from 2015 seems to support that idea. Basically those XCY boxes are high end OTC routers. If they just installed IPFire on them out of the box, they'd be kinda what you're talking about
So if your argument is that in the future you will get a grunty little workhorse box and it will have an Celeron or Atom processor and do a lot for very little money... then we're basically saying the same thing, except I'm saying that if you pay a little more now you can already have it, and it's worth it compared to much of what's actually available in the consumer router space.
The other aspect though, is that for LEDE enthusiasts who are flashing their own software... there is a shrinking price gap between a good consumer router, and a full on PC like the XCY or whatever. And that shrunken gap has lots of implications because much of the special purpose embedded stuff... MTD layouts and soldering your own serial ports and bootloaders and getting around region codes while flashing files and whatnot are just a pain in the ass that goes away when you have standard PC hardware. If the difference in price is $100 for a fancy new x86 Lantiq with 256MB RAM and 1G flash, and $200 for a nice standard PC with 4Gig RAM and 128 Gig removable flash or something... I know how I'm going to spend my money so I can save a lot of hassle and time.
You are absolutely right, we hit idle = zero at around 100Mbps with QoS on, so any headroom above that point is probably really pushing it.
Too bad about the performance step down in LEDE, but out of CPU is out of CPU. Time to look at the box linked above. I kind of like having something totally focused on routing/firewall duties and no WiFi (already have the aforementioned TP-Link EAP's).
Ah, good point, my untold assumption was I need a vdsl modem which I would prefer to be integral part of my router and then the options are extremely limited (well there is lantiq/intel or nothing).
I currently use a lantiq device using lede as bridged modem in front of a ar71xx device as the main router, and just because the anemic CPU in the lantiq device is already exhausted by concurrently running the vdsl modem, nat, pppoe, firewall and wifi; so I am looking forward to a lantiq device with a somewhat more powerful cpu. Without that constraint, I agree with your assessment
Exactly my thoughts. Was looking at pricy OTC routers with multiple cores and all that, but boxes like the one you linked seem tailor made for the typical LEDE crowd.
However, it is great that LEDE supports a wide variety of OTC routers, as some can not swing the $200.
Yes, I think those boxes are perfect for the typical LEDE crowd. Running LEDE itself... well maybe or maybe not depending. IPFire or just a raw Debian or Arch or something might make sense when the embedded small-storage-space restrictions are no longer there.
But, when it comes to Access Points, this I think is where LEDE could shine in the future. Because it makes sense to me that you'd still like more flexibility and custom services and security etc together with specialized RF hardware and antennas and PoE and soforth that you're not going to replace with a custom PC kit.
So I'd really love to see more support for TP-Link APs and the like to compliment the transition to more people using grunty little 4 NIC mini Linux PCs as their dedicated routers.
But that is similar for LEDE on non-x86 hardware, only lantiq XDSL-modem chips are supported, and as far as I can tell there are exactly 0 LEDE supported DOCSIS modem chips. So more often than not one already needs to bridge a proper "modem" (I know, these do not exist anymore these are all routers nowadays, but one can ask them to play dumb ). In that case x86 becomes an option again...
If you are absolutely certain you are able to attain far higher speeds on standard, but older builds, you can do a Git bisect to try and find which commit introduced the regression and file a bug report.
Just compile an old version, without any optimizations, and verify if the speed is indeed much higher. If so, start the bisect process until you find the commit that broke the speed.
The likely result is a Linux kernel update. One of the updates from 3.18 to 4.1 to 4.4 to 4.9 to 4.14.
Example: routing performance deteriorated in a major way between AA12.09 and BB14.07 as the routing cache was remove from Linux kernel in kernel 3.6, so the performance of AA12.09 with kernel 3.3 and of BB14.07 with 3.10 differed quite much. Same has happened with other major kernel version jumps, so finding the Openwrt commit may not help that much, if the commit is a major kernel version upgrade (as I guess is very likely)
We may wonder if you really should upgrade kernel in router oriented projects like openwrt/lede then. Because for me theses bugs make the routers unusable for what I want to do, while they where perfectly usable with openwrt 15.05.
Because newer kernels fix security issues, which would be a pain to keep on back porting and increasingly more difficult as the kernel version in Lede gets more and more behind the upstream version. Newer kernels can also introduce new features that can be useful to have on a router.
I would strongly recommend not running such old software. Your router is the device between your entire network and the big bad world called the internet. Having up-to-date version on such a device is mandatory IMO.
They are not bugs, but negative impact from feature development (mainly due to hardware performance expectations gradually increasing over time).
Also routers have a lifetime. Similarly as your PC/tablet/whatever... An 8-year old PC or tablet might feel a bit ancient, right? So, why should the same not apply to routers?
Old routers will fall behind the usability curve. When I seriously started with Openwrt and started to publish a community build in 2011 ( https://forum.openwrt.org/viewtopic.php?id=28392 ), WNDR3700 was a pretty powerful device with 8 MB flash and 64 MB RAM. Successor models like WNDR3800 doubled flash and RAM. But the core chips in the router family remained the same and the processing power fell behind the current connection speeds. The router still works ok, but its small processing power limits the throughput quite a lot.
Additionally, like said by @Mushoz , security issues are one good reason why we need to move forward and keep updating.
A lot of the topics in this discussion are why I keep plugging the low end of x86 as a dedicated routing platform. You will not go wrong with a $150 to $200 investment that will last 10 years and provide routing ability up to gigabit speeds. Have your wifi be separate via wired PoE access points. The economics today make sense whereas in 2005 the economics pointed towards using low end all-in-one solutions because the bandwidth available was so low.
Part of the problem is that projects like this are always running old kernels,
if people were running current kernels and reported the drop in performance, the
kernel devs would have second thoughts about the changes.
But when the feedback comes years later, it's harder to influence the kernel.
). In that case x86 becomes an option again...