3 ms·

In general for small hardware I go with two options. Qualcomm-Atheros AR9331 based systems, with 64MB ram and 16MB flash (kind of hard to get in that config ch

by rdl 13y ago

In general for small hardware I go with two options. Qualcomm-Atheros AR9331 based systems, with 64MB ram and 16MB flash (kind of hard to get in that config cheaply). You can get decent performance, potentially hang a 5.8/2.4ghz radio on it too, and power it off a single laptop USB port (200mA normal power use, and with stuff, still less than 500mA.) Can put second ethernet, wifi, 3G, etc. there. There's no current commercial hardware in this space which is perfect. That should change shortly. Much beefier systems with "real" amounts of RAM and flash. I generally skip the Broadcom/RPi/etc. and go straight to an Atom or i3, but that's just for convenience, since I only need a small number of them. There's probably something more reasonable to do in the embedded space, but I'm indifferent on qty 1 devices between $500 and $200, especially if I already have most of the components for the $500 thing lying around, and it saves me an hour or two getting it working. There are a lot of things missing from widely available commercial hardware in this space right now.

weland 13y ago
> Qualcomm-Atheros AR9331 based systems, with 64MB ram and 16MB flash (kind of hard to get in that config cheaply). The Carambola 2 board is pretty cheap and it's quite awesome.
- rdl 13y ago
  Yeah; reworked TP-Link 703s are cheaper, though. Missing a couple of things I want, though.
  - nodata 13y ago
    Reworked how?
    - rdl 13y ago
      Someone who is less bad at surface mount soldering than I am can take a 702/703, pop the 2MB flash and 8-16M RAM chips off, and replace with 16/64. There are people in China who do this commercially, and some eBay stores. I've bricked (well, melted) 2 of them so far doing it myself.
      - weland 13y ago
        It sounds pretty hard to do using nothing but a soldering iron. The nice thing about the C2 is that you have all the I/O pins nicely exposed and there's a fairly helpful community built around them. It's not necessarily impressive in terms of documentation, but it's certainly better than something that's pretty much reverse-engineered.
nodata 13y ago
> That should change shortly. Tell us more :)
rwg 13y ago
FWIW, staples.com is selling the (discontinued) WD My Net N750 for US$30. It's an AR9344 box w/ 16 MB of flash, 128 MB of RAM, a 5 port GigE switch, 2 USB 2.0 ports, and a 0.1" pitch TTL serial header already installed on the PCB. It's supported by OpenWRT trunk, and most of the bits to support it are in FreeBSD, so that might be a "just works" option one day.
- rdl 13y ago
  Yeah, I care about being USB powered though.
stephen_g 13y ago
I've been looking at Intel's DPDK recently [1], and it looks really interesting for really fast packet processing in the 'beefier' kind of systems you describe. It would be a bit higher cost, but could mean you could make a router like the EdgeRouter but with good performance without the proprietary accelerator (I believe that Intel's drivers are open source). 1. Data Plane Development Kit, which lets you skip the kernel IP stack (which takes thousands of CPU cycles to process) and do packet processing in userland taking just tens to hundreds of cycles per packet. http://dpdk.org/ http://dpdk.org/
- rdl 13y ago
  I wonder how low the price of an Atom C2358 based system with 4x Intel Gig-E and either mini-PCI or some other solution for wireless would be. That would probably be what I'd build a "high end" home router, or low-end business router, on -- you'd still be able to hit PoE 802.af 2009 (25W). DPDK supports Atom. This Atom has both virtualization and AES-NI, so you could do some interesting security things (the fireeye-style "run malware in a VM", and linerate crypto). ~$120 in parts means a retail price around $500. That doesn't seem at all unreasonable to me, but there's always the "sell it for $250 and make $5-10/mo MRC" option. I think Cisco/Linksys tried that but it was pretty weak.
- pktgen 13y ago
  > 1. Data Plane Development Kit, which lets you skip the kernel IP stack (which takes thousands of CPU cycles to process) and do packet processing in userland taking just tens to hundreds of cycles per packet. http://dpdk.org/ http://dpdk.org/ I wish OS developers saw this as a problem. There is no reason kernel stacks should be so slow for tasks where all processing is done in the kernel. (For packets destined to userspace, you've got the syscall overhead to deal with.) I recently tested the Linux network stack's PPS performance with an Intel X520 10GbE NIC. I used Debian testing, with the 3.12 kernel. My destination machine was an i7-3930K at stock speed. I wrote a simple kernel module adding an NF_IP_PRE_ROUTING hook returning NF_DROP with no processing, which would be the simplest possible code path. For a packet generator, I used another older machine with another X520, using the "pfsend" tool included with PF_RING, and the card in PF_RING DNA mode. That was easily able to saturate the link at line rate (14.8M PPS). The result: the kernel was only able to sustain about 2.8M PPS. I then loaded the DNA driver on the destination machine, used the included "pfcount" tool, and no packet drops - it was receiving the full 14M PPS. I tested DPDK recently and had similar results. I also modified the Linux ixgbe driver ixgbe_clean_rx_irq() function, and added a step in between the "fetch packets from RX ring" and "put packet in SKB and send to network stack" functions. Even when I added a bunch of useless comparisons for each packet, I was able to get ~12-13M PPS. I could get line rate by just dropping and not doing any processing.

by rdl 13y ago

weland 13y ago

> Qualcomm-Atheros AR9331 based systems, with 64MB ram and 16MB flash (kind of hard to get in that config cheaply). The Carambola 2 board is pretty cheap and it's quite awesome.

rdl 13y ago
Yeah; reworked TP-Link 703s are cheaper, though. Missing a couple of things I want, though.
- nodata 13y ago
  Reworked how?
  - rdl 13y ago
    Someone who is less bad at surface mount soldering than I am can take a 702/703, pop the 2MB flash and 8-16M RAM chips off, and replace with 16/64. There are people in China who do this commercially, and some eBay stores. I've bricked (well, melted) 2 of them so far doing it myself.
    - weland 13y ago
      It sounds pretty hard to do using nothing but a soldering iron. The nice thing about the C2 is that you have all the I/O pins nicely exposed and there's a fairly helpful community built around them. It's not necessarily impressive in terms of documentation, but it's certainly better than something that's pretty much reverse-engineered.

nodata 13y ago

> That should change shortly. Tell us more :)

rwg 13y ago

FWIW, staples.com is selling the (discontinued) WD My Net N750 for US$30. It's an AR9344 box w/ 16 MB of flash, 128 MB of RAM, a 5 port GigE switch, 2 USB 2.0 ports, and a 0.1" pitch TTL serial header already installed on the PCB. It's supported by OpenWRT trunk, and most of the bits to support it are in FreeBSD, so that might be a "just works" option one day.

rdl 13y ago
Yeah, I care about being USB powered though.

stephen_g 13y ago

I've been looking at Intel's DPDK recently [1], and it looks really interesting for really fast packet processing in the 'beefier' kind of systems you describe. It would be a bit higher cost, but could mean you could make a router like the EdgeRouter but with good performance without the proprietary accelerator (I believe that Intel's drivers are open source). 1. Data Plane Development Kit, which lets you skip the kernel IP stack (which takes thousands of CPU cycles to process) and do packet processing in userland taking just tens to hundreds of cycles per packet. http://dpdk.org/ http://dpdk.org/

rdl 13y ago
I wonder how low the price of an Atom C2358 based system with 4x Intel Gig-E and either mini-PCI or some other solution for wireless would be. That would probably be what I'd build a "high end" home router, or low-end business router, on -- you'd still be able to hit PoE 802.af 2009 (25W). DPDK supports Atom. This Atom has both virtualization and AES-NI, so you could do some interesting security things (the fireeye-style "run malware in a VM", and linerate crypto). ~$120 in parts means a retail price around $500. That doesn't seem at all unreasonable to me, but there's always the "sell it for $250 and make $5-10/mo MRC" option. I think Cisco/Linksys tried that but it was pretty weak.
pktgen 13y ago
> 1. Data Plane Development Kit, which lets you skip the kernel IP stack (which takes thousands of CPU cycles to process) and do packet processing in userland taking just tens to hundreds of cycles per packet. http://dpdk.org/ http://dpdk.org/ I wish OS developers saw this as a problem. There is no reason kernel stacks should be so slow for tasks where all processing is done in the kernel. (For packets destined to userspace, you've got the syscall overhead to deal with.) I recently tested the Linux network stack's PPS performance with an Intel X520 10GbE NIC. I used Debian testing, with the 3.12 kernel. My destination machine was an i7-3930K at stock speed. I wrote a simple kernel module adding an NF_IP_PRE_ROUTING hook returning NF_DROP with no processing, which would be the simplest possible code path. For a packet generator, I used another older machine with another X520, using the "pfsend" tool included with PF_RING, and the card in PF_RING DNA mode. That was easily able to saturate the link at line rate (14.8M PPS). The result: the kernel was only able to sustain about 2.8M PPS. I then loaded the DNA driver on the destination machine, used the included "pfcount" tool, and no packet drops - it was receiving the full 14M PPS. I tested DPDK recently and had similar results. I also modified the Linux ixgbe driver ixgbe_clean_rx_irq() function, and added a step in between the "fetch packets from RX ring" and "put packet in SKB and send to network stack" functions. Even when I added a bunch of useless comparisons for each packet, I was able to get ~12-13M PPS. I could get line rate by just dropping and not doing any processing.