tech

2017-06-27 · open the original ↗

• A VoIP door station's relay simply interrupts the door strike's power. The host, who says he is not an electrician, explains that a device such as the Axis A8105 network video door station or a CyberData keypad unit opens a door or gate by closing a circuit when a DTMF code is dialled from the phone that answers its call button.
• Whether a strike locks or unlocks without power decides the wiring. Strikes are rated NO (normally open, unlocked without power) or NC (normally closed, locked without power), and some can be toggled; the host's NC strike stays unlocked only while power is applied, which he demonstrates by touching the red lead on and off.
• Splice the relay's CO and NO/NC contacts into one power lead. He cuts the red wire from a plain wall power adapter, splices it to light doorbell wire running into one relay contact, and runs the other contact back to the strike, leaving the black wire direct; he thinks it does not matter which of the two contacts is used. Disconnect power and PoE before wiring.
• The relay holds the door open only for a short timer. In his demo, dialling 99 (his usual code) unlocks the strike for about seven seconds before it relocks; he says five or seven seconds is typical, and that the same green relay terminal block appears on other CyberData gear such as the SIP paging server, equipment he says he sells to customers regularly.

2017-06-23 · open the original ↗

• Mantis is a tap-to-pair Bluetooth sensor kit for classrooms. Walker of HipScience, met by Hanselman at a conference break in Atlanta, describes it as a Bluetooth Low Energy IoT platform for education with a cross-platform app; he pairs a climate sensor by tapping it and shows light, temperature and humidity updating in real time as he covers the light sensor.
• It is a family of interoperating sensors, not one device. He demos a motion device reading gyro, accelerometer, velocity and real-time distance from a wall "like a ruler", and a general-purpose sensor whose plug-in probes cover temperature (shown climbing from 24 as he holds it), pH, conductivity, turbidity, soil moisture and current, with what he calls a very fast sampling rate.
• A mesh gateway relays the data to Azure without the tablet. Per Walker, the devices form a Bluetooth 4 mesh network whose wall-powered gateway speaks MQTT bidirectionally to Azure IoT Hub, feeding Power BI and Stream Analytics dashboards; the app runs on iPad, Android and Windows, including a Universal Windows app using Windows 10's tap-to-pair. He points viewers to mantisopenstem.com.

2017-06-13 · open the original ↗

• The EdgeRouter Lite's weak point is its USB-stick operating system storage. The host, who rates the roughly $91–92 router one of his favourite firewalls, says corruption is not common but not unheard of; the one he recovers was a client's, corrupted in an electrical storm. EdgeOS lives on a removable USB stick inside, reached by removing two case screws.
• Recovery starts from the serial console and a TFTP server. He connects a USB-to-RJ45 console cable (the kind shipped with FreePBX systems), finds its COM port in Device Manager and opens PuTTY at 115200 baud; the three files of Ubiquiti's "last resort recovery" EMRK image go on any TFTP server on the network, in his case his FreePBX system.
• Without DHCP option 66, point U-Boot at the TFTP server by hand. With a cable in eth0, he runs dhcp (or sets a static IP and netmask), then sets the server IP and the EMRK .bin boot file name, runs tftpboot, then bootoctlinux $loadaddr; the documented one-line command works only if DHCP hands out option 66.
• The reinstall must fetch EdgeOS over plain HTTP, not HTTPS. In the recovery software, emrk-reinstall reformats the USB stick and asks for an image URL; he copies the download link for his model from Ubiquiti's site and changes https to http. The download and processing took about 60 seconds before a reboot to a ubnt/ubnt login.
• Protect the router from power events once it is restored. After re-running the setup wizard from the default 192.168.1.1 and moving the LAN off that range, the host recommends a surge protector, battery backup if possible, and the rear grounding screw. He links an Amazon affiliate link for buying one, which he says earns him a referral fee.

2017-05-30 · open the original ↗

• The UAP-AC-HD is Ubiquiti's access point for very dense environments. Against the UAP-AC Pro's 3x3 single-user MIMO (802.11ac Wave 1), the HD has 4x4 multi-user MIMO (Wave 2); the host found the box and the downloaded data sheet disagree on radio rates (800 vs 600 Mbps on 2.4 GHz, 3.4 Gbps vs 1,733 Mbps on 5 GHz), and he could not say which is right.
• Size the HD by Ubiquiti's recommended user limits, not theoretical ones. The HD is rated for 500 users theoretical and 250 recommended, the Pro 200 and 70; MSRP is $349 against $149. The host would put one HD per movie theatre and use HDs in a football stadium, but expects AC Pros to suffice for a hotel. He links Amazon affiliate listings for the units.
• The HD also suits a mesh uplink serving several mesh APs. The host says its 4x4 MU-MIMO can handle several larger mesh access points uplinking wirelessly through it at once, giving far more total throughput than an AC Pro in that role.
• Wave 2 MU-MIMO serves clients simultaneously; Wave 1 serves them in turn. The host's analogy: spatial streams are highway lanes and devices are cars of 1x1 to 4x4 width; under Wave 1, even a 3x3 AP sends to one device at a time, whereas the HD's four lanes can carry, say, four 1x1 phones, two 2x2s, or one 4x4 uplink simultaneously.
• Adoption needed a manual set-inform over SSH to a cloud controller. The discovery tool would not send set-inform to the host's Amazon-hosted controller, so he SSHed in with default ubnt/ubnt credentials and ran set-inform http://<controller>:8080/inform twice. He runs controller 5.4.14 against the latest stable 5.4.16, staying a version or two behind to protect live clients; the HD drew 6.61 W idle, and he did not throughput-test it, lacking dense-client test gear.

2017-04-27 · open the original ↗

• Paging Pro adds features; basic paging is already free. Standard FreePBX lets you create unlimited page groups without the commercial module. The host puts Paging Pro at $95 for a licence covering the whole server, plus, he thinks, annual maintenance of around 10–15% of that price to keep receiving updates, and he judges it good value for what it adds.
• Busy page groups queue pages through valet-style record-then-play paging. The caller records the page and hangs up before it plays, as in noisy airport terminals, which avoids feedback. It can be set to always use valet or only when the page group is already in use, playing the queued page when the current one ends; a page announcement (e.g. "attention everyone") can be prepended.
• Multicast is the module's most important feature. Without it, paging 50 phones means 50 separate calls, which taxes the PBX CPU and can degrade call quality on calls in progress. Multicast sends one stream to an IP address that all the phones listen on, and the host's rule of thumb is to buy Paging Pro once a page group exceeds about 10 phones, depending on server size.
• The scheduler turns the paging system into a school bell. It sets a date range, days of the week, times and excluded dates such as holidays, and plays an announcement file (e.g. a buzzer) at each interval. Standard FreePBX paging does not include it.
• Dialling a chosen outbound route can trigger a page. The example is an emergency route: a 911 call from a warehouse phone goes through while supervisors' phones are paged with an announcement that someone dialled 911. The host says he has never used this feature himself.

2017-04-20 · open the original ↗

• Judge critical decisions through the whole company's eyes. Horowitz's single lesson is that a decision must be weighed from the viewpoint of everyone it touches, including people not in the room. He calls it the mistake CEOs make most consistently at every stage. His advice under pressure is to pause and say you will come back once you have thought it through.
• Demoting an outside executive usually fails the company's sense of fairness. In a case he advised on, a hard-working but out-of-depth sales head held 1.5% equity against engineers' 0.1–0.2%. Demoting him would leave that stake intact or force taking it back, and would cost him the team's respect. When firing, he says, be honest that hiring was the mismatch, and keep the person's dignity, a lesson he credits to Bill Campbell.
• Grant raises through a formal periodic process, never on request. Giving a raise because a good employee asked signals to quieter, possibly better performers that only askers get paid. It also makes asking feel like a duty to one's family. His fix is a review every six months or quarter that gathers everyone's input, with no off-cycle changes and no lobbying afterwards.
• Ten-year option exercise windows deserve reconsideration, not universal adoption. He traces the 90-day window to accounting rule APB 25, replaced in 2004, under which long windows made earnings unforecastable. Weighing about 10% attrition, 6–10% annual dilution and the reward owed to stayers, he rejects Sam Altman's call for every startup to adopt Adam D'Angelo's policy. Either generosity or explicit reward for staying is defensible, if stated up front.
• Toussaint Louverture mastered this by considering even his enemies' views. Horowitz says Louverture made the best officers of defeated British, Spanish and French forces generals, a tactic he says came from Julius Caesar. He let planters keep their land but pay their workers, offset by lower taxes. Horowitz calls it history's only successful slave revolution and claims Haiti then out-exported the US.

2017-04-18 · open the original ↗

• Mesh suits many unwired access points across a large area. The host's example is a stadium with internet at one end and access points dotted around it; a small business usually does fine with a repeater or a standard UniFi access point in isolated mode. A mesh access point wired into the network is just an access point, gaining no mesh capability unless others chain off it.
• The AC Mesh takes external antennas; the AC Mesh Pro does not. Per the boxes, the UAP-AC-M is 2x2 MIMO, 867 Mbps on 5 GHz and 300 Mbps on 2.4 GHz, 183 m (600 ft) range, 802.3af PoE, and accepts any Ubiquiti 5 GHz antenna. The UAP-AC-M-Pro is 3x3 MIMO with the same range, 1,300 plus 450 Mbps, two gigabit ports and a 48 V injector; the host calls its combined "1,750 Mbps" figure misleading for any single device.
• Size the wired uplink access point so it doesn't bottleneck the mesh. His UAP-AC-Pro uplink matches one AC Mesh Pro's 1,300/450 Mbps, which is fine for one but not several. He suggests a UAP-AC-HD (800 Mbps on 2.4 GHz, 1,733 Mbps on 5 GHz) or other antennas, feeding a star of AC Mesh Pros with smaller AC Mesh satellites around them.
• Three wireless hops still delivered about 20 Mbps, which the host deems adequate. He measured 20.8 Mbps on iPerf3 and about 21 Mbps on speedtest.net, through trees at low signal, and calls the results anecdotal. That suits guest Wi-Fi, which he would cap at 2–5 Mbps, or a barn or boathouse; for more bandwidth he recommends a point-to-point bridge, which can do 400–500 Mbps or more.
• Self-healing failover must be switched on; it is off by default. The UniFi setting "enable automatic uplink failover" lets the controller pick a new uplink when one goes offline. When he powered down the AC Mesh Pro, the far AC Mesh went isolated and, after about five minutes, re-linked to the window AC Pro at 17% signal.

2017-04-07 · open the original ↗

• The US-8-60W is a cheap PoE gigabit switch for small deployments. At around $99 it has four standard gigabit ports and four PoE ports sharing a 60 W budget, which the host calls a lot of power for four ports, since by his estimate UniFi access points draw 2–3 W and a standard PoE phone about 5–6 W. The host links it via an Amazon affiliate link.
• Its PoE ports are 802.3af only, with no 24 V passive. Ubiquiti gear powered by 24 V passive, such as NanoStations, some cameras and lower-end access points like the UAP-AC-Lite, needs Ubiquiti's 802.3af-to-24 V passive adapter. The rear also has a grounding point, which the host recommends using.
• Adoption was automatic because the switch shared the controller's network. Otherwise the host says you must find its IP address and run set-inform, or use a Ubiquiti device-discovery tool on the same network. After adoption he upgraded it to firmware 3.7.49 on a 5.4-series UniFi controller running on a Cloud Key.
• The host traced a slow iPerf result to screen-recording overhead. An iPerf3 test to his laptop over a USB 3 network adapter gave about 866 Mbps while he recorded his screen and 941 Mbps without recording, repeatably; he suspects the disk writes from recording and is satisfied the switch delivers gigabit.

2017-03-15 · open the original ↗

• Header editing is a separate mode, and people get lost in it. Hanselman shows that inside the header the body text goes gray and the Header & Footer Tools tab appears only there; even Ctrl+A selects only the header's contents. Double-clicking moves in and out of the mode, and he says turning on paragraph marks helps you see where you are.
• A header is a live template shared by every page. Typing, changing case to uppercase or making text bold in one header changes it on all eight pages of his demo document at once.
• Position header items with tab stops, not by spacing. The blank three-column header uses a left, a center and a right tab; he says you will "never be precise" nudging text with spaces. Because the stops are part of the template, after switching to landscape he drags the center tab once and every header follows.
• Odd, even and first pages can each get their own header. The "different odd and even pages" and "different first page" checkboxes on the Design tab let him lay out book-style page numbers on alternating sides. He inserts page numbers at "current position", because the "top of page" option replaced the header customization he had already done. Footers work the same way.

2017-02-28 · open the original ↗

• Porting is carrier-to-carrier, and the losing carrier decides. The customer gives the new carrier a letter of authorization (LOA: the numbers, address and company details, plus consent) and usually a copy of the latest bill from the losing carrier; the new carrier submits the port, which the losing carrier accepts or rejects. The host notes that some very small rural carriers cannot port at all, so check that first.
• Losing carriers reject any imperfect request because porting loses them business. The host says they are obliged by law to port numbers but have no incentive to make it easy. In his experience about 75% of port requests are accepted first time, and most of the remaining 25% fail for the reasons below.
• An address mismatch is the most common reason for rejection. The address with the new carrier and on the LOA must match the losing carrier's billing address exactly; "Street" spelled out against "St." is enough to have a port rejected, so copy the address from the old bill.
• Bundles, account numbers and contracts each block a port. Phone service bundled with internet or TV may have to be split into its own account first. Where the main number doubles as the account number, add a temporary number as the account number before porting the rest. Under contract, keep one number with the old carrier until the term ends, and watch for automatic renewal at expiry.
• Ports happen on a set day with no time window. Once accepted, the losing carrier gives the new carrier only a date. The host advises having the new service fully configured beforehand and watching it closely that day so the numbers ring through correctly the moment they move.

2017-02-21 · open the original ↗

• Both ends of a fiber link must match in type, speed and fiber. SFP is 1 Gbps and SFP+ is 10 Gbps; the modules on each side and the patch cable must all be the same speed and the same fiber, single-mode or multimode. LC-to-LC is the standard patch-cable connector. The host demonstrates with 1 Gbps multimode between a UniFi Switch 8 150W and a 24-port EdgeSwitch Lite.
• Installing the link is plug-and-click once the parts match. Remove the rubber stoppers from the modules and switch ports and the white caps from the cable ends; modules fit only one way and click home. In his demo a persistent ping dropped when he pulled the Ethernet cross-connect and resumed as soon as the fiber went in.
• The host used 10Gtek modules only because Ubiquiti's were unobtainable. Ubiquiti's own SFPs had been released but were out of stock when he bought, and he says new Ubiquiti products typically take 2 to 3 months to become widely available. He paid for the 10Gtek modules, says they are not a sponsor, and found them about the cheapest available and reliable.
• Ubiquiti's own SFPs are probably the safest bet for Ubiquiti gear. He has not used them yet but believes they cost about the same as 10Gtek. For other third-party modules he points to Ubiquiti knowledge-base articles and forum posts listing compatible SFPs. His description links to both are Amazon affiliate links.

2017-02-07 · open the original ↗

• The host chose UniFi here for remote manageability, not as a default. The client, a new-build retirement home in Texas, had no local IT person able to manage the gear while he is in Oregon; UniFi lets him make changes himself at any hour instead of arranging remote-desktop sessions with someone on site. Price was about the same as the EdgeSwitch/EdgeRouter line, and he says he still picks per customer need and budget. He would not have trusted UniFi at this size before controller 5.3.8.
• He would now use UniFi fiber switches instead of EdgeSwitch XGs. Each of three closets has an EdgeSwitch 16 XG and two UniFi Switch 48s, with a USG Pro on Comcast Business, 26 UAP-AC Pros and a Cloud Key. The UniFi 16 XG was not supported until controller 5.4.9, still in beta when he ordered, so he used EdgeSwitch XGs with every port a trunk. He was disappointed the USG Pro's SFPs are WAN-only, forcing an Ethernet LAN link.
• The cabling contractor switched to single-mode fiber mid-project. Multimode was unavailable, so the 10 Gbps closet-to-closet links became single-mode while switch-to-XG cross-connects stayed multimode, leaving him with spare modules. Third-party 10Gtek SFP and SFP+ modules worked on every combination, bought because Ubiquiti's were out of stock; he rates Beyondtech patch cables well above his old ones. An iperf test across the single-mode link hit 823–833 Mbps, limited by the 1 Gbps client NICs.
• Three networks separate management, main LAN and guest traffic. Management is the native VLAN 1 on 192.168.1.0/24, reachable only from XG ports and ports 1–2 of each 48-port switch; ports 3–42 put devices untagged on VLAN 10, a 172.16 range far larger than needed with DHCP locked down; the last six ports carry access points, addressed on management and tagging VLAN 10 and guest VLAN 20, which is bandwidth-limited and secured with UniFi guest policies.
• Set access-point channels and transmit power manually in dense deployments. Using the UniFi map of the property's 24 placed APs, the host staggered channels on 2.4 and 5 GHz and set power to medium on 2.4 GHz and high on 5 GHz for good overlap without interference. He estimates the network will serve about 200 devices, up to 300 including Wi-Fi.

2017-01-03 · open the original ↗

• Ordinary page groups don't scale, because each page is many calls. FreePBX opens a separate channel to every device in the group at once, and since setting up a call costs Asterisk far more CPU than keeping one going, FreePBX caps page groups at 40 devices. The host says some lower-end servers cannot handle even 40.
• Multicast paging sends one channel that every device listens to. Asterisk opens a single channel to a multicast IP address and port, and the phones and paging devices subscribed to that address all play it. The host calls it "absolutely the way to go" for large paging deployments such as hospitals, schools and retirement homes.
• Any multicast address works if every device in the group matches. The CyberData paging server defaults to 234.2.1.1 on port 2000, then 234.2.1.2 on port 2002, and so on; Wikipedia lists 234.0.0.0/8 as unicast-prefix-based space and 239.0.0.0/8 as private-use under RFC 2365. The host says he is still learning multicast and asks viewers to correct him.
• FreePBX's built-in page-group multicast needs a newer Asterisk. It requires Asterisk 13 or higher, and on his Asterisk 11 the Page Pro commercial module's RTP multicast option is greyed out. Multicast still works by setting the address in Endpoint Manager phone templates: comma-separated addresses join one phone to several groups, and a Sangoma S500 line key can page a multicast address directly. He tested this with an S500 and a Yealink T23G.
• A self-contained paging server is his pick for hospital-scale paging. The CyberData VoIP V3 paging server registers as one FreePBX extension (251 in his lab). Callers dial it, enter a two-digit zone from 00 to 99 and, optionally, a security code. It also bridges legacy analog paging over RCA line in/out, and has a relay, a night ringer and up to 10 custom audio files.

2016-12-23 · open the original ↗

• Holiday time conditions need updating for this year's observed dates. The host warns that Christmas 2016 is observed on Monday 26 December rather than the 25th, and New Year's Day on 2 January, so FreePBX holiday entries left on the usual dates will not route calls to the holiday destination on the actual days off.
• Put a holiday time condition ahead of normal call routing. In the host's design, inbound calls hit the holiday time condition first: on a match they go to a holiday announcement (then voicemail or hang-up); otherwise they flow on to an extension, the main IVR, or a further business-hours time condition.
• Holidays live in a time group keyed only on dates. Under Applications > Time Groups the host builds one "holidays" group using only month-day and month ranges, ignoring times and weekdays, then points a time condition at it. Because holidays such as Thanksgiving and Labor Day move each year, he sets a calendar reminder for early January to re-enter the whole year's dates once.

2016-12-22 · open the original ↗

• The CyberData SIP paging amplifier makes a loud, simple SIP paging system. The host lists MSRP at about $550 for the amplifier and about $76 for a CyberData horn speaker, adding that nobody pays MSRP. In his own test, without the +4 dB volume boost, a page was clearly audible outdoors at about 30 yards from one horn.
• Power comes only over PoE, and the PoE standard limits horns. The unit has no separate power input, so it needs a PoE switch or injector; it runs one horn on 802.3af and two on 802.3at, and auto-senses which. CyberData's documentation gives a 25-foot maximum per speaker run and claims about 5,000 sq ft of coverage at full volume.
• FreePBX setup uses two extensions and Endpoint Manager. The host registers account one as the pager (extension 250) and account two as a "night ringer" (extension 252) that just plays a stored audio file. He provisions with the CyberData template plus DHCP option 66; SIP accounts and two backup servers can also be set by hand.
• Endpoint Manager leaves the default admin/admin login in place. Unlike other devices it provisions, where the password gets changed, the host found this unit's web login stays admin/admin even after configuration, so it has to be changed separately.
• The unit does far more than the host uses it for. He points to a DTMF-triggered relay (code 123), a slot for a CyberData strobe, a sensor input that can fire the relay, 10 multicast groups, and up to 10 custom audio files. He has not worked out how to dial in and play stored files, or what the relay is for in a paging setup.

2016-12-15 · open the original ↗

• Factory-defaulted UniFi gear works at once and shows solid white. The host's build is a USG, a UniFi Switch 8 150W, a UAP-AC-Pro and a Cloud Key, each reset by holding the button about 10 seconds. A solid white LED means "ready for adoption", blue means adopted. The USG defaults to 192.168.1.1 and hands out DHCP, so a DHCP client gets internet before any configuration.
• The Cloud Key's setup wizard adopts everything in one pass. The host sets WAN type (DHCP, static or PPPoE) on the USG's own page first. He then finds the Cloud Key with Ubiquiti's Discovery Chrome plugin and runs its wizard on controller version 5.3.8. The wizard adopts all three devices, creates the main and guest SSIDs, and links to Ubiquiti cloud access.
• Move the LAN off the default subnet, with fixed infrastructure addresses. He re-addresses to 192.168.99.0/24, giving the USG .1, the switch .2 and the Cloud Key .3, while the access point stays on DHCP. DHCP runs from .100 to .254 so statics have room, and DNS points to the USG first and 8.8.8.8 second.
• Changing the subnet strands the DHCP devices until rebooted. A running ping shows when connectivity returns. Power-cycling the PoE switch also reboots the Cloud Key and access point it powers. When the discovery tool failed to find the Cloud Key again, the host deduced its new address from the order in which DHCP had handed out leases.
• The Cloud Key's static IP is set on its own configure page. The host sets it there, not in the UniFi controller, and sets the switch's static IP under its configuration in the controller. Each change briefly drops whatever hangs off the switch. He finishes by upgrading each device's firmware, and says his product links are Amazon affiliate links paying him about 3–4%.

2016-11-23 · open the original ↗

• ASP.NET Core is finally on TechEmpower, but only tops plaintext. Edwards says round 13 is the first to include ASP.NET Core, measured on 1.0.1 rather than the just-shipped 1.1, and it landed 10th in plaintext but "kind of halfway down" in JSON, database and fortunes. He hopes round 14 lands in January on 1.1, and says the goal is top ten in every test, with much larger gains expected from the version after 1.1.
• Pipelines invert the Stream model so the framework owns memory. Edwards describes pipelines (formerly "channels", in the corefxlab repo) as handing the caller a pooled buffer instead of the caller allocating one, with providers planned for sockets, the file system, libuv and Windows registered I/O. He relays an MVP Summit demo where naive stream code allocated "many hundreds of megabytes" against about 14 KB with pipelines; MVC users would get the benefit without code changes.
• A four-month community PR made Kestrel about 5% faster. A community member, Ben, replaced a binary search for the matched byte in Kestrel's vectorized HTTP parsing with a de Bruijn-sequence "magic number"; Edwards puts the plaintext result at roughly 5.25 million requests per second. Review found that reordering two lines broke Kestrel because lock synchronization was uncommented, and the JIT team made fixes as a result.
• Developers want the SDK; minor versions do not roll forward. Edwards says the SDK carries only the single runtime it was built against, so anyone wanting both 1.0 (LTS) and 1.1 (current) must install both. An app built for netcoreapp1.0 will roll forward to new patch versions but will not load on a machine with only 1.1 unless forced with a dotnet.exe argument, a behaviour he says he is not satisfied with and the team plans to revisit.
• The new MSBuild tooling is alpha; csproj aims to RTM with VS 2017. Edwards calls the Preview 3 SDK shipping in the VS 2017 RC "alpha quality", with side-by-side issues alongside VS 2015, while the plan of record is for csproj tooling to ship final with VS 2017 RTM. He says dotnet build --native (CoreRT) is a much longer-term effort outside both the VS 2017 and post-1.1 timeframes.

2016-11-08 · open the original ↗

• The 150-watt UniFi Switch 8 suits a small office powering phones. It has 8 gigabit ports with 802.3af PoE and 2 SFP ports, any port settable to Ubiquiti's proprietary 24V passive PoE for its cameras and lower-end access points; MSRP is $199. The host reckons the 150 W total budget would easily carry four or five demanding devices such as UAP-AC Pros or VoIP phones.
• Cheaper eight-port UniFi switches are coming, per Reddit photos. One, shown at WISPAPALOOZA in Las Vegas, takes AC or 802.3af PoE in on its first port and passes PoE out on its last, with no PoE to other devices, useful where there is no AC power, such as an attic. A 60 W version powers only its last four ports. The host assumes both will undercut $199 but has no price or ETA.
• The host's first review unit arrived dead on arrival. It showed no fans or lights across several cables and outlets, and the RMA replacement took about 20 days; meanwhile he bought two new units for the customers they were meant for, and the replacement powered up normally.
• Adoption needs the set-inform command sent twice. He found the switch in Ubiquiti's Device Discovery Chrome plugin only under the UniFi family filter, sent set-inform to http://<controller>:8080/inform with the default ubnt/ubnt credentials, and adopted it; it then went disconnected until set-inform was repeated, which he says also applies over SSH.
• The host still prefers full EdgeSwitches to UniFi switches. He concedes the UniFi controller is quickly catching up in capability, offering per-port VLAN trunking, mirroring, aggregation, port isolation and storm control, but thinks much of the UniFi switch's popularity is Ubiquiti marketing: users liking every bubble green on the controller dashboard.

2016-10-17 · open the original ↗

• The EdgeSwitch 48 is a data-center switch priced well below HP or Cisco equivalents. The host quotes about $760 US for the 48-port 500 W model and about $875 US for the 750 W model; each has 48 ports plus two SFP+ and two SFP ports, and he says he has installed a number of them without problems. Ubiquiti forum threads list which SFP modules are compatible.
• Size the PoE budget by how many ports carry PoE. Rather than adding up each device's wattage, the host tells customers to buy the 500 W version if half or fewer of the 48 ports will power PoE devices (phones, access points, cameras), and the 750 W version if half or more will; he says the rule has not let him down so far.
• Out of the box the switch takes a DHCP address and default credentials. Find it with the Ubiquiti Discovery tool or the DHCP leases and log in as ubnt/ubnt. The host changes the password first (best practice would also change the username, which he skips here), then sets a static IP under System > Connectivity > IPv4; the browser does not follow the switch to its new address.
• Changes must be saved explicitly with Save Configuration. On the shipped 1.0.1 firmware a popup warns of unsaved changes; on later firmware the Save Configuration button in the upper right glows red instead.
• Firmware is upgraded through the backup image slot, then a manual restart. The host uploads the new image (version 1.6.0, as of 19 September 2016) as the backup firmware, which takes about four to five minutes, and sets it as next active; contrary to what he expected, the switch does not reboot on its own and must be restarted under System > Utilities.

2016-10-11 · open the original ↗

• The ASP.NET Core minor release is locking down its features. An ASP.NET team member (Hanselman is travelling and absent; Jon presents the community links) lists what is on the roadmap for what "effectively" becomes 1.1, after the June 1.0 release: URL rewriting and response caching middleware, the WebListener server, middleware usable as MVC filters (so request localization can run after routing and read a culture token from the URL), view components as tag helpers, and a single Azure integration method on WebHostBuilder that turns on App Service logging and IIS integration and is planned to cover Key Vault. A command-line Razor view pre-compilation tool, run at publish, is meant to cut startup time by avoiding loading and JIT-ing Roslyn, which he says costs "multiple seconds".
• A cookie-based TempData provider fixes MVC's out-of-the-box failure. The default TempData provider goes through session, which MVC does not depend on, so calling TempData in a new MVC project "explodes" unless session is configured. The new provider round-trips the data in a cookie encrypted with the app's data protection setup. He concedes cookies have size limits, but argues they suit post-redirect-get status messages, and notes that session in ASP.NET Core does not use System.Web's pessimistic per-session lock.
• SignalR is being rebuilt on a new byte-pumping abstraction. Working name "ASP.NET Core Sockets", built on "channels": a layer for memory-efficient movement and parsing of bytes, with pooling and few or no allocations. He says it could let Kestrel switch between libuv, .NET sockets or Windows RIO without changes further up the stack. As of the recording, a SignalR preview is expected "probably early next year" and RTM "mid next year", with Razor Pages, now past its prototype, on the same timeline.
• Several large efforts are running in parallel, mostly landing next year. He names the MSBuild move, which is happening in public repos but has not been summarized because of heavy churn; .NET Standard 2, which promises to bring back many APIs; and the web API token and Azure identity story, where IdentityServer4 is the current option. A post on release cadence under the new LTS/FTS support policy is promised "soon". In Q&A he says tooling reaches RTM on the VS "15" timeline and that the docs are intended to move to docs.microsoft.com.
• Kestrel is still not approved as a public-facing edge server. Discussing Rick Strahl's post on external access to Kestrel, the team member says they always recommend treating it as an application server behind a reverse proxy such as IIS or nginx. Other links note that EF Core lacks the old database initializers and seeding, and that a formal CLI "global tools" install story is wanted but "doesn't exist yet".