Increasing Energy Efficiency in Networking End Points

Networking end points such as NAS boxes, printers demand higher performance with the increasing throughput and processing requirements. This is achieved by parallelizing work loads across multiple processing units working at higher frequencies, which in turn demand greater power. Regulatory authorities around the world mandate energy efficiency for Green operation. These regulations are two dimensional: consuming minimum power in standby mode and optimizing the ratio of performance to power consumption.

For a typical use case of an enterprise network printer in an office environment, the printer utilization goes through different levels of workload. These levels are depicted in the figure below. Apart from the fluctuating daytime utilization levels, during the night the printer is idle or not utilized and thus, it is wasteful to keep the equipment and devices fully throttled up. Note that in such environments, it is not feasible to shut down and boot-up the printer again on regular basis. Thus, the printer should be able to intelligently shut down the majority portion of the components on board and major sections of the processing units to effect power savings. It would be also be desirable to have techniques to monitor the traffic profile and adjust the frequencies of the processing elements on-the-fly to reduce runtime power.


Click to enlarge

Based on the traffic flows, the power consumption based operational modes of the printer can be categorized as follows:

Run
The system gives peak performance in this mode. All the devices on board run at the peak frequencies, consuming maximum power.

Jog
The system typically operates in this mode in low work load scenarios. On detecting low workloads the monitoring application configures the system to migrate to this mode. The devices on board work at reduced frequencies to reduce dynamic power consumption. The computing elements which are idle can be powered down for further power savings.

Sleep
The system typically operates in this mode when there are no work loads pending, but needs to be still connected to network. The energy consumption is very LOW. In this mode the DDR is kept in self refresh state. The critical interfaces such as Ethernet which receive packets are kept ON. On reception of the packet, the DDR is taken out of self-refresh mode, packet is stored and DDR is again put back in to self refresh mode.

Hibernate
This is the most aggressive power saving mode achieved by powering down almost all of the components except a very small portion needed to wake the system up. The power consumption is reduced to bare minimum. The system does not respond automatically to the packets or work loads. On board logic (such as FPGA) triggers the wakeup, based on programmed timer expiry or any network activity.

System Design for Network based Power management


Click to enlarge

The above figure illustrates the main logical components of an energy efficient system for a printer. The printer board consists of one or more processing elements (cores), which run an operating system such as Linux, stored in the DDR. An Ethernet controller connects the printer to the network and all the print requests come from different entities on the network. There are applications running on this system which do the page processing, font rendering and printing activity. These applications can be load balanced on one or more cores of the system. The statistics of the print requests is collected by the system monitor application at regular intervals of time.

The board has a Power Management Controller (PMC) which manages the power consumption based operating modes listed earlier, such as Run, Jog etc. System monitor application directs the PMC to transition to a specific mode based on the System’s Power Management Policy. The power management policy is very generic and configurable. The system monitor directs the PMC to switch between the modes on the basis of the processing requirements, which are governed by the workload.


Click to enlarge

Power Management state transitions
The mode transitions can be described with the above state diagram. The traffic of page requests can be accounted in terms of pages per millisecond (Pg/ms).

- Run Mode
Upon reset the printer is initialized and starts processing the print requests. The print application divides its tasks into individual parallel threads of execution running on separate cores. If the system monitor detects the Pg/ms rate falling below “n”, which suggests that the overall system is taking more power than it is required, it directs the PM Controller to transition the system into the jog mode.

- Jog Mode
In the jog mode the processing units as well as other system blocks run at reduced frequencies. This is programmed by the system monitor. The frequency at which the processing unit runs to process a thread of execution can be set as per the processing requirement. If the page requests fall below a number “k” where all the threads can be run on a single core , a mode called Jog+ could be deployed – wherein the remaining cores could be powered off. If at any point the number of requests rises above “k”, the remaining cores are powered on one at a time and the threads are migrated to the specific cores. Similarly if the Pg/ms rises above “n” the system is made to transition to normal frequency run mode.

- Sleep Mode
If the Pg/ms falls to zero for time ‘t’, the system monitor requests the operating system to save the context in DDR and directs the PMC to switch off the cores and put the system in sleep mode. The PMC stays ON along with a few other interfaces like ethernet waiting to get a print job or any other user defined packet, which marks the start of traffic, to wakeup the printer and move to jog or run mode.

- Hibernate Mode
During night hours the traffic goes to zero Pg/ms. This is an everyday scenario and during this slot the printer and moreover the network is not utilized at all. Thus instead of going to the sleep mode the system monitor stores all context into DDR, puts it into self refresh and requests the PMC to shut off everything on board. As in figure below, an on board FPGA (ACTimon), monitors the network activity. If there is any activity on the network, the FPGA powers on the board components and the system wakes up to the run mode. Alternatively the FPGA logic can trigger wakeup based on expiry of a pre-programmed timer.


A typical networking endpoint system.
Click to enlarge

In the late 1970s, physicist Amory Lovins popularized the concept of a "soft energy path", with a strong focus on energy efficiency. He popularized the notion of negawatts—the idea of meeting energy needs by increasing efficiency instead of increasing energy production. Consider this - if all households in Europe changed their more than ten year old appliances into newer efficient ones, 20 billion kWh of electricity would be saved annually, hence reducing CO₂ emissions by almost 18 billion kg! Thus energy efficiency is an extremely effective solution to issues like pollution, global warming, fossil fuel depletion, etc. As mankind continues to innovate, connect and build a hyper-digital future – we need to pause and rethink, replan and rebuild so we do not run out of gas!