DIY home router

[evh909]

Thought I would share this article with everyone.....

http://arstechnica.com/gadgets/2016/01/numbers-dont-lie-its-time-to-build-your-own-router/

Link speaks for itself, found it interesting.

 

[L&LD]

What will be interesting is the subsequent article on how to configure Ubuntu for that kind of performance.

In the end though, you still need really deep pockets, a lot of knowledge and multiple devices to get what any mid or high end router offers most people today, with a setup wizard included.

A dedicated system running as a router and a great / excellent AP to handle the WiFi is still well north of $500 if you have to buy both new. Not that the performance that combination would offer isn't worth it, but on top of that you need to be a Linux kung fu ninja too.

These type of articles will force router manufacturers to give us better routers with real desktop class hardware and performance to match what ISP's are offering to more and more of us. But not yet.

 

[coxhaus]

I have found over the years running software routers back when I needed SPAM filtering for my mail server, software routers ran with lower latency than consumer routers which translated to lower ping times. The problem was all the patching, an ongoing experience using a Linux editor; almost every month there was a patch of some kind. And I was running a canned router system. No free style required. I can't even imagine trying to keep up with all the holes in software that needs to be patched. Just trying to stay on top it by yourself would be a nightmare.

 

[thiggins]

Thanks for the link. I'll be updating the Router Benchmark process soon and there are some good ideas here.

 

[sfx2000]

This would almost be a nice segue over to the $500 dollar router thread...

The take away I saw from the article over on ArsTechinica is that even high-end AC5300 class router - we're at a point where the SoC's are getting saturated - not just the cores, but memory and the switch fabric on these ARM based SoC's (Broadcom, QC-Atheros, and Marvell) - it's not really a fair fight to compare a dual-core Cortex-A9, even at 1.6GHz to a dual Core Celeron based on 3rd Generation Intel Core (Ivy Bridge) architecture - just no comparing that in any way that looks good for the SoC's...

Then went looking at the board support - mainstream Linux (Ubuntu 14.04LTS 64Bit) vs....

 

[System Error Message]

Its one of the benchmarks i would really like to do with very high end routers. its also something very worrying price wise because it means that the consumer routers have been getting more expensive, starting to cost a lot more than a x86 solution with a cheap wifi AP. $300 is the absolute max a consumer router should cost (you can still get low end i3 machines for around that price though). Its one of the reasons why i havent even bothered with any newer consumer routers. Even ubiquiti and mikrotik offer something way faster in that price range.

 

[Blade Runner]

A dedicated system running as a router and a great / excellent AP to handle the WiFi is still well north of $500 if you have to buy both new.

Perhaps separate devices is the better option for wireless and router functions.

These type of articles will force router manufacturers to give us better routers with real desktop class hardware and performance to match what ISP's are offering to more and more of us. But not yet.

IMO unlikely because it's a numbers game. Too many fanboys who will spend whatever for NetAsusGear’s Super Kraken 10GigaBlop wireless router with half-baked software.

 

[System Error Message]

There seems to be intel NUCs with dual NICs and wifi for less than ASUS current midrange routers (like the AC88U). So for the price of a midrange router from asus or netgear you can actually make a fanless compact x86 router and add a decent wifi AP behind it. If it doesnt need to be compact you can just salvage even from the core2 series and make a very cheap and fast router. The price of core2 based xeons are very cheap and linus even made a video on an 8core xeon core2 based system for $150. Unless people at least take the effort to install their own OS and configure things than consumer router manufacturers will just keep charging lots of money for wifi routers. Its also possible to get used 10Gb/s for around $30. SFP+ is usually better if you want 10Gb and SFP+ to SFP+ direct cables are also cheap.

So this should be in the $300 thread instead.

 

[rajl]

Thought I would share this article with everyone.....

http://arstechnica.com/gadgets/2016/01/numbers-dont-lie-its-time-to-build-your-own-router/

Link speaks for itself, found it interesting.

It's a good article, and illuminating. I definitely think that lays out a better testing methodology for evaluating router performance than what is currently done here. However, I think the article is making a bit of a mountain out of a molehill.

The tests are a little over the top. It's evaluating router performance for sustained bouts of 10,000 simultaneous connections. Most home users will rarely see that. For example, on my network the average load is around 300 connections at any given time, with maximum spikes up to 1,500 connections occurring once in a blue moon and the more normal spike being around 1,000 connections. These spikes are usuall no longer than a minute or two in duration and are usually linked to beginning a episode in Hulu it Netflix. I would have to set up an absurd scenario to hit 10,000 simultaneous connections in a sustained burst and I suspect most (but not all) home users would too. In the real world, these MIPS and ARM routers will perform just as well as when you build your own (excluding VPN performance).

 

[System Error Message]

There are a few important factors that make building your router better. It is less likely to freeze and you wont need to upgrade it for a while and theres more savings there. You also have a wide selection based on your budget from recycling old hardware that is still capable to using low powered small form factor machines. Upgrading is also cheap and you can cheaply customise it based on your needs. an upgrade to 10G would mean just adding a card and if you have a fibre optic connection than you could get SFP and the SFP module for it and not need a modem. I havent yet seen a consumer router with SFP.

So while performance is one factor the other factors are life span(how long you can use it before needing to upgrade), reliability, re-usability. So even if building your own router can be more expensive but you would last 2 or 3 more iterations longer before having to upgrade your hardware compared to using an embedded consumer router. Having a router that doesnt need to be restarted to change settings and not requiring random reboots helps and customisability lets you add more ram, storage or other features you need since they have PCIe. An interesting comparison would be the mikrotik CCR1009 vs consumer routers vs PC based routers in cost, performance and hardware features (excluding wifi). Ubiquiti is a little shameful since they only had dual core routers and never bothered to take the architecture further (no overclocking, no quad or 8 cores since the CPU has those variants), they used a good MIPS which even though unstable never used the 4 or 8 core variants so it is starting to fall behind in software processing. Software processing speed is very important because QoS and firewall are important and cant be done in hardware.

 

[Blade Runner]

It's a good article, and illuminating. I definitely think that lays out a better testing methodology for evaluating router performance than what is currently done here. However, I think the article is making a bit of a mountain out of a molehill.

The tests are a little over the top. It's evaluating router performance for sustained bouts of 10,000 simultaneous connections. Most home users will rarely see that. For example, on my network the average load is around 300 connections at any given time, with maximum spikes up to 1,500 connections occurring once in a blue moon and the more normal spike being around 1,000 connections. These spikes are usuall no longer than a minute or two in duration and are usually linked to beginning a episode in Hulu it Netflix. I would have to set up an absurd scenario to hit 10,000 simultaneous connections in a sustained burst and I suspect most (but not all) home users would too. In the real world, these MIPS and ARM routers will perform just as well as when you build your own (excluding VPN performance).

IMO the author's point was one client doing P2P activity can bring down a network. People are doing P2P activities such as gaming and torrents. Stress testing P2P activities should be part of every review.

 

[L&LD]

IMO the author's point was one client doing P2P activity can bring down a network. People are doing P2P activities such as gaming and torrents. Stress testing P2P activities should be part of every review.

That isn't the authors point at all. Just yours.

The point of the article was that routers now need to be replaced to keep up to ISP speeds because of their low performance hardware vs. what one may have laying around (or roughly equivalently to buy new).

In addition to having some (high) level of Linux and iptable knowledge too.

Testing P2P activity on routers is not what I would call critical testing of such hardware. But if it is a simple test to include for reviewers, why not?

 

[rajl]

IMO the author's point was one client doing P2P activity can bring down a network. People are doing P2P activities such as gaming and torrents. Stress testing P2P activities should be part of every review.

Actually, the author's point was that he found himself replacing routers more frequently than "when it dies" because his customers were increasing their internet speeds in a regular basis. The p2p example later in the article was how ~15 years ago, the dial-up modems his ISP ran could handle a lot of concurrent users, but got hosed by a single Limewire user. Not saying that BitTorrent can't hose a consumer router and a consumer connection, but the thesis seemed to be that consumer routers can't route at the speeds currently available and one should therefore build your own. I disagree with the thesis.

However, as note above, there are a number of other reasons to build your own. For example, maximizing throughput with QoS enabled is one reason as QoS disables hardware acceleration. Increasing VPN speeds is another, as most consumer routers lack hardware accelerated cryptographic instructions (e.g., AES-NI). However, if all you are doing is basic NAT and SPI firewalling, any $100 router is more than up to the task of routing at WAN speed for the vast majority of households. We on SNB tend to be the exception, rather than the rule.

 

[Blade Runner]

That isn't the authors point at all. Just yours.

The point of the article was that routers now need to be replaced to keep up to ISP speeds because of their low performance hardware vs. what one may have laying around (or roughly equivalently to buy new).

In addition to having some (high) level of Linux and iptable knowledge too.

Testing P2P activity on routers is not what I would call critical testing of such hardware. But if it is a simple test to include for reviewers, why not?

We'll agree to disagree. IMO author's anecdotal evidence was one client doing P2P activity can bring down a network. P2P activities are a large part of internet traffic even if the numbers/percentages are not accurate.
https://en.wikipedia.org/wiki/BitTorrent
I know very few people who limit their internet activities to email and CNN.

 

[L&LD]

We'll agree to disagree. IMO author's anecdotal evidence was one client doing P2P activity can bring down a network. P2P activities are a large part of internet traffic even if the numbers/percentages are not accurate.
https://en.wikipedia.org/wiki/BitTorrent
I know very few people who limit their internet activities to email and CNN.

You ignored the post above yours. And the fact that the P2P example was from 15 years ago.

 

[rajl]

We'll agree to disagree. IMO author's anecdotal evidence was one client doing P2P activity can bring down a network. P2P activities are a large part of internet traffic even if the numbers/percentages are not accurate.
https://en.wikipedia.org/wiki/BitTorrent
I know very few people who limit their internet activities to email and CNN.

And I know very few people who still use BitTorrent - probably just as many that limit their internet usage to email and CNN. Goes to show how varied the users of the Internet are.

I remember reading somewhere that, in North America at least, the majority of traffic is video streaming (Netflix, Hulu, YouTube, etc) during the evening. This kind of traffic is moderately high bandwidth due to H.264 compression (~5 megabit per stream), and involves a low number of concurrent connections (burst of ~300 per stream when the stream starts to buffer the stream, conduct authentication, etc. followed by a couple dozen ongoing connections as the stream is received based on my RRD graphs). So a household that streams in 3 devices at night plus web browsing and email is looking at using 20 megabits of bandwidth and bursting at 1,500-2,000 connections and sustaining 500 or so. These are hardly the tortuous conditions tested in the article and something that even my 6 year old WNDR3700 router can route at speed without breaking a sweat.

So I continue to disagree with the article's thesis - that consumer routers are unable to effectively route at speed. It's a relatively small sliver of individuals that have consumer routers that can't keep up with their internet speeds.