Switch power module failure: symptoms, tests and spares planning
A failing switch power module shows up as a dark or red PSU LED, a chassis power alarm, unequal load sharing between twin supplies, or repeated thermal shutdowns under load. Confirm the fault with input and output voltage, ripple and load tests, then size cold spares from the installed base and its MTBF.
Symptoms: lamp states, log timing and thermal evidence
A switch can lose a power supply and keep forwarding for weeks. The evidence arrives as a lamp state, a log event or a temperature reading. The combination narrows the fault to the input side, the output stage or the load path.
| Symptom | Field evidence | Threshold or reference | Likely fault |
|---|---|---|---|
| Dark STAT lamp | PAC600S12-EB lamp off | Input under/overvoltage, output OC/OV, short, OTP [1] | Input or output protection |
| Red AC OK lamp | Cisco 9300 AC/DC OK red | “Output has failed” [2] | Output stage |
| Repeated reloads | Hiccup-mode OV/OC | Output off/on until fault clears [1] | Output regulation |
| Load-share split | 70/30 at equal input | QFX5110-48S-AFO pair is 2 × 650W, 1+1 [4] | Regulation loop or ORing FET |
| Fan speed | READ_FAN_SPEED_1 (0x90) | Well below twin at same inlet [3] | Fan bearing or heatsink |
| Ripple | 12V rail above 120mVpp at 50% load | ATX12V 2.4 limit | Output capacitor ESR |
| PoE shedding | Data up, powered devices off | PoE budget exhausted [4] | 56V module or budget |
| Breaker trip | Input OCP latched | No auto-recovery [1] | PFC stage or rectifier |
A dark LED with a tripped breaker or PDU alarm points upstream, at the feed rather than the module. A green LED with repeated reloads or PoE shedding points to output regulation or thermal margin. Huawei’s protection table separates the two cases: input undervoltage recovers when the input returns, input overcurrent stays latched, output overvoltage and overcurrent are hiccup-mode, and overtemperature shuts down until the module cools [1].
The 12V and 56V rails fail differently. A 12V module fault takes the control plane with it, so the switch stops forwarding. A 56V PoE module fault leaves data intact while powered devices drop, which looks like a cabling or PD problem until the PoE budget or module telemetry is checked [4].
Load share produces no lamp state at all. In a 1+1 pair, one module carrying 70% of the current at equal input voltage is running closer to its thermal limit. The pair in a QFX5110-48S-AFO is specified as 2 × 650W, 1+1 [4]. A PMBus current log exposes the split before the chassis reports a fault.
Log timing separates thermal from electrical faults. A PSU event repeating every afternoon tracks solar gain or a chiller setback. One repeating at PoE power-up tracks inrush or a 56V rail at its limit. A single event during a UPS transfer tracks hold-up time. A green lamp carries no margin data, and a module can pass a visual check and fail a 75% load step.
Bench tests and fault isolation
The module comes out of the chassis, the bulk capacitors discharge, and the feed voltage requires a CAT III meter. Opening the module exposes a switching stage at hundreds of volts and a heatsink that may be live.
On the input side, the PAC600S12-EB accepts 90V to 290V AC at 45Hz to 65Hz, draws up to 8A and delivers 600W at 12V and 50A [1]. A 5V drop between the PDU outlet and the PSU inlet at 8A dissipates 40W in the cord or plug, enough to cause a false input alarm.
On the output side, the ATX12V 2.4 guide sets a ±5% band on the 12V rail, or 11.40V to 12.60V, with a 120mVpp ripple limit at 20MHz bandwidth [5]. It is a desktop-PSU specification, and switch vendors may set tighter windows, so the figures work as bench ceilings. A module that reads 12.05V at 25% load and 11.30V at 75% load has crossed the lower limit. A 750mV drop from 25% to 75% load points to a regulation loop losing gain.
Load testing runs 25%, 50%, 75% and 100% steps for 30 minutes each, recording input power, output voltage, ripple and case temperature. Efficiency follows from DC output power divided by AC input power. The 80 PLUS floor is 80% at 10%, 20%, 50% and 100% of rated load with power factor at least 0.9 [10]. The 02313PAC is listed at 94% [4].
Fan and thermal checks use READ_FAN_SPEED_1 (0x90) at no load and at 50% load [3]. A fan that does not change speed, or sits well below its twin, points to bearing drag or a blocked heatsink. The 02313PAC record lists -40°C to 70°C operating and an MTBF above 200,000 hours [4]. Those are datasheet limits at controlled conditions.
Ripple readings depend on probe technique. A long ground lead picks up switching noise and inflates the reading, so the measurement uses a coaxial tip-and-barrel probe or a short twisted pair across the output capacitor, with the scope set to 20MHz bandwidth limit [5].
Hold-up is measured with a scope on the output while the AC input is switched off. The output must stay inside its regulation band for at least 17ms at full load [5]. Insulation resistance and dielectric withstand tests belong on the bench at IEC 62368-1 levels, with the module disconnected from the chassis.
The bench pass bands are short enough to keep in a notebook. The 12V rail stays between 11.40V and 12.60V. Ripple stays at or below 120mVpp at 20MHz bandwidth. The AC input stays within 90V to 290V at 45Hz to 65Hz. A -48V feed stays between -40.5V and -57.0V DC [8]. Efficiency stays at or above 80% at 50% load. Hold-up lasts at least 17ms at full load [5]. A Type 4 PoE port allocates 90W at the PSE [7]. Surge and dip immunity belong to the installed system, with IEC 61000-4-5 and IEC 61000-4-11 as the test references and ITU-T K.21 for customer-premises telecom resistibility [6].
Moving the suspect module into a known-good slot produces one of four results. A failure in both slots means the module is faulty. A pass in the second slot, with the known-good module failing in the original slot, means the backplane, connector or slot power path is faulty. Both modules passing alone while the pair fails together means load sharing, firmware or a power-budget problem. Both modules failing in the original slot means the feed or backplane is faulty.
Two-point measurements catch connector loss. A 12V reading of 12.05V at the PSU edge and 11.60V at the backplane is a 450mV drop across pins that should measure in milliohms. At 30A that is 15mΩ of contact resistance. On the AC side, a 5V drop at 8A is 40W of heat in a cord or plug.
Feed faults mimic module faults because both produce a dark lamp. The PDU breaker, cord, inlet fuse and UPS transfer log are checked first. Huawei’s input overcurrent protection does not auto-recover when the input returns [1], so a module that stays dark after a breaker reset may be latched in a protection state and can be retested in a known-good chassis. Backplane faults leave pitted pins, discolored plastic or a burnt smell. Photographing the connector before cleaning preserves the pitting pattern that shows which pins carried the current.
How much load headroom should a switch PSU carry?
A 48-port PoE+ switch with every port at 30W allocates 1440W at the PSE [7]. The same 48 ports at IEEE 802.3bt Type 4 allocate 4320W at 90W per PSE port [7]. The powered device receives 71.3W at Type 4 because the cable loses the difference, and the PSU still supplies the PSE side. Allocation follows the negotiated class: a Class 4 port draws up to 30W at the PSE under 802.3at, while a Type 4 port draws up to 90W under 802.3bt [7].
The YT CIG 48-port PoE+ reference design lists a 740W PoE budget [4]. The figure is set by PSU capacity, the PoE controller limit and chassis thermals. Adding a second 600W module does not raise it when the controller or airflow is the binding constraint.
Non-PoE data centre switches carry smaller loads. The Juniper QFX5110-48S-AFO is listed with two 650W supplies in a 1+1 configuration, and the Huawei CE6855-48S6Q-HI with two built-in AC supplies and a 216W maximum draw [4]. The CE6855’s 216W maximum is the chassis figure; the QFX5110’s 1+1 label means each 650W module is rated for the full chassis load.
For 1+1, each module must be rated for the full worst-case chassis load at the highest ambient in the datasheet. For N+1, N modules cover the load and the extra module provides redundancy. A 2000W load with 600W modules needs N=4 (2400W) plus one, for 3000W installed.
Nameplate wattage is quoted at an ambient. Above the datasheet’s full-power ambient, the derating curve cuts the available wattage. The -40°C to 70°C range on the 02313PAC record is an operating limit, and full power may need a lower ambient [4].
The 80 PLUS program requires at least 80% at 10%, 20%, 50% and 100% of rated load with a power factor of 0.9 or higher [10]. The floor applies at all four load points, so a module run at 20% load is judged by the same 80% line as one at 50%. Port counts and topology for access and data-centre roles sit in the IP fabric switches and routers guide; PSU sizing follows the module and PoE controller figures.
How should redundant supplies and spare modules be planned?
A redundant pair fails the chassis when the two modules share something they should not. Two PSUs on the same PDU or breaker are one supply with two cords, and a breaker trip or a UPS transfer takes both down. A 1+1 pair is redundant only when one module carries the full load; a pair of 600W modules in a 1000W chassis carries 1200W total but loses redundancy at 600W of load. Different airflow directions, firmware or load-share gains can leave one module carrying most of the load while the other idles.
Both modules ride through a short AC dropout on their output capacitors. ATX12V 2.4 asks for 17ms at full load, and a transfer switch that takes longer resets the chassis even with two healthy modules [5]. Two modules in a 1+1 pair also draw the same inlet air, so a blocked filter or a recirculating hot aisle raises both temperatures.
Capacity arithmetic separates 1+1 from N+1. A 1+1 pair of 600W modules installs 1200W and covers a 2000W load only if the chassis never exceeds 1200W. A 4+1 set installs 3000W and delivers 2400W after one failure, which covers the 2000W load with 400W of margin. A 4+4 set installs 4800W and still delivers 2400W after one whole group fails. The N+1 count uses the module rating at the site’s ambient. A 25°C nameplate overstates the available power. 1+1 suits access and aggregation switches whose chassis draw sits well below one module’s rating. N+1 suits modular chassis switches whose load grows with line cards.
Spare stock starts from the MTBF. A module rated at 200,000 hours has an annual failure rate of 8,760 / MTBF = 4.38% under the prediction’s conditions, with Telcordia SR-332 Issue 4 as the usual basis [9]. For 120 installed modules, expected failures are 120 × 0.0438 = 5.26 per year. A Poisson calculation for a 95% probability of a spare gives 5.26 + 1.645 × 2.29 = 9.03, so 10 modules. For 99%, 5.26 + 2.326 × 2.29 = 10.6, so 11 modules.
Lead time adds pipeline stock. At 12 weeks and a steady failure rate, 5.26 × 12 / 52 = 1.21 modules are in transit or on order, so the pool needs two more. Criticality can raise the number: emergency-call or hospital PoE switches justify a cold spare per 10 chassis even when the Poisson result is lower, while lab switches can share a central pool. The 02313PAC storage rating of -40°C to 85°C and 5% to 95% relative humidity sets the warehouse envelope [4].
Mixing wattage is a separate question from mixing firmware. Some chassis allow a 600W and an 1100W module with the pair limited to the smaller rating, while others reject the combination at power-on. The FRU list and the power-supply configuration guide decide.
Chassis fit is decided by the FRU list, airflow direction and connector keying. Wattage alone does not settle it.
| Part number | Family or use | Output | Input | Efficiency | Notes |
|---|---|---|---|---|---|
| PAC600S12-EB | Huawei S6700 | 12V / 50A, 600W | 90V to 290V AC | About 96% peak | Hot-swap, 90 x 215 x 40 mm |
| PAC600S56-EB | Huawei PoE | 56V, 600W | 100V to 240V AC | Not listed | Overtemperature protection |
| 02313PAC | Huawei PoE | 56V DC, 600W | 90V to 290V AC | 94% | MTBF above 200,000 hours |
| PAC-2200WF | Huawei CloudEngine/NetEngine | 2200W | Not listed | Not listed | Catalogue compatibility |
| PWR-M120-DC-S | Juniper M120 | 2100W, 48V DC | Not listed | Not listed | DC module |
| SmartPowerI48/3000 | Emerson | 3000W, 48V DC | Not listed | Not listed | 48V DC output |
The Huawei 600W family includes 12V and 56V PoE modules that are not interchangeable. PAC600S12-EB delivers 12V at 50A, while PAC600S56-EB is 56V [4]. PAC600S12-EB measures 90 × 215 × 40 mm and weighs 0.985 kg [1]. A back-to-front module in a front-to-back chassis pushes air against the chassis fans and can trip overtemperature protection; the 02313PAC record specifies back-to-front airflow [4].
How should a spare PSU be stored and commissioned?
The 02313PAC is rated for -40°C to 85°C storage and 5% to 95% relative humidity [4]. The module stays in its ESD bag with desiccant until it goes into a chassis.
Commissioning runs 30 minutes at no load, then two hours at 50% load, with READ_VIN, READ_VOUT, READ_IOUT and READ_TEMPERATURE_1 read over PMBus. A module in storage for years may need capacitor reforming, and the capacitor manufacturer’s shelf-life note sets the procedure.
The serial number and FRU revision go into the asset register, because a spare that fits one hardware revision may not fit the next. The cooling fan tray and DC power cable assembly are the other consumables to stock.
Frequently asked questions
Can a failing power supply still show a green LED?
Yes. A green lamp reports a threshold crossing at one instant. A module with rising ripple, a lazy fan or a load-share offset can hold 12.0V at 25% load and fail at 75%, and the bench load test records the difference.
How long can a switch ride through an AC dropout?
ATX12V 2.4 asks for 17ms at full load [5]. Switch vendors vary around that figure. That covers a UPS transfer, not a generator start that takes seconds.
Can a 600W module be replaced with a higher-wattage module?
Only within the chassis FRU list. A higher-wattage module with a different airflow direction, connector keying or latch geometry will not fit, and the chassis power-supply configuration guide decides fit.
What MTBF basis should a spares plan use?
The vendor datasheet figure and one calculation standard, normally Telcordia SR-332 Issue 4 [9]. Figures computed at different temperatures or load points are not comparable.
Should power supplies be replaced on a calendar schedule?
No IEC or ITU standard sets a PSU replacement interval. Replacement evidence is fan-speed drift, ripple above 120mVpp on the 12V rail, repeated overtemperature trips or a growing load-share offset.
Sources
[1] Huawei S6700 Series Hardware Description, PAC600S12-EB.
[2] Cisco Catalyst 9300 Hardware Installation Guide.
[3] PMBus Specification 1.3.
[4] YT CIG catalogue records.
[5] Intel ATX12V Design Guide 2.4.
[6] IEC 61000-4-5; IEC 61000-4-11; ITU-T K.21.
[7] IEEE 802.3at-2009; IEEE 802.3bt-2018.
[8] ETSI EN 300 132-2.
[9] Telcordia SR-332 Issue 4.
[10] 80 PLUS program.
Need a specific part?
Send the part number and quantity — we confirm condition, availability and price before you commit.