“With the cloud, individuals and small businesses can snap their fingers and instantly set up enterprise-class services,” says Roy Stephan, founder of PierceMatrix.
Dataproc offers you a sophisticated, effortless cluster management service for your Spark and Hadoop projects. It is a managed service on the Google Cloud Platform (GCP).
Dataproc is a boon for Spark, which is functional and fast to handle the enormous volume, diverse variety, and high velocity of today’s data.
Spark is resource hungry. Buying, networking, and managing many machines for data analytics can be costly, disrupting, and time-consuming. Running your data projects on fewer machines burns up more of another precious, irrecoverable resource—time.
To mitigate this challenge, Google’s Dataproc is an economical yet powerful, managed service for Spark and Hadoop. With each virtual machine (VM) priced as low as a cent an hour, adopting the service does not require a second thought.
How Dataproc seamlessly complements cutting-edge technologies like Spark
So, how does Apache Spark execute so fast?
While the architecture of older technologies like Hadoop is disk I/O intensive, Spark uses RAM, that is working memory, to execute jobs. However, RAM is much more expensive than hard disk space. Dataproc offers a powerful yet inexpensive solution.
Dataproc harnesses the virtually limitless computing resources of Google Cloud. That’s not all. As explained below, it is a managed service that facilitates optimal, low-cost utilization of these vast resources.
At the heart of Dataproc are its easy cluster scalability functions. You have the option to scale up or scale down a cluster at any time with Dataproc, even when jobs are running on it.
Dataproc lets you configure two kinds of VMs, called workers, in a cluster on which Spark runs. The first is the primary worker; the second, the secondary, or the preemptible worker.
While it is difficult to accurately calculate the actual number of VMs necessary for a given data analytics job, using Dataproc, you begin with an approximate number of primary and secondary workers. Notably, the cost of a secondary worker is much less than that of a primary worker.
Thus, you can afford to begin with a higher number of secondary workers at the start of a job, when typically more VMs are required and shut down the unwanted secondary workers towards the end of the job. This way, you enjoy the twin benefits of high resource availability and low costs. As a result, you save big on both the time to run jobs and the cost per job.
This orchestration of VM’s comes with reliable connectivity to Google Cloud for storage. Using the Google Storage Connector, an open-source Java library, you can ensure seamless interoperability between Spark, Hadoop, and Google Cloud. Through this option, you can directly access data stored in Google Cloud, without transferring it into HDFS first. Moreover, the data stays intact on the cloud, long after you have shut down the cluster.
Faster cluster configuration
Besides, Dataproc reduces the time to execute each job by drastically reducing its setup time. This is a big advantage for data scientists, given the complexity and volume of jobs that they deal with daily. With a point-and-click GUI console to set up your cluster, Dataproc takes just 2–3 minutes to configure a job compared to about 30 minutes for the same process when using other technologies.
Scalability options in Dataproc
In addition to manual scaling of a cluster, Dataproc uses Autoscaling to dynamically control the number of VMs corresponding to load demands. Autoscaling works on a predefined policy to control how Dataproc responds to load demand.
Using measures like maximum CPU utilization, the number of HTTP requests, and stackdriver metrics, the policy tells the autoscaler when to spin up or shut down VMs. It is possible to define up to 5 policies for a single cluster, giving you fine-grained control over VMs for running jobs.
The autoscaler uses log data of the last 10 minutes of cluster operation to calculate the optimal cluster size. If this data indicates that the same load can run on a cluster with fewer VMs, autoscaler performs a controlled scale down of the cluster to continue executing the job with a lower number of VMs.
Moreover, you can forcefully decommission a preemptible worker at any time, giving you full control over a running job and cluster size.
Dataproc logs—to the last second—the duration for which each VM is used and bills you accordingly. This technology’s autoscaling and close-grained pricing help you focus on your business priorities by eliminating digression to pointless technical challenges.
Occasionally, when downsizing a cluster, a worker may terminate while work is in progress. Graceful Decommissioning addresses this issue. Using graceful decommissioning, a worker in a decommissioning state will wait for the work in progress to end before shutting down. Such decommissioning prevents time-consuming recovery procedures, thus ensuring your jobs run faster.
Ease of use, Security, and Reliability with Dataproc
Managing a complex, heterogeneous infrastructure is a nightmare. Containers help engineers run applications independent of the underlying OS. While VMs emulate the underlying hardware, containers emulate the underlying OS. This helps bring virtual homogeneity and ease of management to a heterogeneous computing environment.
Nonetheless, managing containers is an additional overhead in data analytics projects. Traditionally, data scientists had to deal with more than one container management system to run a data analytics job. For instance, when Spark ran on Google Cloud, data scientists had to use both Kubernetes and YARN to manage many containers in the underlying cluster.
Dataproc has a unified console that makes container management easy. The unified console simplifies container management while saving you precious time and effort.
Lastly but importantly, Dataproc easily integrates with Spark. You can move existing spark jobs by simply changing the storage path from “hdfs://” to “gs://”.
Dataproc uses Google Cloud’s Identity and Access Management (IAM) for comprehensive security. IAM gives you control over which user has access to what resource. With IAM, permissions are not granted to an individual user. Rather, permissions are categorized into access groups called roles. Each user belongs to a role and every resource on the Cloud has a specific role attached to it. This way, IAM controls the level of access to each resource.
Some of the other security features Dataproc offers are as follow:
- Default at-rest encryption
- Data in flight between the machines in a cluster are encrypted
- Your engineers use Secure Shell (SSH) to access machines on the cloud
- End-to-end authorization with GCP token broker
For reliability, Dataproc restores a machine that has crashed suddenly. Using its cluster configuration data, Dataproc detects machines that have stopped working for reasons other than user action. Then, it restores the machine into the cluster automatically. Thus, the integrity of the cluster is maintained for the duration of the job.
A Stable, Scalable, Low-Cost Option
Dataproc offers you vast resources at extremely low initial costs. Moreover, with Dataproc, it only takes 2-3 minutes to set up a large, managed cluster. These benefits are easily justifiable to top management to get your big data projects going.
Now, leverage the disruptive power of big data at costs that do not burn a hole in your pocket.
Contact ACME technology Solution for powerful and low-cost cloud solutions for your every computing need. With D3VTEC, sourcing cutting-edge technology is easy.