| 07:29 | topi` | rabeeh: is it possible to order a custom build of ClearFog with 2 GB or 4 GB of RAM? |
| 07:37 | jnettlet | topi`, I believe we are doing a respin of the SOM that will support 2 GB. I don't think 4GB will be supported. |
| 07:38 | topi` | jnettlet: 2 GB would be totally acceptable. We just happen to have a CTO who thinks that Linux needs 4 or 8 GB to be useful ;) |
| 07:39 | jnettlet | with no gui? |
| 07:39 | topi` | we usually cache a lot of stuff to RAM |
| 07:39 | topi` | since writing to disks (even flash) is prone to failure in vehicles with erratic power, etc |
| 07:40 | jnettlet | sure |
| 07:40 | topi` | jnettlet: our current x86-based solution, that the ClearFog would replace, actually uses 14-20 W power when idling. Can the CF beat this figure? |
| 07:41 | jnettlet | let me double check and get back to you. I know we had discussed it but sometimes production timetables change. |
| 07:41 | jnettlet | topi`, yes...but you do know we have an x86 based solution that is a 7W TDP |
| 07:41 | jnettlet | it can be configured down to 4W |
| 07:41 | topi` | we haven't been paying attention to power use previously, but having minimum idle power would open up new use cases |
| 07:41 | topi` | jnettlet: we have no specific need for x86, it's linux after all :) however, we do have customers who demand 2 LTE modems |
| 07:42 | topi` | (this is because of redundancy) |
| 07:42 | topi` | since the mobile link is the weakest point in our system, by far |
| 07:42 | jnettlet | the clear fog isn't our most power efficient but idle is still only about 3-4W |
| 07:43 | topi` | yeah, I'd figure it is a decent number |
| 07:43 | topi` | the scaleway guys from France use ARMADAs in their baremetal cloud server nodes, and they thought it was pretty power efficient |
| 07:43 | jnettlet | obviously adding a couple of modems will increase that, but the Telit modems I use only draws like 40 mA at idle, ramping up to 80-100mA when transmitting |
| 07:44 | jnettlet | we are still working with Marvell to fix the cpuidle support to drop that number down. |
| 07:44 | topi` | yes, I'dthink around 50 mA idle for the Huawei modems we use |
| 07:45 | jnettlet | I have my power monitor hooked up to an iMX6 right now. When I get a chance I will drop a couple of modems in a ClearFog and test some numbers for you. |
| 07:45 | jnettlet | Although would the ClearFog Base be a better fit for you? |
| 07:45 | topi` | this is not *bad* for a stupid modem, if I think of past times when the Intel guys came to us (Nokia) and offered some samples of Medfield Atom and the bloody chip was *idling* at 20 mA |
| 07:46 | jnettlet | It removes the switch |
| 07:46 | topi` | not idling, in deep sleep |
| 07:46 | jnettlet | yeah. the iMX6 does deep sleep at about 30mA |
| 07:46 | topi` | jnettlet: we're also interested in the switch, and mostly it's about having 2 modems |
| 07:46 | jnettlet | oh okay. |
| 07:47 | topi` | and, for robustness, we would be using the eMMC (built-to-order) |
| 07:47 | topi` | however, is it possible for the ARMADA to boot from M.2 disk? |
| 07:47 | jnettlet | yes |
| 07:48 | topi` | great news |
| 07:48 | jnettlet | you can choose the boot medium from the dip switches on the top |
| 07:48 | jnettlet | options are SPI, eMMC/SDHC, m.2, UART |
| 07:48 | topi` | however, the eMMC is a very cost-effective way of providing reliable boot medium |
| 07:49 | topi` | getting a proper SLC based 8 or 16 GB M.2 disk will cost $50-$100 |
| 07:49 | jnettlet | yes, although m.2 is dropping in price like crazy. Especially for smaller cards |
| 07:49 | jnettlet | oh I think we can get them much cheaper. |
| 07:49 | topi` | they're really milking the "industrial grade disk" premium |
| 07:51 | topi` | oh, I've experienced strange HW problem with the HB Gate. I was hooking up a keyboard to one of the USB slots and the device rebooted itself. I wonder why? |
| 07:51 | topi` | it was definitely not a kernel panic, but some kind of HW glitch |
| 07:51 | topi` | after it, the MMC did not boot, but it exhibited millions of I/O errors |
| 07:52 | topi` | no amount of rebooting fixed it, I thought I had fried the SD card. but I took it off of power, let it be for half an hour, and then it worked. |
| 07:52 | jnettlet | that is odd. MMC and not SDHC? |
| 07:52 | topi` | SDHC |
| 07:52 | topi` | maybe some oddity with the power supply I used... no figure |
| 07:52 | topi` | go figure |
| 07:53 | topi` | unfortunately, I don't have another Gate to test this on |
| 07:53 | topi` | I might pick up one now that I'm visiting our offices |
| 07:53 | jnettlet | ah okay...yeah I have seen this. Generally I realize I have bumped the SDHC or jarred the board enough that the SDHC jiggles and it crashes the firmware on the card. |
| 07:54 | jnettlet | you need to unplug everything to actually let the SDHC's ARM chip fully reset |
| 07:55 | jnettlet | I have never seen this if the board is fully secured or enclosed, but with just bare boards I manage to do it occasionally. The card is right on the corner where I want to grab the board to plug USB drives in and out of the far side usb plugs |
| 07:56 | topi` | jnettlet: agreed, I have grabbed the Gate several times from the SD card holder :) |
| 07:56 | jnettlet | like you said eMMC is more robust in this case |
| 07:56 | topi` | this is most probably what happened, the ARM core is probably so low power that a mere few seconds of cooling off isn't enough |
| 07:57 | topi` | btw, which kernel do you offer for the ARMADA? |
| 07:57 | topi` | I don't know about Marvell's mainline efforts |
| 07:57 | jnettlet | right now most are images are built around 3.10...much of Marvell's work has been pushed mainline. |
| 07:58 | topi` | I do know they used to employ a couple of linux kernel hackers back in 2007 or so (I met those guys at Ottawa) |
| 07:58 | jnettlet | We are currently working to finalize a 4.4.y LTS branch that has the rest that hasn't been mainlined yet |
| 07:58 | topi` | that's great news |
| 07:58 | topi` | would that be a suitable starting point to start building prototypes on? |
| 07:58 | jnettlet | We are working closely with Marvell. https://github.com/MarvellEmbeddedProcessors |
| 07:59 | topi` | I don't mind having to add a couple of patches every now and then :) |
| 07:59 | jnettlet | do you want an embedded OS, or a full development OS for now? |
| 08:00 | topi` | we base our products on Debian |
| 08:00 | jnettlet | Then I would grab images here that have our initial 4.4 kernel. https://images.solid-build.xyz/A38X/ |
| 08:00 | topi` | we have a list of packages from Jessie that we build the rootfs from |
| 08:00 | jnettlet | it will be upgraded very soon, we are just waiting to get approval from Marvell to release some code. |
| 08:01 | jnettlet | It is all OSS/GPL but there is always a legal process for big companies |
| 08:02 | topi` | interesting that the Armada 388 microSOM is just as cheap as the Braswell one |
| 08:02 | topi` | even though, by judging the amount of I/O, the Armada is clearly a beefier chip. Braswell wins on pure CPU perf. |
| 08:03 | topi` | jnettlet: yeah, legal processes take time. I used to work on Linux at Nokia, and we had to take a "legal course" and consult the lawyers before publishing |
| 08:09 | jnettlet | yeah, different use cases. |
| 08:11 | jnettlet | Braswell has a GPU and hardware video decoding. I/O throughput is probably a wash but I haven't fully pushed the Braswell. It doesn't have much of the advanced TCP or storage functions that the Armada has. |
| 08:11 | jnettlet | and the Armada can handle many more dedicated lanes for I/O |
| 08:12 | jnettlet | and the Braswell doesn't have hardly any additional I/O like i2c, gpio etc |
| 08:12 | topi` | yeah, different use cases |
| 08:12 | jnettlet | topi`, will you be using the PoE functionality of the ClearFog? |
| 08:13 | topi` | if we get a customer from passenger train business, then yeah |
| 08:13 | topi` | they frequently employ PoE |
| 08:13 | topi` | we're currently negotiating with a major EUropean rail company |
| 08:14 | topi` | problem is, the Railway certs like EN50155 are expensive to get |
| 08:15 | jnettlet | oh is this for onboard wifi uplink? |
| 08:16 | topi` | they have entire systems that communicat to ground station |
| 08:16 | topi` | like alerts, error codes, etc |
| 08:16 | topi` | not just wifi uplinks :) |
| 08:16 | jnettlet | sure of course. most of that is automated at this point. |
| 08:16 | jnettlet | well good luck. Let us know if there is anything we can help with. |
| 08:16 | topi` | the rail company I know uses expensive Cisco equipment, we'll be able to compete in price, definitely |
| 08:17 | jnettlet | certainly. |
| 08:18 | topi` | do you have the metal case for CF? how many slots it has for wifi/4G antennas? |
| 08:18 | jnettlet | topi`, if you are looking for more advanced networking running on a Debian base. We are working with the VyOS project. |
| 08:19 | jnettlet | we do have a metal case. it has 4 antenna punch outs...but it shouldn't be hard to customize it. |
| 08:21 | jnettlet | but for VyOS our hardware inspired their first ARM32 port of the distro. |
| 08:22 | topi` | cool |
| 08:22 | topi` | we "kind of " compete with VyOS :) |
| 08:22 | topi` | int he sense that we have implemented ourselves functinality on top of debian |
| 08:27 | jnettlet | ah, I didn't know you guys were doing advanced networking/router functionality |
| 08:29 | topi` | jnettlet: not really, but many customers ask for some tailor-made networking things, and we try to make everything possible :) |
| 08:30 | topi` | many have some specific needs, like routing UDP traffic from a camera from point A to B |
| 11:02 | heap_ | hi |
| 11:02 | heap_ | jnettlet: pls so is there still some kernel+linux os for old cubox prov1? |
| 11:36 | jnettlet | heap_, are you having a problem with the upstream kernel? Or is there functionality there that is missing other than A/V? You only use the Cubox as a NAS right? |
| 11:47 | heap_ | no i bought regullar x86 pc |
| 11:47 | heap_ | jnettlet: i found old cubox prov1 and i want to use it as a offsitebackup via rsync |
| 13:20 | jnettlet | heap_, okay so just network and storage. Then the upstream kernel should have full support for that. |
| 13:20 | jnettlet | Sorry I had a lunch to attend to. |
| 13:20 | jnettlet | are you having build problems with the mainline kernel? |
| 14:30 | heap_ | jnettlet: are we talking about old cubox device? |
| 14:42 | jnettlet | yes |