Load Shedding

Load shedding status

We are currently not Load Shedding due to high demand or urgent maintenance being performed at certain power stations. To determine the time/s that you will be affected, please view the Schedule for your area. However, if you are an Eskom customer and do not have power, please register on our App to log your fault: Click here to download the Android or OR iOS version of the App.

Load shedding (loadshedding) is a way to distribute demand for electrical power across multiple power sources. Load shedding is used to relieve stress on a primary energy source when demand for electricity is greater than the primary power source can supply.

Load Shedding

DOWNLOAD 🔥 https://tiurll.com/2y2R84 🔥

The goal of load shedding is to prevent a power grid or power source from overloading. As a type of load management, load shedding works by rotating power outages or reducing power consumption from primary sources until demand decreases and more capacity becomes available. Buildings such as data centers often rely on backup power systems during times of load shed to prevent backup failures.

Load shedding is frequently planned, though it is sometimes a strategy used in the aftermath of a natural disaster or severe weather. Load shedding power cuts can last from a few minutes to a few hours. It can disrupt businesses, services, reduce productivity and upset customers or clients.

Load shedding is often planned and negotiated with local building owners. Utility providers monitor electricity demand and identify when it exceeds supply or nears capacity limits. They then create a load shedding plan that entails rotating power outages, temporary current disconnections and incentives to building owners for complying. Once demand decreases or additional power resources become available, the utility provider restores power to the affected areas.

Load shedding can also happen without prior planning. Power customers might experience involuntary load shedding when a utility electrical provider lowers or stops electricity distribution across a coverage area for a short period of time. This type of load shedding is commonly referred to as a rolling blackout. Brownouts, another type of involuntary load shedding, are caused by a power supplier lowering voltage distribution during peak usage times to balance supply and demand.

Most buildings, including data centers which use 1.8% of the United States' electricity, purchase electrical power from a power utility provider. To reduce the cost of power while also ensuring continuous operation, a building operator may negotiate an agreement with a power provider to voluntarily load shed on a pre-scheduled or on-demand basis.

During load shedding events, the building draws power from its secondary source(s) -- typically on-site diesel generators -- rather than from the utility. Some data centers are switching to green energy sources such as on-site or contracted solar photovoltaics or wind-based renewable power.

Many utilities' load management programs to shift or curtail power usage offer cost incentives to building operators who voluntarily load shed during peak periods. Load management programs are a good option for energy-intensive building operations that also have high-quality power distribution control and secondary power sources, such as a data center.

Building operators sometimes agree to demand response plans, where instead of load shedding, they promise to use less power during certain periods of time. Some are also transitioning to smart buildings, which are more energy efficient than traditional buildings.

To prevent disruption to the systems in a building, an operator can rely on uninterruptible power supply systems and power distribution units (PDUs) that moderate the flow of electricity to sensitive equipment. Small to midsize businesses and residential buildings with backup power generation might also be candidates for load management programs. Environmental protection bodies define and regulate load shedding as nonemergency use of nonprimary power in countries such as the U.S.

Power outages and load shedding are similar, which can be confusing. Both, for instance, require advance planning: load shedding involves planning with utility providers while power outages require strong continuity plans. Despite each term involving power cuts and plans, they are two different phenomena.

Unlike load shedding, which is temporary, a power outage lasts until power is restored. This can take a few hours, days, weeks or longer. In the event of both load shedding and power outages, some buildings use automatic transfer switches to switch to secondary power sources.

One common question we struggled with was determining the default number of connections the server would allow to be open to clients at the same time. This setting was designed to prevent a server from taking on too much work and becoming overloaded. More specifically, we wanted to configure the maximum connections settings for the server in proportion to the maximum connections for the load balancer. This was before the days of Elastic Load Balancing, so hardware load balancers were in widespread use.

We set out to help Amazon service owners and service clients figure out the ideal value for maximum connections to set on the load balancer, and the corresponding value to set in the frameworks we provided. We decided that if we could figure out how to use human judgment to make a choice, we could then write software to emulate that judgment.

Determining the ideal value ended up being very challenging. When maximum connections were set too low, the load balancer might cut off increases in the number of requests, even when the service had plenty of capacity. When maximum connections were set too high, servers would become slow and unresponsive. When maximum connections were set just right for a workload, the workload would shift or dependency performance would change. Then the values would be wrong again, resulting in unnecessary outages or overloads.

During tests, we make sure to measure client-perceived availability and latency in addition to server-side availability and latency. When client-side availability begins to decrease, we push the load far beyond that point. If load shedding is working, goodput will remain steady even as offered throughput increases well beyond the scaled capabilities of the service.

In terms of visibility, when load shedding rejects requests, we make sure that we have proper instrumentation to know who the client was, which operation they were calling, and any other information that will help us tune our protection measures. We also use alarms to detect whether the countermeasures are rejecting any significant volume of traffic. When there is a brownout, our priority is to add capacity and to address the current bottleneck.

If misconfigured, load shedding can disable reactive automatic scaling. Consider the following example: a service is configured for CPU-based reactive scaling and also has load shedding configured to reject requests at a similar CPU target. In this case, the load shedding system will reduce the number of requests to keep the CPU load low, and reactive scaling will never receive or get a delayed signal to launch new instances.

We are also careful to consider load shedding logic when we set automatic scaling limits for handling Availability Zone failures. Services are scaled to a point where an Availability Zone's worth of their capacity can become unavailable while preserving our latency goals. Amazon teams often look at system metrics like CPU to approximate how close a service is to reaching its capacity limit. However, with load shedding, a fleet might run much closer to the point at which requests would be rejected than system metrics indicate, and might not have the excess capacity provisioned to handle an Availability Zone failure. With load shedding, we need to be extra sure to test our services to breakage to understand our fleet's capacity and headroom at any point in time.

When a server is overloaded, it has an opportunity to triage incoming requests to decide which ones to accept and which ones to turn away. The most important request that a server will receive is a ping request from a load balancer. If the server doesn't respond to ping requests in time, the load balancer will stop sending new requests to that server for a period of time, and the server will sit idle. And in a brownout scenario, the last thing we want to do is to reduce the size of our fleets. Beyond ping requests, request prioritization options vary from service to service.

Prioritization and throttling can be used together to avoid strict throttling ceilings while still protecting a service from overload. At Amazon, in cases where we allow clients to burst above their configured throttle limits, the excess requests from these clients might be prioritized lower than within-quota requests from other clients. We spend a lot of time focusing on placement algorithms to minimize the probability of burst capacity becoming unavailable, but given the tradeoffs, we favor the predictable provisioned workload over the unpredictable workload.

Load balancers might also queue incoming requests or connections when services are overloaded, using a feature called surge queues. These queues can lead to brownout, because when a server finally gets a request, it has no idea how long the request was in the queue. A generally safe default is to use a spillover configuration, which fast-fails instead of queueing excess requests. At Amazon, this learning was baked into the next generation of the Elastic Load Balancing (ELB) service. The Classic Load Balancer used a surge queue, but the Application Load Balancer rejects excess traffic. Regardless of configuration, teams at Amazon monitor the relevant load balancer metrics, like surge queue depth or spillover count, for their services.

In the beginning of this article, I described a challenge from my time on the Service Frameworks team. We were trying to provide Amazon teams with a recommended default for maximum connections to configure on their load balancers. In the end, we suggested that teams set maximum connections for their load balancer and proxy high, and let the server implement more accurate load shedding algorithms with local information. However, it was also important for the maximum connections value to not exceed the number of listener threads, listener processes, or file descriptors on a server, so the server had the resources to handle critical health check requests from the load balancer. ff782bc1db