203 According to IDC, by 2023 over 500 million cloud-native apps and services will be developed and deployed. That’s equivalent to the total number of apps created in the last 40 years. This accelerated growth will define new competitive requirements in every industry. It will be critical for IT to prepare to ride this oncoming wave, and fast object storage is here to help. Object storage has now been available in various forms, generations and delivery models for more than two decades. With the introduction of AWS S3 in 2006 — the world’s first and most successful cloud object storage service — application developers started to embrace the object model. For cloud services, object storage is a fit given the massive amounts of data these applications tend to generate and manage. Plus, and perhaps most importantly, the use of RESTful APIs for accessing storage really fits the cloud services model, given that it leverages the natural language of internet communications, HTTP. With the rise in popularity of the AWS S3 API, we saw another key obstacle to the adoption of object storage become minimized: a standard API for application access to object storage. Just as Network Attached Storage (NAS) vendors have embraced network file protocols such as NFS and SMB, object storage also needed a de facto API that could be supported in the cloud and in on-premises object storage solutions. Today, the AWS S3 is a tablestake for every object storage vendor on the planet — and is also now widely used by key enterprise applications ranging from backup management to big data analytics, and even in vertical industry solutions in media and healthcare. Object storage misconceptions and the performance metrics that really matter For the last decade, object storage has still been associated for use with “secondary” data — backups and long-term archives to be specific. This is based on a somewhat outdated notion that while object storage could provide advantages in “scale” (growth of capacity), it had the connotation that performance would be insufficient for higher-performance applications. The truth is that object storage, whether in the cloud or on-premises, can provide very high levels of throughput (fast delivery of data per second). This is the performance metric that is the most important for big data payloads, like those seen in cloud applications (images, videos, audio files and documents). In systems such as Scality RING, deploying very high levels of throughput for a wide range of applications has been a standard capability for years. Applications in video content delivery, analytics for big data and even for cloud-based email services have depended on RING to deliver 10’s of gigabytes per second of throughput — and even the performance needed to store petabytes of data per day! The other key performance metric more suitable for smaller data payloads is “latency” — that is, the response time (typically measured in milliseconds) for the application to retrieve data from the storage system. Leveraging flash media to deliver super-fast latencies with object storage In the world of on-premises storage arrays, flash media has now been used for many years to deliver super-fast latencies (below 1 millisecond) for many years. So, why hasn’t object storage leveraged this type of media to deliver low latency before? The first part of the answer is that, in many cases, object storage has used flash media for many years, but mainly to accelerate metadata access. A portion of the storage hardware platform was used to store metadata on flash media, but the larger data payloads would still be stored on traditional hard disk drives (HDD) to get the lowest-cost and best economics. The second part of the answer is that now we are seeing the emergence of a new class of applications that can really benefit from having the actual data stored on flash media in object storage solutions. These applications will be delivered through the model we now term “cloud-native,” which is really based on the use of containers as made popular by Docker to deliver software as a set of services. Plus, it involves the use of container orchestration frameworks — the most popular, by far, now being Kubernetes. Types of application workloads that benefit from fast object storage Next, we should consider the types of application workloads that can benefit from fast object storage. The common thread here is that these workloads will often require access to millions/billions of smaller data objects (files), as well as larger media files (as in the past). Therefore, access latency becomes a bigger issue when dealing with huge volumes of small files. These workloads include: Live tier analytics – Here, the data is analyzed “in place” on the object store rather than being extracted/moved from the object store to another storage tier for access. Tier1 backup – Typically, these are shorter term backups and more critical for restore purposes, so they depend on truly fast access. AI & machine learning (ML) applications – This will often generate and operate on millions of smaller data files. IoT & edge applications – These will require fast access to sensor data, log files, time series data and many other types of smaller files These new application workloads are coming online now and will be naturally deployed through the cloud native model. Cloud-native applications will also make use of RESTful APIs, making the AWS S3 API even more popular than it is today. And, the marriage of object storage and flash media will deliver the combination of high throughput and low latency, ideal for this new world of cloud-native workloads. Surfing the tsunami will be easy with solutions that deliver fast object storage for cloud-native apps New research from leading analysts — like in this Evaluator Group report — document how solutions, such as Scality RING, can now leverage all-flash storage servers to deliver this new exciting level of storage performance to fully support cloud-native applications. Now is the time to prepare!
