10:49 on Mar 22, 2023

BlankSystemDaemon
Mar 13, 2009



For Ethernet using TCP, linerate is ~116MBps, for UDP it's 125MBps, assuming MTU is 1500 bytes.

BlankSystemDaemon
Mar 13, 2009



Traceroute isn't really a useful diagnostic tool anymore, since almost every router on the internet will de-prioritize the ICMP echo/replies with decrementing TTLs, going from attempting to not answer it as fast as possible (or at all) all the way up to sending the traffic via an entirely different route - both of which result in different results packet switched networks compared with normal traffic

This can sometimes be worked around by using TCP, but very few implementations of traceroute support doing that because it requires keeping track of RSTs.
This, of course, requires that the routers are configured to send RSTs on a closed port, instead of simply not responding - which is the better option, and is usually the default on OS' that implement half-open connections, which not every OS does, and routers don't tend to like to accept being the destination for.

BlankSystemDaemon
Mar 13, 2009



wolrah posted:

What's everyone with stupid fast home fiber doing hardware-wise these days?

I've been using pfSense for years on a Netgate SG2440 which is fine for up to gigabit speeds, but now I've moved to a neighborhood which has AT&T's 2/2 and 5/5 service available. I got the 2/2 for now and probably won't upgrade any time soon, but for the sake of futureproofing I'd really like to build something that could at least handle a full 5/5 if I ever do choose to make the jump. Unfortunately that knocks out all of the cheap quad 2.5G Atom boxes that are everywhere on the internet.

I have a strong preference for x86 hardware just due to the number of choices on the software side, but I'm open to anything if it makes sense.

Is there any appliance-style hardware in this range worth looking at, or should I just pick up a few NICs and find some compact PC hardware to stick it in? Or maybe just stick it in my server and virtualize the whole thing?
If you wanna stick with something FreeBSD-based, it can do in excess of 20Gbps bidirectional stateful firewalling.

BlankSystemDaemon
Mar 13, 2009



That Works posted:

I’m not familiar. What’s that one?
It was the original FreeBSD based firewall appliance, but it used ipfilter by Darren Reed instead of ipfw or pf.
It was quite popular on Soekris hardware.

EDIT: The way it saved its configuration to a separate filesystem in /cfg was also quite revolutionary for its time, and is the reason why things like NanoBSD and TrueNAS is partitioned the way it is, to this day.

BlankSystemDaemon fucked around with this message at 01:19 on Apr 1, 2023

BlankSystemDaemon
Mar 13, 2009



priznat posted:

What’d be a good router option to connect to a fibre pon modem if it is 10G SFP? This was one of the main reasons I was eyeing a udm se.

I am assuming there are aliexpress options that have 10G instead of the 2.5G ports?

My area is getting fibre rolled out and would like to move over from cablemodem at some point.
When you say router, I assume you mean you're looking to do both stateful and stateless firewalling at 10G?

Forwarding at 10G isn't difficult, firewalling (especially statefully) can be - so I'd recommend going with commodity hardware and a 10G SFP+ PON adapter from FiberStore, then installing BSDRP on it.
It's based on FreeBSD -CURRENT so has all the speed-ups that the pf firewall has gotten (which makes it about as fast as ipfw is), and has everything you should need.

BlankSystemDaemon
Mar 13, 2009



Just remember that an iperf test (no matter if it's version 2 or 3, though they're different) doesn't measure the firewalling speed of the hardware, which is what matters.

BlankSystemDaemon
Mar 13, 2009



Speaking LTE networks, even if they're for backup - I wanna get a hold of a LTE-A EM160R-GL.

It does LTE-A Cat 16, meaning it's supposed to be capable of the full 1Gbps/150Mbps uplink that's advertised for stationary devices.

BlankSystemDaemon
Mar 13, 2009



unknown posted:

IIRC, for unifi, they only channel scan on startup since they don't have a secondary radio. (or if you hit the scan button, but that shuts down normal usage for a minute). Maybe their higher end gear has the bonus radio for scanning though.
What the gently caress does this even mean.

I refuse to believe that you can't use multiple device nodes with different operating modes, so that one is a hostap and another is station and a third is monitor.

BlankSystemDaemon
Mar 13, 2009



It probably also needs to be said that if you set things up properly with pf, pfsync, and carp, a firewall isn't the single point of failure that a lot of people think it is.
It is also entirely doable on off-the-shelf consumer hardware for a multi-gigabit FTTH connections.

If the internet is as important as it seems like it is for them, it's worth finding an IT guy that can ensure it stays up.

chocolateTHUNDER posted:

ISP Modem/ONT > Firewall > Switch > WAP
CPE - Firewall - Router - Switch - WAP.
Even if you're doing everything on a pair of boxen using the above method using a pair of GPON SFP+ modules, you still need to conceptualize it like that - otherwise, you risk configuring it wrong.

BlankSystemDaemon fucked around with this message at 12:57 on May 25, 2023

BlankSystemDaemon
Mar 13, 2009



namlosh posted:

That’s interesting… what would be the cheapest/simplest setup that could do this?
Like I know I can set my UDMPro up with redundant WAN connections from say, 2 ISPs but that’s still my UDM as single point of failure. What (two) consumer device(s) will do BGP?
Also can it be done with only one WAN connection?
UDMPro (and everything Ubiquiti) is Linux, not FreeBSD - so no pf, pfsync and carp.

On FreeBSD, pf is nowadays (as in, on the stable/13 branch) a very fast firewall, whereas pfsync exists to synchronize packet state over a out-of-band (usually direct, non-switched) connection, and carp is a alternative to the proprietary Cisco VRRP option.

opnSense is a fork of pfsense that's using a much more modern version of FreeBSD and it can do CARP - and they'll sell you ready-made appliances, too.

The above method doesn't involve BGP at all.

BlankSystemDaemon fucked around with this message at 09:41 on May 26, 2023

BlankSystemDaemon
Mar 13, 2009



Eletriarnation posted:

VRRP is actually from the IETF, although it's extremely similar to Cisco's HSRP.
Right.

CARP is still better, because Cisco only pinky-promised to not pursue any claims on their patent - which had the chilling effect Cisco intended.

BlankSystemDaemon
Mar 13, 2009



There's still 802.11a and 802.11b out there.

BlankSystemDaemon
Mar 13, 2009



There's a very classic network design called "router-on-stick" where you have one LAN port connected to a L2 or L3 switch, and then the switch is responsible for, well, switching traffic between the LAN devices, whereas the router does the actual routing (and firewalling) - and it's effectively what you get in any router with more than a couple ports, because it's such a simple design it's very hard to gently caress up.

It seems to me that the UDM Pro would do very well for what you're hoping to have it do, and with the SFP+ ports, you can even add one or more switches.
If you add a switch with PoE (or, better yet, use the PoE ports on the device, and add a switch for all other devices), you also have the option of doing Power-over-Ethernet for the unifi cameras which can record onto the storage that you can install into the UDM Pro - similar to how the access point you linked can be do power-over-ethernet.

Do also remember to make liberal use of VLANs - one for management, one for regular devices you trust, one for home surveillance, one for IoT devices you don't trust, one for guests if you end up using the hotspot functionality, and so on and so forth.

They don't require cloud access at all (though, if memory serves, you have to explicitly tell it that you don't want it).

BlankSystemDaemon fucked around with this message at 10:42 on Jun 20, 2023

Adbot
ADBOT LOVES YOU

BlankSystemDaemon
Mar 13, 2009



Sniep posted:

I still doubt they'd let you use that port to it's capacity but regardless.. nice.
Yeah, it's one thing for an ISP to say "you've got N Gbps now", it's another thing entirely to handle that the linerate 24/7/365.

I'm kinda surprised my ISP puts up with me using the 1/1Gbps@~$78/month at linerate.

  • 1
  • 2
  • 3
  • 4
  • 5
  • Post
  • Reply