[Cerowrt-devel] stanford talk/deluged in hardware/yurtlab

[Dave Taht]

On Sun, Feb 3, 2013 at 10:09 PM, Mark Constable <markc at renta.net> wrote:

> On Sun, 3 Feb 2013 10:17:42 AM Dave Taht wrote:
> > Well, I see it for 320.
>
> Yes, too much. At $220 it's good value and with a newer model out now
> I expect this price to drop under $200 AUD (inc delivery) during 2013.
>
> > Then you need to add a SSD,
>
> I've got a 4 Bay 2.5" SATA Removable Rack for the CD slot so I see the
> N40L as a catchall for my future hand me down 3.5" and 2.5" SATA drives
> on a real 3GB/s (maybe 6GB/s) SATA bus.
>
> > and a decent network card,
>
> Broadcom Corporation NetXtreme BCM5723 Gigabit Ethernet PCIe (rev 10)
>
> > Awful big, tho, in an era where I can get 1/2TB on an 2.5 inch SSD.
>
> I see it as a centralised private cloud server with up to 12TB of RAID1
> that is all-round useful enough to be just about anything, even a primary
> workstation in a pinch with Radeon HD4200 VGA out.
>
> Hereabouts (AU) they are very popular so I was kind of hoping that maybe
> the N40L (at ~$200) could be the basis of an "official" 64 bit version
> of CeroWrt based on the UML target.
>

Well, that sort of begs the question of why go through all the pain of
porting openwrt to an x86 product, when you can just install
ubuntu/fedora/any of a zillion other products on x86.

About the only compelling argument I can make for openwrt on x86 is the gui
interfaces and the 'light on flash' argument. The first makes it
theoretically easier for someone familiar with openwrt to configure an x86
based version.

The second is increasingly without point.

If you have compelling arguments for a high end x86 box for a 64 bit
cerowrt... go for it!

Please note if I could find a decent router box in the sub 120 dollar range
on x86 (with wifi, and 4 port switch) I'd dump mips in an instant. I think
cero has adequately proved that all these fancy new algorithms CAN run on
consumer hardware, already... although it seems as though the next
generation of all this consumer hardware puts WAY too much stuff in
hardware where we can't fix it in software. This latter point is my largest
concern going forward - trying to find/make/use hardware that can be
debloated.

At the moment I think however we're talking about two different things.

1) I was trying to spec a box (specifically for the yurtlab, (but others
have asked for it, too)) that would let me do stuff like do packet captures
at line rate and run mrtg  ( http://www.lns.com/papers/mrtg/ ) and that can
easily be an x86_64 box like this one... but god, the idea of making it run
openwrt when I can boot one up in ubuntu in 15 minutes....


vs

2) finding some sexy hardware that could be found at retail and improved.

As for 2, it's really looking grim. The arm folk treat ethernet as an
obsolete interface (which it is getting to be in the home!) and hook up one
chip via a usb bus, if that. The mips area is in disarray. x86 folk think
floating point and heat sinks are important in a router. ppc... ya know, I
haven't looked at life in ppc-land in a while....



>
> _______________________________________________
> Cerowrt-devel mailing list
> Cerowrt-devel at lists.bufferbloat.net
> https://lists.bufferbloat.net/listinfo/cerowrt-devel
>



-- 
Dave Täht

Fixing bufferbloat with cerowrt:
http://www.teklibre.com/cerowrt/subscribe.html
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.bufferbloat.net/pipermail/cerowrt-devel/attachments/20130203/cac10477/attachment-0002.html>