okaaay so the plan is to restart the eoma68 router project, this time with a pre-existing reference design based on the QCA9531. that has a PCIe interface, USB2 and a 5-port GbE *and* a 2x2 2.4ghz WIFI antenna. full source is available for everything so it can be entirely libre and RYF Certified.
the advantage of having an EOMA68 Card in the router should be clear: the Card will have considerably more resources: RAM, CPU cycles etc. meaning that VPNs can be done without high latency, yet take advantage of the LAN capabilities of the 5-port... you could put in a MiniPCIe Card (a *proper* one) e.g. a 3G/4G/LTE Modem, WIMAX, 802.11ac... blah blah.
the tricky bit: connecting the EOMA68 Card to the QCA9531. now, i took a look at the Reference Design and i *really* do not want to touch the layout for the GbE, WIFI or PCIe. so i figured, why not connect the EOMA68 USB2 host interface back-to-back with the QCA9531's USB host?
turns out that something called the Cypress AN2720 can do exactly that, and it comes up as a cdc_subset of the usbnet linux kernel driver. yay! quick search online: the datasheet is publicly available, easy to find on digikey. yay!
so i would assume, because it's not an actual 10/100 ethernet, that it would run at (saturate) the full 480mb/sec of USB2. so not quite GbE speeds but pretty damn close. yay!
anyway should be quite straightforward.
l.
--- crowd-funded eco-conscious hardware: https://www.crowdsupply.com/eoma68
Replying by phone; please forgive the top-posting (and potential typos) that result from that.
Would it be possible to have *two* PCIe Mini Card slots, say one for WiFi and one for 4g/LTE?
That would be awesome. Mom hates cables, and I mostly agree with her on that... so look much easier, with the router in the front hall, to run WiFi rather than Ethernet to the kitchen table, two rooms and ~30ft away...
On May 5, 2017 10:12 AM, "Luke Kenneth Casson Leighton" lkcl@lkcl.net wrote:
okaaay so the plan is to restart the eoma68 router project, this time with a pre-existing reference design based on the QCA9531. that has a PCIe interface, USB2 and a 5-port GbE *and* a 2x2 2.4ghz WIFI antenna. full source is available for everything so it can be entirely libre and RYF Certified.
the advantage of having an EOMA68 Card in the router should be clear: the Card will have considerably more resources: RAM, CPU cycles etc. meaning that VPNs can be done without high latency, yet take advantage of the LAN capabilities of the 5-port... you could put in a MiniPCIe Card (a *proper* one) e.g. a 3G/4G/LTE Modem, WIMAX, 802.11ac... blah blah.
the tricky bit: connecting the EOMA68 Card to the QCA9531. now, i took a look at the Reference Design and i *really* do not want to touch the layout for the GbE, WIFI or PCIe. so i figured, why not connect the EOMA68 USB2 host interface back-to-back with the QCA9531's USB host?
turns out that something called the Cypress AN2720 can do exactly that, and it comes up as a cdc_subset of the usbnet linux kernel driver. yay! quick search online: the datasheet is publicly available, easy to find on digikey. yay!
so i would assume, because it's not an actual 10/100 ethernet, that it would run at (saturate) the full 480mb/sec of USB2. so not quite GbE speeds but pretty damn close. yay!
anyway should be quite straightforward.
l.
crowd-funded eco-conscious hardware: https://www.crowdsupply.com/eoma68
arm-netbook mailing list arm-netbook@lists.phcomp.co.uk http://lists.phcomp.co.uk/mailman/listinfo/arm-netbook Send large attachments to arm-netbook@files.phcomp.co.uk
--- crowd-funded eco-conscious hardware: https://www.crowdsupply.com/eoma68
On Fri, May 5, 2017 at 3:19 PM, Christopher Havel laserhawk64@gmail.com wrote:
Replying by phone; please forgive the top-posting (and potential typos) that result from that.
Would it be possible to have *two* PCIe Mini Card slots, say one for WiFi and one for 4g/LTE?
yes i was just considering that. one wired for USB (only).
actually now that i think about it, it would be sensible to put a GL850G onto the QCA9531, then connect both the MiniPCIe (USB only) and AN2720 onto that, rather than have the MiniPCIe wired directly to an EOMA68 USB.
that way the MiniPCIe is still accessible by the QCA9531 even if there's no EOMA68 Card in use.
That would be awesome. Mom hates cables, and I mostly agree with her on that... so look much easier, with the router in the front hall, to run WiFi rather than Ethernet to the kitchen table, two rooms and ~30ft away...
yeahyeah. well in this case 2.4ghz WIFI will already be on-board, but if you wanted 5.4ghz or 802.11ac (assuming you can get a MiniPCIe 802.11ac Card) that's what you'd use that slot for.
l.
https://wireless.wiki.kernel.org/en/users/drivers/ath10k QCA9880. ath10k. proprietary. blegh. but it's MiniPCIe.
so... doable.
l.
Better, in my opinion, to have WiFi on a card. I've yet to meet a WiFi adapter that didn't burn itself out within a year or two. In fact, the one I'm using right now (Rosewill RNX-EasyN1 from Newegg) is barely piping data at all... two weeks ago I could stream Pandora on here, but not today. I've a replacement on the way, but I only ordered it yesterday so it's going to be a (long) week...
For some reason, this card wants to work better if I plug it in after the system completely boots, rather than if I boot the system with the card attached. Oy. Machine is a Lenovo X420, before anyone asks -- its internal card is even more horrible than the burnt-out USB one I'm griping about... and I can't really change that without serious BIOS mucking (hooray for whitelists, not!), which I pretty well don't want to do.
On Fri, May 5, 2017 at 3:47 PM, Christopher Havel laserhawk64@gmail.com wrote:
Better, in my opinion, to have WiFi on a card.
QCA9531 it's built-in and i'm not messing with the layout, so it stays. MiniPCIe (actual PCIe) is already on the reference design so _that_ stays. USB2 MiniPCIe (with a SIM card) is something to be added, definitely.
l.
Hi Luke have you looked into the Atheros AR9590 and AR9580? I think those are the fastest i could find in the wild
On Fri, May 5, 2017 at 6:03 PM, Luke Kenneth Casson Leighton lkcl@lkcl.net wrote:
On Fri, May 5, 2017 at 3:47 PM, Christopher Havel laserhawk64@gmail.com wrote:
Better, in my opinion, to have WiFi on a card.
QCA9531 it's built-in and i'm not messing with the layout, so it stays. MiniPCIe (actual PCIe) is already on the reference design so _that_ stays. USB2 MiniPCIe (with a SIM card) is something to be added, definitely.
l.
arm-netbook mailing list arm-netbook@lists.phcomp.co.uk http://lists.phcomp.co.uk/mailman/listinfo/arm-netbook Send large attachments to arm-netbook@files.phcomp.co.uk
On Fri, May 5, 2017 at 7:16 PM, Allan Mwenda allanitomwesh@gmail.com wrote:
Hi Luke have you looked into the Atheros AR9590 and AR9580? I think those are the fastest i could find in the wild
it's more a practical matter of what's accessible. have a quick look: search for any combination of keywords "AR9580 datasheet" or anything along those lines. see if you can find something - anything. even an indication of what chipset's in it... ah! "AR9580 linux" turns up this: https://wiki.debian.org/ath9k whew ok so it's an ath9k which means firmware's on-board (whew).
l.
Is msata okay? I know its sold as Compex msata cards
On 5 May 2017 21:30:00 GMT+03:00, Luke Kenneth Casson Leighton lkcl@lkcl.net wrote:
On Fri, May 5, 2017 at 7:16 PM, Allan Mwenda allanitomwesh@gmail.com wrote:
Hi Luke have you looked into the Atheros AR9590 and AR9580? I think
those
are the fastest i could find in the wild
it's more a practical matter of what's accessible. have a quick look: search for any combination of keywords "AR9580 datasheet" or anything along those lines. see if you can find something - anything. even an indication of what chipset's in it... ah! "AR9580 linux" turns up this: https://wiki.debian.org/ath9k whew ok so it's an ath9k which means firmware's on-board (whew).
l.
arm-netbook mailing list arm-netbook@lists.phcomp.co.uk http://lists.phcomp.co.uk/mailman/listinfo/arm-netbook Send large attachments to arm-netbook@files.phcomp.co.uk
so i would assume, because it's not an actual 10/100 ethernet, that it would run at (saturate) the full 480mb/sec of USB2. so not quite GbE speeds but pretty damn close. yay!
Based on what USB2 gives us with "mass-storage" devices, 30MB/s is basically the upper bound. And FWIW, when I connect my desktop to my A20-based router via USB2 on one side and USB-OTG on the other (using the "gether" gadget), I'm getting about 10MB/s, so "faster than fast-ethernet" maybe, but be surprised if you get "close" to GbE speeds.
Stefan
On Fri, May 5, 2017 at 8:04 PM, Stefan Monnier monnier@iro.umontreal.ca wrote:
so i would assume, because it's not an actual 10/100 ethernet, that it would run at (saturate) the full 480mb/sec of USB2. so not quite GbE speeds but pretty damn close. yay!
Based on what USB2 gives us with "mass-storage" devices, 30MB/s is basically the upper bound. And FWIW, when I connect my desktop to my A20-based router via USB2 on one side and USB-OTG on the other (using the "gether" gadget), I'm getting about 10MB/s, so "faster than fast-ethernet" maybe, but be surprised if you get "close" to GbE speeds.
g_ether over the A20's musb (Mentor USB) interface is... very broken. musb is a ridiculously-low-cost OTG controller that has to be partially-implemented in softtware. the current state of the linux driver for musb is completely fucked-up. for example: if you plug a USB3 hub into it, then plug in a USB 1.1 keyboard, it goes "aargh fuck i have no fucking idea what to do, aaiyaaa splurgh". likewise the usb speed-allication code is all screwed up: you can just about get away with plugging in one device but a hub and then multiple devices: forget it.
so what you are probably running into is the musb driver going "ha! you plugged in a *what*?? pffh i have no idea what speed that is so let's just assume it's USB 1.0 mkaay?"
a proper USB2 host controller should have none of these difficulties.
l.
g_ether over the A20's musb (Mentor USB) interface is... very broken.
No doubt. But IME that tends to apply to a large proportion of USB thingies.
so what you are probably running into is the musb driver going "ha! you plugged in a *what*?? pffh i have no idea what speed that is so let's just assume it's USB 1.0 mkaay?"
I don't know what's the source of the 10MB/s performance, but note that 10MB/s is still 7 times faster than the theoretical upper bound of USB1, so it's definitely not putting itself in USB1 mode.
a proper USB2 host controller should have none of these difficulties.
Let's hope so,
Stefan
--- crowd-funded eco-conscious hardware: https://www.crowdsupply.com/eoma68
On Fri, May 5, 2017 at 8:30 PM, Stefan Monnier monnier@iro.umontreal.ca wrote:
g_ether over the A20's musb (Mentor USB) interface is... very broken.
No doubt. But IME that tends to apply to a large proportion of USB thingies.
so what you are probably running into is the musb driver going "ha! you plugged in a *what*?? pffh i have no idea what speed that is so let's just assume it's USB 1.0 mkaay?"
I don't know what's the source of the 10MB/s performance, but note that 10MB/s is still 7 times faster than the theoretical upper bound of USB1, so it's definitely not putting itself in USB1 mode.
which one's the... wait... you said 10MB/s - is that 10 Megabytes/sec? which is 100mbits / sec?
I don't know what's the source of the 10MB/s performance, but note that 10MB/s is still 7 times faster than the theoretical upper bound of USB1, so it's definitely not putting itself in USB1 mode.
which one's the... wait... you said 10MB/s - is that 10 Megabytes/sec? which is 100mbits / sec?
That's right.
Stefan
On Fri, May 5, 2017 at 9:29 PM, Stefan Monnier monnier@iro.umontreal.ca wrote:
I don't know what's the source of the 10MB/s performance, but note that 10MB/s is still 7 times faster than the theoretical upper bound of USB1, so it's definitely not putting itself in USB1 mode.
which one's the... wait... you said 10MB/s - is that 10 Megabytes/sec? which is 100mbits / sec?
That's right.
ok so that's odd, why limit a virtual device's speed?
Hi,
On Fri, May 05, 2017 at 03:04:37PM -0400, Stefan Monnier wrote:
so i would assume, because it's not an actual 10/100 ethernet, that it would run at (saturate) the full 480mb/sec of USB2. so not quite GbE speeds but pretty damn close. yay!
Based on what USB2 gives us with "mass-storage" devices, 30MB/s is basically the upper bound. And FWIW, when I connect my desktop to my A20-based router via USB2 on one side and USB-OTG on the other (using the "gether" gadget), I'm getting about 10MB/s, so "faster than fast-ethernet" maybe, but be surprised if you get "close" to GbE speeds.
That is slower than Fast Ethernet which is 12.5MB/s.
Erik
http://rhombus-tech.net/community_ideas/router/news/
ok so i've done a preliminary layout and it's to be a 160x160 PCB. the QCA9531 Reference Design's layout - i'm not touching it with a 10ft barge pole. it contains *really* complex R.F. layout and PCIe connectivity. i'll be routing round this layout through some of the gaps, to re-route some of the GPIO and also the USB connections, but that's it: nothing else.
it's actually a really powerful board, with connectivity similar to that of pcengines.ch alix boards and some of mikrotic's routers.
l.
Well, the router looks really good. Finally a possible competitor for https://omnia.turris.cz/en/ .
I think there might be quite a huge user based interested in such a router (judging based on the interest in Turris routers - see e.g. their Kickstarter campaign).
--- crowd-funded eco-conscious hardware: https://www.crowdsupply.com/eoma68
On Sun, May 7, 2017 at 1:37 PM, dumblob dumblob@gmail.com wrote:
Well, the router looks really good. Finally a possible competitor for https://omnia.turris.cz/en/ .
shrieeek! $EUR 339!! *splutter*! the qca9531 is $10, the rest of the components would bring it to a mass-volum retail cost of say... $75, and an EOMA68-A20 Card at mass-volume retail pricing would be around $30.
I think there might be quite a huge user based interested in such a router (judging based on the interest in Turris routers - see e.g. their Kickstarter campaign).
good to hear.
arm-netbook@lists.phcomp.co.uk
-
Allan Mwenda -
Christopher Havel -
dumblob -
Erik Auerswald -
Luke Kenneth Casson Leighton -
Stefan Monnier