With gridscale.io, the Cologne-based software company gridscale provides the technology basis for the fully automated operation of virtualized and container-based environments at system houses, service providers and, more recently, for local installation in corporate data centers. By means of automation, including real-time provisioning and statistical evaluation of the performance values, gridscale has been able to achieve virtually 100% availability of the IT infrastructure.
Gridscale sees itself as a software supplier for modern cloud operations, in which the infrastructure can be operated with stability and scaled up and down very quickly using real-time provisioning, autoscaling and intelligently calculated capacity forecasts.
The software can be installed in the company data center or hosted by gridscale. Henrik Hasenkamp, co-founder and CEO of gridscale, told LANline that the software stack will be loaded and the operation including all operational processes and capacity management will be taken over.
In 2014, Hasenkamp reports, gridscale decided to develop a highly available decentralized system to enable companies to operate autonomously IaaS/PaaS (Infrastructure/Platform as a Service) without operational problems. At that time, the market did not meet this requirement despite numerous cloud software providers.
For this reason, Hasenkamp continues, algorithms have been put together to process the data center operation as a decentralized system without personnel, to identify anomalies and to react to them with a catalog of measures. The gridscale technology monitors all relevant metrics of the cloud environment - for example, the CPU times a database consumes - and reacts automatically in case of anomalies.
An example: If there is no more I/O activity after a CPU load peak, the system evaluates this as an indication of a kernel panic, as the gridscale CEO confirms. It then automatically reloads the entire workload. In case of malfunctions, the customer data is saved to another storage medium via live migration. According to Hasenkamp, this concept of high availability by means of automated operation and failover processes runs through the entire operation of the infrastructure.
For highest possible performance gridscale used a profiling of the workloads, which was dynamized by means of a classification algorithm and statistical methods: According to Hasenkamp, gridscale uses "statistical methods" - marketers in the industry would say: machine learning - with the aim of identifying anomalies as emerging disorders. From the telemetry data such as CPU utilization, I/O behavior, etc., the software calculates a forecast for the workload behavior and reacts immediately in the event of a target/actual delta. According to Hasenkamp, this makes it possible to operate thousands of data centers "mainly autonomously".
Gridscale also offers automated Kubernetes as a platform service for container operations. According to Hasenkamp, orchestration with Kubernetes is suitable for almost all container scenarios, but the services for logging, authentication, etc. are "not so great". These problems are solved by integration, for example by means of certificates for individual services and securing the APIs. However, this is still in the closed beta phase.
Gridscale hosts the solution from two georedundant data centers of E-Shelter and InterXion. According to Hasenkamp, the infrastructure uses standard 19-inch rack servers with a 10/40G network as well as redundant power supply and connectivity. Since the system runs decentralized, the storage is used from the compute nodes. For larger environments (from 50 systems), however, separate storage nodes can also be used. Gridscale.io supports virtualization and containers as well as their mixed operation. Provide two reference designs that are as manufacturer-neutral as possible.
According to Hasenkamp, the provider gives an SLA of 100 percent on the availability of primary services, 99 percent on secondary services. There would be a refund for every violation. Up to now, one hundred percent availability has actually been achieved in one's own data centers. Thanks to the distributed architecture, there would be no loss of workloads or data even in the event of a complete failure of a zone, emphasizes the gridscale CEO.
The Cologne-based company supplies the technology primarily to large providers and system houses, including the Komsa Group and Controlware, according to Hasenkamp. Whitelabel purchases from a gridscale-DC are licensed according to usage (pay per use), purchases from a partner-DC according to managed RAM. A mix of the models is also possible, for example operation at the partner and backup at gridscale.
The original article in german can be found here.