Increased number of users ideally means an increase in benefit for an online service. However, if the service cannot scale sufficiently well or fast enough, all users suffer from the growing pains the service is experiencing. What should have been an opportunity for greater exposure and a larger user base, can instead turn into loss of sales, customer unwillingness to return, and a bad reputation. Using the elastisys cloud platform and its predictive auto-scaling features, your cloud deployment is always kept at the right size. The elastisys cloud platform will also proactively increase your cloud deployment to make sure that new cloud instances are fully operational as new users arrive.
Typical auto-scaling solutions today only work reactively, typically based on some threshold values: “if the service is under this much load for X amount of time, scale up by Y amount”. This does not ensure that sufficient capacity is allocated to your cloud deployment in case of a sudden large increase in users. Essentially, the reactive approach is like determining a safe altitude by looking directly down at the ground below and trying to keep some safe margin above it.
This looking down-approach works as long as the bumps are small enough to be well within the safety margin. But when a usage spike appears, just looking down is useless. Looking forward, and predictively maneuvering ahead of time to compensate, is the only way to avoid being hit by the spike.
What this means for service deployment, is that cloud instances need to be made operational well in advance. Even using modern technologies like containers that start more or less instantly or virtual machines that start within minutes, this means planning ahead.
Elastisys cloud platform supports multiple advanced auto-scaling algorithms, which can be combined to give your operations team full control, and peace of mind. We provide algorithms tailored for:
- Recurring workloads, where our algorithms and statistical methods determine if there are high correlations between historical workloads and the current one, and if so, uses those previous workloads as an indication of what the future load will be. These are very good for repeating workloads with periodical patterns, such as higher load during office hours, increased activity at the end of the current month, or similar.
- Irregular workloads, where our algorithms emit short-term predictions of future load based on observed recent load. These are very good for managing irregular workload variations, such as a sudden influx of users due to a promotion or a campaign that has gone viral.
- Rule-based directives, which provisions in response to exceeded thresholds. These are not predictive, but rather, good to keep certain key performance indicators (KPIs) within certain bounds.
In spite of how advanced the mathematical and statistical models may be, using the elastisys cloud platform’s auto-scaling features is simple: the elastisys auto-scaling engine uses an estimation of how high load a single cloud instance can successfully handle and monitoring data from your cloud instances. Based on this, it performs advanced calculations and employs machine learning techniques that ensure that you always have a sufficient amount of cloud instances operational as users come and go.
Elastisys cloud platform does predictive auto-scaling right. With it, you can regard insufficient capacity as a problem of the past. It just works.