The Use Cases of Stealth

Stealth’s industry-leading efficiency unlocks significant savings in a number of applications - from simple data archival to complex multi-data center environments. The four key use cases below are those which we have spent developing through the five years of Cyborg.


1 | Stealth for Database

Databases are the hungriest systems in the cloud - in terms of computing, memory, storage, and bandwidth costs. Using compression to alleviate some costs is nothing new. However, using compression for all four categories, is a first.

web_db.png

Stealth’s best-in-class efficiency means that more can be done with less: when Stealth replaces a legacy algorithm, it completes the same compression tasks in the DBMS storage engine at a fraction of the cost. While a legacy algorithm, such as gzip, might require 15% of CPU for compression and decompression, Stealth would only need a mere 5% — opening up an extra 10% which can be repurposed for revenue-generating purposes.

For instance, on a single AWS DB server, this 10% can represent $3,000/year. On a cloud scale, this number can easily reach millions.

We have developed, or are currently developing, support for the following DBMS:

 
  • MySQL
  • PostgreSQL
  • MongoDB
  • Cassandra
  • MariaDB
  • Hadoop (HDFS, HBase)
 

2 | Stealth for Machine Learning

web_ml.png

Before machine learning (ML) can achieve the incredible results we see today, each model needs to be trained. This training involves streaming massive amounts of data, typically on a terabyte-scale, prior to reaching production status. This poses a new set of challenges: storing this training data, streaming it without bottleneck, and maintaining reasonable costs.

In comes Stealth. By combining our high-efficiency algorithm and our versatile plugins for DBMS, we are able to embed Stealth directly into ML libraries to perform compression and decompression. The advantage: training data can be stored in a compressed format (saving 50-95% in storage); training data can be streamed compressed (saving 50-95% network bandwidth); and shortening training time for IO-bottlenecked systems.

Our development has largely been focused on Tensorflow so far. We hope to expand to other ML libraries in the near future.


3 | Stealth for Data Migration

Migrating petabyte-scale loads, whether from on-premise to the cloud, or from cloud to cloud, is an expensive affair. Countless solutions exist, from dedicated fiber uplinks to 18-wheeler data transfer devices. Compression is also used in the field, however, it is often used at the mercy of the data source.

web_dm.png

The utter compression performance of Stealth, besting the closest competitor by around 50%, opens up the possibility of real-time compression. Petabyte-scale loads, which may take days, weeks, or months to transfer, can now be compressed in real-time during transfer. This effectively reduces bandwidth use by 50-95%, shortens transfer time by the same amount, and has little effect on compute footprint.

We are currently working with cloud computing providers to embed Stealth into their data migration products. No release date has been set yet.


4 | Stealth for Internet of Things

web_iot.png

The Internet of Things (IoT) has had a profound impact on every industry. Its worldwide rollout is the driving factor behind technologies like 5G and millimeter-wave wireless. According to Ericsson, IoT devices will make up 70% of wide-area networks by 2022. This explosive growth is creating a boom in wireless bottlenecks globally.

The lightweight nature of Stealth, combined with its aptitude for small-data compression, makes it the perfect candidate for embedded IoT compression. By running locally on IoT devices, Stealth can perform real-time compression, reducing bandwidth load by up to 75% - while having negligible effect on latency.


I’m sold! Where can I get it?

Stealth is in active development, being deployed to a small number of beta customers. These include large cloud computing providers and small businesses alike.

If your organization is interested in learning more, or joining our private beta, please get in touch with us.