Carputer 2015 - System Design

The carputer design is coming together. I've collected a decent set of system components, cabling, connectors, and tools that will be useful for prototyping. System design has been a bit up in the air as I'm coming to understand the limited selection of components and their constraints. In particular, finding acceptable display panels and controllers has been a problem.

The following is my current iteration on the overall design.

Car outline clip art image by Cliparts.co. Diagram produced using Inkscape.

1. Vehicle battery.
   Note the new cable run through the engine firewall to the rear of the vehicle. This will allow maintenance of a low-power sleep mode to quickly bring the master carputer online when the vehicle is turned on or a remote network query is made.
2. Vehicle system bus and accessory power distribution module.
   This represents a magical (to me) black box that provides 12V accessory power and vehicle systems signaling.
3. Vehicle A/V control unit.This module processes input and output for non-essential vehicle systems, such as climate control and audio. This is designed to interface with a display unit that will be entirely replaced by new components for this project.
4. Flat panel touchscreen display and controller module.
   The existing LCD panel is 8 years old and the backlight is weak. Readily available panels in the 7" size seem to go from 800x480 to 1280x800. For this application, panel visibility in a variety of lighting conditions is much more important than resolution.
5. Legacy display signal conversion and digital capture module.
   The A/V control unit generates what appears to be a 15kHz RGBs video output (like an old arcade game). This will be converted to composite video and then digitally captured with a USB dongle. The master carputer needs to be able to process and display these video frames.
6. Legacy wiring interface microcontroller.
   This is a controller like an Arduino that interfaces signals from the A/V wiring harness to USB. The master carputer will then be able to determine things like vehicle speed and reverse state.
7. Powered USB hub.
   This supplies the power and data bus for components in the front and center consoles of the vehicle. The uplink is wired to the master carputer host at the rear of the vehicle.
8. Switchable DC to DC power conversion module.
   The master carputer requires a stable DC power supply, possibly with multiple voltages and possibly with battery backup support.
9. Master carputer.
   This is the main computer that handles all USB I/O devices and generates audio/video output to the front display over HDMI. The carputer setup is as yet unspecified. This could be a repurposed coreboot plus Linux Chromebook, an embedded PC, or a Raspberry Pi. Potentially the frontend (audio/video) and backend (input processing and control) could be separate computers.
10. Rear-view camera module.
    This vehicle did not originally come with a backup camera. A camera will be added to provide rear-facing video to the master carputer for display or recording.

Notice that the long cable runs between the front and rear of the vehicle use digital signaling. I had originally planned to run composite video from the rear camera and analog audio from the master carputer to the front of the vehicle. The flatscreen display module I'm most strongly considering now only has HDMI input and breaks out HDMI audio to a D/A converter. Display controllers often have a board with switches to select between inputs and configure the display like a computer monitor. In the carputer case, fewer switches might be a good thing. I also like the idea of using only digital signals over distance. Solving glitchy or noisy analog audio/video problems is no fun at all.

Carputer 2015 - Motivation and Goals

Background

I've long wanted to do an in-car computer system but the timing was never right. Recently I decided that my midlife crisis car will be the car I already have: a 2007 Infiniti G35x sedan. Being willing to invest time and money into this car means I can finally have the carputer I want. Interestingly the carputer I want now bears only a passing resemblance to the carputer I wanted 5 years ago.

Mobile Device Integration

In-dash navigation used to be a really big deal. The factory Navigation Package including a 7" LCD touchscreen and Sirius XM Radio was a $2,100 option for my car. Needless to say I'm happy I declined. Now that mobile touchscreen devices are so ubiquitous, why would anyone want to be locked into an in-dash system that works differently and more awkwardly than one's own smartphone or tablet? The industry has clearly figured this out with integration platforms like Android Auto.

Starting Points

This is my car and its sorry 7" non-touch LCD screen. Status display is 15 kHz RGB video, just like old '80s arcade games! I still need access to status information but this washed-out and washed-up LCD needs to go.

The original Bose audio system is still good enough to my aging ears. I'm planning to integrate with the existing audio instead of replace it.

Philosophy and Goals

Having a software and systems background, I like to go into new projects with a set of guidelines. Here's what I've come up with for this carputer:

• Incrementally add and augment features such that at no point are existing features lost.
  e.g. Kickass visualizations aren't as fun if there's no way to know cabin thermostat settings.
• Integrate rather than re-create. Follow the mobile device integration philosophy.
• Carefully plan the physical installation.
  Do I really want to open up the dash and interior panels more than, say, twice?
• Be mindful of how and when components need to be powered.
• Run multiplexed buses over distance rather than individual signals.
• Be mindful of component placement and possibilities for modularity.
• Maintain a professional look when the vehicle is turned off. Hide wires and dongles.
• Start cheap and be able to justify increases to the BOM.
• Prototype outside of the vehicle until a minimally acceptable feature set can be achieved.
  Yeah, this probably isn't as fun when it comes to posting progress pictures.

The following is my set of non-negotiable requirements:

• The main computer shall run a Linux kernel. I understand this environment.
• There must be a highly visible touchscreen within usable distance from the driver.
• As a new feature, backup camera functionality must be integrated.

Continuation

I'm in the process of ordering parts for initial prototyping. I've tested a few system components to make sure they work and will satisfy requirements. I intend to continue posting status updates here on my blog and on Google+. It would be great if this becomes something of an open project with a degree of collaboration. You let me know what does/doesn't work and I'll let you know. Please let me know when I'm being stupid; I can take it. I'll attempt not to dwell on deep specifics of my vehicle so as to be useful to other carputer enthusiasts. Any generally useful software I develop will be made available, although I can't yet guarantee it will be open source or free.

Taking Control of Your Wi-Fi

Ubiquiti UniFi APs and Controller Software on Linux

Background

For years, I bought Wi-Fi routers and access points from the likes of Linksys, Netgear, D-Link, and Asus. Warranties would immediately be voided as I loaded alternative, Linux-based firmwares onto these devices. I started with DD-WRT before moving on to Tomato-based firmwares and OpenWrt. Tomato by Shibby is my choice for Broadcom chipset devices. OpenWrt supports a wide variety of hardware and has DIY geek appeal.

Gradually, I came to expand my network of Wi-Fi APs to cover the entire house, immediate yard, outbuildings, etc. If you're in an apartment or crowded suburbia, absolutely get yourself a quality device that supports 5GHz, like the higher end models from Asus. It's really crowded out there on 2.4GHz, so only use that as a fallback. Locate your single 5GHz device somewhere central and (hopefully) enjoy the speed and lack of hiccups.

On the other hand, some of us want to expand our coverage to several acres. No dead spots in certain corners of the house. Decent Wi-Fi coverage while out in the garden or mowing the lawn. Minimum lot size in my town is two acres, and it's not unreasonable to want that (or more) covered. This desire for increased coverage and easy management of multiple APs led me to try the UniFi products by Ubiquiti.

Dumb APs, Smart Control

Ubiquiti addresses the management and cost concerns of multiple APs by dumbing down the AP. The UniFi APs are simple, don't directly support web-based management, and don't really have the resources to run (comparatively heavyweight) Linux firmwares. What these APs do support is Java-based management software in the form of the UniFi Controller. The cross-platform nature of Java allows this software to run on Windows PCs, Macs, Linux boxes, and so on. You can fire up the software once to get things configured, and then not really worry about it. Alternatively, you can run the controller 24/7 to provide monitoring and captive portal functionality.

Now that I have a decently low-power yet powerful box on which to run this, in the form of a coreboot Chromebox, it's time to make the UniFi Controller run as a service on Linux. Ubiquiti somewhat supports this, but it still takes effort and research to make it right. For the benefit of myself and others, I'm going to document everything in one place.

Minimum System Requirements

Resource requirements are not especially light, considering you'll be running Java and MongoDB.

• 2GiB RAM
• 10GB storage
• Single 64-bit x86 processor core

In theory, other architectures should work, but don't expect somebody to have created packages or compiled binaries. Given RAM and CPU requirements, I don't recommend planning to deploy on an embedded ARM platform like the Raspberry Pi or BeagleBone Black. If you do manage to get this running on an alternative platform like ARM or MIPS, please let me know!

Selecting a Linux Distro

Ubiquiti appears to support Ubuntu and Debian best, providing repositories from which to obtain packages. The software itself is also available in a ZIP archive, for those who are pledged to another distro and willing to put in some work. I'll be documenting the procedure for Debian 7 (wheezy) here. Debian a good, solid choice for this. If you're able run this all in a virtual machine, I highly recommend it.

Installing and Configuring the Software

We start with a relatively stock Debian 7 install. I accepted most of the defaults, deselecting the desktop and selecting ssh and standard utilities package sets. Configure networking as you'd like it once the install has completed.

We'll add the apt repositories for Ubiquiti and MongoDB.

apt-key adv --keyserver keyserver.ubuntu.com --recv C0A52C50
echo "deb http://www.ubnt.com/downloads/unifi/distros/deb/wheezy wheezy ubiquiti" >/etc/apt/sources.list.d/ubiquiti.list
apt-key adv --keyserver keyserver.ubuntu.com --recv 7F0CEB10
echo "deb http://downloads-distro.mongodb.org/repo/debian-sysvinit dist 10gen" >/etc/apt/sources.list.d/mongodb.list
aptitude update

Let's disable startup of the default MongoDB instance in advance. It's going to want to pre-allocate multiple GB of journal files, so best to avoid that.

echo "ENABLE_MONGODB=no" >>/etc/default/mongodb

Now we'll install all the packages, watching the UniFi controller fail to start. This is due to the unifi init script setting an outdated JAVA_HOME path. We can fix that in the init script or create a symlink. I'm going to recommend the symlink so we can safely upgrade the unmodified package init script in the future.

aptitude install unifi
ln -sf java-6-openjdk-amd64 /usr/lib/jvm/java-6-openjdk

Just to be safe, we can make sure services are stopped and database journal files are removed.

service unifi stop
service mongodb stop
rm -Rf /var/lib/mongodb/journal
rm -Rf /var/lib/unifi/db/journal

Now we'll edit UniFi controller properties to disable DB journaling, then start the service.

echo "unifi.db.nojournal=true" >>/var/lib/unifi/system.properties
service unifi restart

Now point your browser at http://<ip-address>:8080/ and accept the certificate to access the software interface. I hope you found this useful. Please let me know if you have problems or improvements.

Last Year's Phones: Nexus 5 and Moto X

The Lay of the Android Land

These phone comparisons have been done to death, so I'm going to come at this from a different angle. The touchscreen smartphone market has matured. In the Android space, the new Samsung Galaxy S5 and HTC One M8 are both rather incremental improvements over their predecessors. With unlocked prices well above USD $600 for these new flagships, one might reasonably consider paying half that amount for one of last year's finest.

Personally, I'm a stickler for a near-stock Android experience. Samsung provides a hodgepodge of their own multiple UI efforts along with the mandated Google pieces. Other vendors provide their own enhanced experience. All of this equates to lock-in. After many years of Apple's iDevice, iTunes, App Store lock-in, I refuse to go back. Thankfully, we have our Google Nexus branded devices, Google Play Edition devices, and soon Android Silver. Google's acquisition of Motorola Mobility aided the demise of the Motoblur UI in favor of a value-added, near-stock Android experience. Let's hope Lenovo, as the new owner of Motorola, stays the course.

Anyway, enough background. As an early adopter of the LG-made Nexus 5, I'm now transitioning to the slightly older and lower-spec Moto X. My home's metal siding and roof make for very poor indoor signal. Republic Wireless provides a clever solution, and their customization requires a Moto X or G. Thankfully, prices on the Moto X have dropped significantly since its introduction. Both the Moto X and Nexus 5 are in the $300-400 range without contract. Now, on to the phones.

Aesthetics and Philosophy of Use

Nexus 5 (left) and Moto X (right)

With its 4.95-inch display, the Nexus 5 is a somewhat larger device than the 4.7-inch Moto X. Some have critiqued the N5's "lack of style." To this, I must respectfully disagree. The Nexus 5 is a debadged, dechromed, monochromatic slab of a device. This in itself is a style statement. This is a device that exudes power without ostentation, making some competitors look like they're overcompensating. The only other distinguishing features are the the circular ear hole speaker and the unfortunate bezel "chin" extending below the display. The slightly rubberized back are sides are nicely grippy, if a little bit wide and angular in the hand.

The Moto X, on the other hand, has an understated class... and sexy curves. Its slightly rubberized back has a simulated carbon fiber look. The curvature not only feels perfect in the hand but is practical as well, encasing a stacked battery. The front glass extends to the sides, making them somewhat more slick than the N5. My only complaint would be the slightly too obvious microphone hole in the lower left of the front glass. If the N5 is a muscle car, the X is a sleek luxury coupe.

Upon use, the relatively minor size difference between these two devices appears to magnify. The ~5-inch display and full HD resolution bring the N5 into borderline phablet territory. Phone apps can feel like they're underutilizing the display. As an example, the Amazon store app feels limiting in a way that it just doesn't on the Moto X. The N5 is crying out for the ability to choose between phone and tablet apps. Thus, in the current state of the Android ecosystem, the Moto X is great as a phone and the Nexus 5 is a near-phablet that can't run tablet apps. Here's hoping that Google figures this out. This issue may not be obvious to users until they've had their phone for a while.

Display Comparison

Aside from the 0.25-inch size difference, display resolution and technology contribute to the perceived differences mentioned above. The N5's 1080p IPS LCD display makes small text and detail highly visible. Extended periods of reading are not a problem. The front glass is a perhaps a little more reflective than ideal, and the backlight can only get so dim for nighttime use, a typical issue with LCDs.

The Moto X sports a 720p AMOLED display. Text is less clear than on the N5, partly owing to the lower resolution and partly to obvious color fringing around object edges. It's not clear to me whether this is more an artifact of intentional anti-aliasing or the nature of the AMOLED subpixel matrix itself. The AnandTech review of the Moto X is informative.

Thus, the Moto X is definitely no e-reader. That all said, AMOLED has some benefits that are put to good advantage by Motorola. Blacks are black (no backlight), and display power draw is proportional to the number of subpixels lit. This sets the stage for Motorola's exclusive Active Display feature. With the display off, motion sensing and software are able to detect when the user has picked up or unpocketed the phone. The display then kicks in to show the current time and notification updates in white on an otherwise dark background. Just pick up or nudge the device and you'll get status without touching a single button or tapping the display. This is both useful and seriously cool.

Voice Controls: Ok Google (Now?)

Both the Nexus 5 and Moto X allow voice control of a number of phone and search operations. With the N5 switched on, unlocked, and at the home screen, the Google Now Launcher will respond to a spoken command of "Ok Google" followed by a directive or query. This experience is awkward to say the least. Once one has pressed the button to wake the phone and exited the current app by touching the home icon, is it really helpful to speak to the phone instead of just use it?

Once again, Motorola adds a feature that addresses real world use cases. The Moto X contains a subprocessor that is always listening for the phrase "Ok Google Now." It's necessary to train it, and perhaps the additional "now" was added to further prevent this feature from triggering accidentally. Still, this turns voice control from a novelty into a truly useful feature. I'm even finding myself saying "Ok Google Now, launch app-name" as I reach to pick the phone up from the desk, saving precious seconds. Can you tell I like this feature yet? I'm hoping this works decently with background noise in the car.

Miscellany and Conclusion

Cameras on both phones are passable, not great. It's a pity that so many OEMs are cutting costs and corners on sensor elements, optics, and imaging software/firmware. There are many reviews and comparisons out there with images taken from these and competitive phones.

Processing performance on both the N5 and Moto X seems entirely acceptable. They are subjectively faster than the 2013 Nexus 7. Benchmark results are available out there. The Moto X loses the spec war by having only two general purpose processing cores versus the more common four (or eight) in today's flagship devices. I can attest that browsing with Chrome, an activity that benefits from parallelism, does appear faster on the four core N5. For other activities it's kind of a wash. Looking purely at specs yields little practical information, as mobile processing is severely limited by power and cooling constraints.

Battery performance is somewhat low on both of these devices. The Moto X should do slightly better than the N5, but I don't have enough data yet to draw any conclusions. I'd generally accept a little more thickness and weight on all fixed battery devices to reduce mid/late-day charging hassle.

To draw some final conclusions, the Moto X is a great last-gen phone with some very unique features. Discounts have allowed this device to stay appealing despite it's age. The LG-made Nexus 5 is a very capable near-phablet with great potential. One hopes that future firmware updates will permit the N5 to come closer to realizing the potential of its hardware.

BeagleBone Black Exploration

The BeagleBone Black is an inexpensive ($45 US), ARM processor based hobbyist/developer board that runs Linux. This board is similar to the better known Raspberry Pi, enough so that's it's hard not to compare them. The BBB employs the embedded control oriented Texas Instruments AM3359 Sitara SoC, while the RPi employs the video oriented Broadcom BCM2835. I have both boards, so I'll definitely be comparing them.

I ordered the full BeagleBone Black (Rev A5C) starter kit from Adafruit including the 5V power adapter, prototyping breadboard and backplate and jumper wires, and "cape" PCB. I also recommend a microSD card to boot/install alternative Linux OS distros and a microHDMI cable/adapter and USB keyboard for debugging networking configuration.

BeagleBone Black and ChronoDot RTC on Prototyping Breadboard

The BBB comes standard with Ångström Linux on its onboard 2GB eMMC flash storage. Nod to the BBB over the RPi here. The Raspberry Pi requires a prepared SD flash card to get started.

For my purposes, I immediately loaded Arch Linux ARM onto a microSD card, booted from SD, and installed Arch onto the eMMC. I have to give a lot of credit to the developer(s) who produced the Arch image: it's nicely pre-configured to bring up wired Ethernet and SSH using DHCP. Arch thus allows a completely headless install and configuration. Alas, I messed up static IP configuration and had to attach a display and keyboard to work through it.

During the installation process, I felt motivated to benchmark both the eMMC and my microSD card (ADATA 32GB UHS-I).

Device	Read MB/s	Write MB/s
2GB eMMC	21.2	3.9
32GB μSD (BBB)	4.3	1.7
32GB μSD (PC USB2)	19.6	19.2

The SD interface is incredibly slow. The same card performs tremendously better on my PC monitor's USB2 adapter. I wouldn't want to run the OS from SD anyway, as reliability is always questionable. The microSD interface is clearly intended primarily as a means to load alternative distros onto the eMMC. Even using an SD card for a data/media filesystem takes a bit of effort.

Once running from the eMMC, the BeagleBone Black feels subjectively quick, certainly quicker than the Raspberry Pi. This is no surprise. The TI Sitara contains a more recent, higher-clocked, superscalar processor core with DDR3 RAM support.

My intended use for the BBB is as an internal network services (DHCP, DNS, NTP) and home automation controller. Like the RPi, the BBB doesn't have a battery backed real-time clock (RTC). I easily added one via the ChronoDot RTC and Lemoneer's excellent guide. The BBB has no lack of I/O and should be awesome for monitoring and control. There are even two dedicated PRU cores for true real-time control. I haven't explored this yet, but the PRUs may eliminate much of the need to attach Arduino microcontrollers a la the Raspberry Pi.

For my purposes, the BBB looks to be solidly better than the RPi. That said, the RPi still has some key advantages. There is a huge Raspberry Pi installed base and community. The RPi camera module is appealing. The RPi is decidedly superior for graphics and especially video tasks like XBMC. At these prices, there's something to be said for having one (or more) of each.

A Night and a Day on Backup Power

Power Management for Home Computing and Networking

In this installment, I discuss strategies by which computing enthusiasts may reduce power and retain or improve availability of home network services.

TL;DR TBD ;-)

Low Power Computing

There's never been a better time for low power computing. A lot of us are optimizing electricity use, not just because power is expensive, but because we've become mindful of waste. Smartphones, tablets, and ultrabooks use a lot less power than desktop PC setups. Mobile devices train us to be aware of power use. We can't always plug in, and it's inconvenient to plug in, so we optimize and prioritize around battery life. For those of us geeks who run home servers, more networking than a single WiFi router, whole house DVR backends, and home automation, it may be time to apply our power management skills to the home. I have some additional motivation: disruptive power outages that take my home off-grid, ranging from momentarily to minutes, hours, or several days.

On-Demand Battery Backup

I've owned a number of small UPSes from APC and CyberPower. These are a great help for those whose power is less than reliable. For many, this is all the power protection they need. Check out the recent "line-interactive" models with "active PFC". These are typically the best match for modern electronic devices where UPS cost is an issue.

If your equipment is mission-critical, especially sensitive, or you need to run on dirty power (as from a generator), you should be considering a "double conversion on-line" UPS. These units rectify current to charge batteries and feed an inverter that produces stable AC output. On-line UPSes are more expensive and less efficient than their line-interactive counterparts, but they serve their niche.

Generating Electric Power

If you live in remote or disaster-prone area, you may be considering the ability to generate your own power. You may be considering a photovoltaic solar installation even if you live in a center of civilization. Make no mistake, generating reliable AC power is a challenge. Some small or remote installations may be better off sticking to DC power and equipment. I won't write more about DC here, but may blog about in the future if I get around to implementing my dream DC system.

Typically, generating your own power means getting a generator setup that runs on fossil or bio-fuel. Good generators are expensive. Around here, some people spring for whole-house standby generators with automatic control systems and transfer switches. These setups usually run on natural gas or propane and cost $10K USD or more. Many more people have smaller "portable" generators that they backfeed into house power, a manual transfer switch, or just replug equipment into the generator itself. If backfeeding, it's critical to turn main line power off at or before the breaker/fuse box.

My generator is a 6500W (max) tri-fuel (gasoline, natural gas, or propane) unit that cost about $2.5K. It's got electric start, but I need to manually engage the key switch, flip switches in the basement, etc. to bring it on line. The generated AC power isn't very clean. I've seen frequency range from 60-64Hz and voltage from 115-130V. If I had an oscilloscope, I could see the ugliness of the "simulated sine wave" output. AC motors will sometimes vibrate on generator, lights flicker, A/V equipment exhibit audio noise, and cheap UPSes beep and refuse to charge their batteries or pass generator power through.

Real UPSes That Work

I can deal with most of the downsides to dirty, generated power. It's temporary; most things work, just not optimally. UPSes that discharge during the initial outage then refuse to charge or pass generator power are a huge annoyance. You'd have to shutdown, unplug, and replug each piece of equipment on each UPS to switch to generator power, then repeat the process to switch back to the UPS! This somewhat defeats the purpose of having a UPS in the first place. It took me a few years to take the hint and solve the problem.

My solution was to purchase a double conversion UPS with extended runtime battery pack. My critical equipment plugs into this UPS and will continue to run on generator power. Non-critical equipment that would suffer from unexpected outages goes on cheap, line-interactive UPSes. This non-critical equipment either survives short outages or stays off until line power returns.

• SU1500RTXL2UA Double Conversion UPS
• BP48V24-2U External Battery Pack

I went with Tripp Lite instead of APC because the equivalent (SURTA series) APC UPS was literally $400+ more.

Data for a Case Study

So now that you've heard the rambling details of why and how I addressed my problem, here are the specifics from recent testing. This takes us to the title of this lengthy post. I spent the night with the UPS unplugged to see how long it would run my critical equipment. Much of the next day was then spent on generator to 1) ensure batteries would charge and equipment run, and 2) determine how long it would take batteries to charge from 10% to 100%.

Equipment consists of:

• Comcast Business Class network gateway
• D-Link DIR-825 running as dedicated firewall/router
• Asus RT-N66U running as WiFi access point and network switch
• Raspberry Pi running critical network services and monitoring
• Thin Mini-ITX Intel "Ivy Bridge" PC running as DVR backend
• HDHomeRun network-attached dual digital TV tuner

I don't have exact power consumption figures all devices individually or all devices together. The UPS won't guess loads under 100W, so I know this is less than that. My back-of-envelope guesstimate is an average draw of ~60W. The runtime figures bear this out.

The equipment ran for just over 12 hours on battery. I was hoping for closer to 16 hours, but there are optimizations left to try. The UPS batteries charged well from generator while equipment continued to run. During the just over 5 hours it took the batteries to charge from 10% to 100%, there was one 19 second interval where input power went out of spec. The generator kicked up to 64Hz and 130V during this time and engine RPM and generator whine were obviously increased. I have no idea why. Anyway, I'm happy enough with these results.

Power Optimization Tips and Tricks

So, how to get from 12 hours to my goal of ~16 hours on battery? Tripp Lite has a nice calculator. Basically, I need to stay closer to 40W than 60W. To PC gamers that probably sounds impossibly low. Well, good thing I'm not gaming on batteries!

Most obviously, functions can sometimes be consolidated onto a single device. For instance, I'm running a separate firewall/router and WiFi access point. Most home users don't do that. Ah, but there's a method to my madness! During some extended power outages cable internet will go out; not always at first, but eventually power backup goes out at the cable headend too. In these situations I want to be able to "shed" the load of my Comcast gateway and firewall/router. And sometimes I just want to be able to power cycle them without taking down my internal network.

So separating and assigning related services that fail together into power/availability groups is sometimes useful. I have an "internet connectivity" group and a "home networking" group. Now, how to lower overall power use?

My PC-based DVR backend and TV tuners pull a decent amount of power. Yet when it comes down to it, DVRs only need to be powered up when they're being used.

Wake-on-LAN/Timer to the Rescue

Some PC-based equipment simply doesn't need to be fully powered up at all times. Modern PCs can generally be put to sleep (suspended to RAM for reduced power draw and quick resume) or hibernated (suspended to disk for no power draw and slower resume). Failing that, systems can be shutdown and booted back up when needed. This is all possible because modern PCs can keep a trickle of power going to a clock and system/network controllers while the PC is still plugged in.

Think of this like your TV. When the TV is off, it's really in "standby". A sensor and controller are powered and waiting to detect your press of the "on" button on the TV remote.

Let's go back to my PC-based DVR example. If this system drew a lot of power, it might make sense to keep it powered off until it's needed to record or stream content. Since my particular system only pulls 15-45W, it's more convenient to keep it on and immediately available. That equation changes when I'm running from battery. This calls for a little automation:

1. When line power is available, power on, stay on, and run normally. Done.
2. When running from battery and the system is idle, set the wakeup timer to just before the next scheduled event. Go to sleep or hibernate.
3. When the pre-set timer goes off, wake up if sleeping or hibernating.
4. When receiving a magic network packet, wake up if sleeping or hibernating.

The "magic packet" allows another computer or device to wake or boot an otherwise unavailable system from a standby state on demand. There are a number of utilities available to send magic packets through a manual UI or scripted interface. Some systems also support the ability to wake on keyboard/mouse activity or selective USB activity. These are very cool and underutilized features outside the world of portable PCs.

Power Control Hacks

I also mentioned that I'm running a network-attached digital TV tuner. Logically, this is part of a "DVR" power/availability group. The tuner box only needs to be powered when the DVR backend is recording or streaming live TV. Unfortunately, the tuner box doesn't support power management. It's either plugged in and on, or unplugged and off. Some devices are like this.

A little research shows that the HDHomeRun tuner runs on 5V and under 2A. My PC-based DVR already contains cables that can supply the proper voltage and current! Let's say that I hacked a SATA power cable to provide 5V at 2A to a barrel connector that could be plugged into the tuner. Now (in theory) I have a tuner that can power off and on as the PC sleeps and wakes.

Software/Firmware Power Tuning

Intel, AMD, and the various ARM vendors have gotten pretty good at making the hardware and firmware recognize demand and selectively power up/down functional units and communications buses. This used to be done with software (if at all). Still, there are sometimes power saving drivers to install or driver options to tweak to save a little juice. A blunter approach is to disable (in the BIOS/firmware) or remove devices that aren't needed. Underclocking and/or undervolting CPUs, GPUs, or RAM is also an option.

Consolidation and Virtualization

As I stated before, sometimes combining and consolidation is detrimental to sensible grouping and availability of services. Oftentimes, consolidation is brilliantly appropriate. System and container virtualization can make consolidation even more applicable. Let's say I want to add a software PBX like Asterisk to my mix of services. I could easily install this on the DVR system, preferably under virtualization to isolate it from the DVR system image. The PBX would go up and down as the DVR system wakes and sleeps during outages. That said, my voice-over-IP (VoIP) provider already forwards calls to my cell phone when my PBX is down. This makes the PBX a useful but not critical service. And there are potentially ways to proxy voice service and wake the PBX system on incoming calls.

Roll Credits

If you made it this far, I admire your fortitude. Believe me, I wasn't planning to write such a voluminous tome. As always, questions and comments are welcome via Google+.

Repurposing the HTPC

Several month ago, I built a slim form factor Windows 8 PC to act as a media streaming frontend. Notes here:
http://travesh.blogspot.com/2013/03/super-slim-intel-htpc-build.html
http://travesh.blogspot.com/2013/03/super-slim-htpc-followup.html

Total overkill as an HTPC for everything but games, yet I hadn't found a better solution to stream MythTV recordings, Netflix, Amazon Instant Video, etc. to single frontend. Android solutions were flakey or slow. Netflix uses Microsoft Silverlight, so good luck getting that on Linux. After a while, I added Aereo and Plex to my streaming requirements. Windows 8 on a small PC can do all of this, but it can't do it like an appliance with a unified UI. I bought a Roku 3 and the PC sat for a bit.

Having torn down my big MythTV backend when the summer TV doldrums went into effect, I'd been starting to feel the pressure to put MythTV recording back into service. Aereo's cloud DVR has mostly worked for me, but not 100%. Fundamentally, their service is a legal and technological hack; I'll enjoy it for now, but can't totally trust it.

So the now unused PC frontend has become the new DVR backend. The SDD got replaced with a 500GB 2.5" 7200rpm drive for video storage and Windows 8 got wiped in favor of Mythbuntu. My goal this time around is low power recording. When I lose electric service (as I often do), I want my DVR to keep recording on UPS and/or generator. There's also just the cool factor of the low power x86 and ARM trend. Everyone is starting to care at least a little about heat and power.

All said and done, the new DVR backend idles at 16W. That's insanely good! And this is Ivy Bridge, so who needs Haswell? And especially, who needs Atom? My old Atom 330 always idled at about 35W. The later Mini-ITX Atom boards were better, but never great. ARM isn't quite there as a DVR backend; disks over USB, 10/100 Ethernet on most boards, weak transcoding speed and support. I was planning this elaborate sleep/wake configuration for the Myth backend, but I might not bother. Always-on is just easier.

I'm planning my next blog post about separating the DVR recording backend from transcoding, archival storage, and streaming. The slim form factor Mini-ITX Sandy, Ivy, and Haswell get a huge thumbs up from me as a near ideal solution for DVR recording. And at 16W idle, you may want to run some other always-on services from this system as well. Power tuning is still in progress. How low can we go?