Table of Contents
At intermix.io, we work with companies that build data pipelines and data lakes in the cloud. Some start “cloud-native”, others migrate from on-premise solutions like Oracle or Teradata. What they all have in common though is the one question they ask us at the very beginning:
“How do other companies build their data pipelines?”
And so that’s why we decided to compile and publish a list of publicly available blog posts about how companies build their data pipelines. In those posts, the companies talk in detail about how they’re using data in their business, and how they’ve become “data-centric”.
Here’s the short list of the 14 companies:
11. Dollar Shave Club
And we’re sorry if we missed your post – we’re happy to include it, just fill out this form (take less than a minute). And with that – please meet the 14 examples of data pipelines from the world’s most data-centric companies
Getting data-driven is the main goal for Simple. It’s important for the entire company to have access to data internally. Instead of the analytics and engineering teams to jump from one problem to another, a unified data architecture spreading across all departments in the company allows building a unified way of doing analytics. The main problem then is how to ingest data from multiple sources, process it, store it in a central data warehouse, and present it to staff across the company. Similar to many solutions nowadays, data is ingested from multiple sources into Kafka, before passing it to compute and storage systems. The warehouse of choice is Redshift, selected because of its SQL interfaces, and the ease to process petabytes of data. Finally reports, analytics and visualizations are powered using Periscope Data. In such a way, the data is easily spread across different teams allowing them to make decisions based on data.
Clearbit was a rapidly growing, early-stage startup when it started thinking of expanding its data infrastructure and analytics. After trying out a few out-of-the-box analytics tools (and each of them failed to satisfy the company’s demands) they took building the infrastructure in their own hands. Their efforts converged into a trio of providers: Segment, Redshift, and Mode. Segment is responsible for ingesting all kind of data, combining it, and syncing it daily into a Redshift instance. The main data storage is obviously left to Redshift, with backups into AWS S3. Finally, since Redshift supports SQL, Mode is a perfectly fitted tool which is used to running queries (while using Redshift’s powerful data processing abilities) and creating data insights.
Mode make it easy to explore, visualize and share that data across your organization.
But as data volume grows, that’s when data warehouse performance goes down. With ever increasing calls to your data from analysts, your cloud warehouse becomes the bottleneck. Uncertainty why queries take longer and longer to complete frustrates analysts and engineers alike.
That’s why we’ve built intermix.io, so that Mode users get all the tools they need to optimize their queries running on Amazon Redshift. Here one of our dashboards that shows you how you can track queries from Mode down to the single user:
The whole data architecture at 500px is mainly based on two tools: Redshift – for data storage, and Periscope – for analytics, reporting, and visualization. From a customer-facing side, the company’s web and mobile apps run on top of a few API servers, backed by several databases – mostly MySQL. Data in these DBs is then processed through a Luigi ETL, before storing it to S3 and Redshift. Splunk here does a great job in querying and summarizing text-based logs. Periscope Data is responsible for building data insights and sharing them across different teams in the company. All in all, this infrastructure supports around 60 people distributed across a couple of teams within the company, as of 2015.
The data infrastructure at Netflix is certainly one of the most complex ones, having in mind that they serve over 550 billion events per day, equaling roughly to 1.3 petabytes of data. In general, Netflix’s architecture is broken down into smaller systems, such as systems for data ingestion, analytics, predictive modeling etc. The data stack employed in the core of Netflix is mainly based on Apache Kafka for real-time (sub-minute) processing of events and data. Data needed in the long-term is sent from Kafka to AWS’s S3 and EMR for persistent storage, but also to Redshift, Hive, Snowflake, RDS and other services for storage regarding different sub-systems. Metacat is built to make sure the data platform can interoperate across these data sets as a one “single” data warehouse. Its task is to actually connect different data sources (RDS, Redshift, Hive, Snowflake, Druid) with different compute engines (Spark, Hive, Presto, Pig). Other Kafka outputs lead to a secondary Kafka sub-system, predictive modeling with Apache Spark, and Elasticsearch. Operational metrics don’t flow through the data pipeline but through a separate telemetry system named Atlas.
The tech world has changed dramatically since Yelp was launched back in 2004. Eight years later and Yelp started its change. It transformed from running a huge monolithic application on-premises to one built on microservices running in the AWS cloud. By the end of 2014, there were more than 150 production services running, with over 100 of them owning data. Its main part of the cloud stack is better known as PaaSTA, based on Mesos and Docker, offloading data to warehouses such as Redshift, Salesforce and Marketo. Data enters the pipeline through Kafka, which in turn receives it from multiple different “producer” sources.
Gusto, founded in 2011, is a company that provides a cloud-based payroll, benefits and workers’ compensation solution for businesses. Their business has grown steadily over the years, currently topping to around 60 thousand customers. By early 2015, there was a growing demand within the company for access to data. Up until then, the engineering team and product managers were running their own ad-hoc SQL scripts on production databases. There was obviously a need to build a data-informed culture, both internally and for their customers. When coming to the crossroad to either build a data science or data engineering team, Gusto seems to have done the right choice: first build a data infrastructure which then would support analysts in generating insights and drawing prediction models.
The first step for Gusto was to replicate and pipe all of their major data sources into a single warehouse. The warehouse choice landed on an AWS Redshift cluster, with S3 as underlying data lake. Moving data from production app databases into Redshift was then facilitated with Amazon’s Database Migration Service. On the other side of the pipeline, Looker is used as a BI front-end that teams throughout the company can use to explore data and build core dashboards. Aleph is a shared web-based tool for writing ad-hoc SQL queries. Finally, monitoring (in the form of event tracking) is done by Snowplow, which can easily integrate with Redshift, and as usual, Airflow is used to orchestrate the work through the pipeline.
Building such pipeline massively simplified data access and manipulation across departments. For instance, analysts can simply build their own datasets as part of an Airflow task, and expose it to Looker to use in dashboards and further analyses.
Cloudflare is a web performance and security company that provides online services to protect and accelerate websites online. Online content distribution, web optimization, web security, and analytics are a few examples of the company’s business range.
While different services may require different data stacks to work on, they are all built on top of Cloudflare’s core infrastructure. In the core of their data stack there are Kafka clusters as a streaming platform, and CitusDB as a data warehouse – a scaled up version of PostgreSQL. Data is ingested through Cloudflare’s edge services using HTTP requests, then passed on to Kafka clusters, before getting stored in CitusDB warehouse. A nice example of a service working on top of this infrastructure is the DNS Analysis – a service which processes around 1 million DNS queries per second! The DNS edge service pre-processes and aggregates data, before sending it encrypted to one of Cloudflare’s data centers. Within the data center, data is de-multiplexed and pushed into several Apache Kafka clusters, which in turn pushes data to consumers grouped by Kafka topic. Consumers can store processed information into corresponding DBs which are later queried by the company’s API services and information delivered to customers.
Cloudflare gives their services to millions of websites around the world, processing and storing hundreds of terabytes of data daily. Interestingly, Cloudflare is not a fan of commercial cloud technologies, but they implement their own data centers across the world, in total 152 as of this moment.
Teads is a video advertising marketplace, often ranked as the number 1 video platform in the world. Working with data-heavy videos must be supported by a powerful data infrastructure, but that’s not the end of the story. Teads’ business needs to log user interactions with their videos through the browser (like play, pause, resume, complete…), which count up to 10 million events per day. Another source of data is video auctions (real-time bidding processes) which generate another 60 million events per day. To build their complex data infrastructure, Teads has turned to both Google and Amazon for help.
Originally the data stack at Teads was based on a lambda architecture, using Storm, Spark and Cassandra. This architecture couldn’t scale well, so the company turned toward Google’s BigQuery in 2016. They already had their Kafka clusters on AWS, which was also running some of their ad delivery components, so the company chose a multi-cloud infrastructure. Transferring data between different cloud providers can get expensive and slow. To address the second part of this issue, Teads placed their AWS and GCP clouds as close as possible and connected them with managed VPNs.
So how does their complex multi-cloud data stack look like? Well, first of all, data coming from users’ browsers and data coming from ad auctions is enqueued in Kafka topics in AWS. Then using an inter-cloud link, data is passed over to GCP’s Dataflow which is then well paired with BigQuery in the next step. Having all data in a single warehouse means half of the work is done. The next step would be to deliver data to consumers, and Analytics is one of them. The Analytics service at Teads is a Scala-based app that queries data from the warehouse and stores it to tailored data marts. Interestingly, the data marts are actually AWS Redshift servers. In the final step, data is presented into intra-company dashboards, and the user’s web apps.
You'll get additional resources like:
Remind’s data engineering team strives to provide the whole company with access to the data they need, as big as 10 million daily events, and empower them to directly make decisions. They initially started with Redshift as its source of truth resource for data, and AWS S3 to optimize for cost. While S3 is used for long-term storage of historical data in JSON format, Redshift only stores the most valuable data, not older than 3 months. The company uses Interana to run custom queries on their JSON files on S3, but they’ve also recently started using AWS Athena, as a fully managed Presto system – to query both S3 and Redshift databases. The move for Athena also triggered a change in the data format – from JSON to Parquet, which they say was the hardest step in building up their data platform. An EMR/Hive system is responsible for doing the needed data transformations between S3 and Athena. In the data ingestion part of the story, Remind gathers data through their APIs from both mobile devices and personal computers, as the company business targets schools, parents and students. This data is then passed to a streaming Kinesis Firehose system, before streaming it out to S3 and Redshift.
Remind’s future plans are probably focused on facilitating data format conversions using AWS Glue. This step would allow them to replace EMR/Hive from their architecture and use Spark SQL instead of Athena for diverse ETL tasks.
Robinhood is a stock brokerage application that democratizes access to the financial markets, which enables its customers to buy and sell U.S. listed stocks and ETFs with zero commission. The company debuted with a waiting list of nearly 1 million people, which means they had to pay attention to scale from the very beginning.
Robinhood’s data stack is hosted on AWS, and the core technology they use is ELK (Elasticsearch, Logstash, and Kibana) – a tool for powering search and analytics. Logstash is responsible for collecting, parsing and transforming logs, before passing them on to Elasticsearch, while data is visualized through Kibana. They grew up from a single ELK cluster with a few GBs of data to three clusters with over 15 TBs. Before data goes to ELK clusters, it is buffered in Kafka, as the rates of which documents enter vary significantly between different data sources. Kafka also shields the system from failures and communicates its state with data producers and consumers. As with many other companies, Robinhood uses Airflow to schedule various jobs across the stack, beating competition such as Pinball, Azkaban and Luigi. Robinhood data science team uses Amazon Redshift to help identify possible instances of fraud and money laundering.
Dollar Shave Club (DSC) is a lifestyle brand and e-commerce company that’s revolutionizing the bathroom by inventing smart, affordable products. Don’t be fooled by their name, they have a pretty cool data architecture, for a company in the shaving business. Their business model works with online sales through a subscription service. Currently, they serve around 3 million subscribed customers.
DSC’s web applications, internal services, and data infrastructure are 100% hosted on AWS. A Redshift cluster serves as the central data warehouse, receiving data from various systems. Data movement is facilitated with Apache Kafka and can move in different directions – from production DBs into the warehouse, in between different apps and in between internal pipeline components. There’s also Snowplow which collects data from the web and mobile clients. Once data reaches Redshift, it is accessed through various analytics platforms for monitoring, visualization, and insights. The main tool for the job is, of course, Apache Spark, which is mainly used to build predictive models, such as recommender systems for future sales.
Coursera is an education company that partners with the top universities and organizations in the world to offer online courses. They started building their data architecture somewhere around 2013, as both numbers of users and available courses increased. As of late 2017, Coursera provides courses to 27 million worldwide users.
Coursera collects data from their users through API calls coming from mobile and web apps, their production DBs, and logs gathered from monitoring. A backend service called “eventing” periodically uploads all received events to S3 and continuously publishes events to Kafka. The engineering team has selected Redshift a central warehouse, offering much lower operational cost when compared with Spark or Hadoop at the time.
On the analytics end, the engineering team created an internal web-based query page where people across the company can write SQL queries to the warehouse and get the needed information. Of course, there are analytics dashboard across the company which are refreshed on a daily basis. Finally, many decisions made in Coursera are based on machine learning algorithms, such as A/B testing, course recommendations, understanding student dropouts and others.
Wish is a mobile commerce platform. It provides online services that include media sharing and communication tools, personalized and other content, as well as e-commerce. During the last few years, it grew up to 500 million users, making their data architecture out of date.
The data architecture at Wish, before scaling up, had two different production databases: a MongoDB one storing user data, and a Hive/Presto cluster for logging data. Data engineers had to manually query both to respond to ad-hoc data requests, and this took weeks at some points. Another small pipeline orchestrated by Python crons, also queried both DBs and generated Email reports.
After rethinking their data architecture, Wish decided to build a single warehouse using Redshift. Data from both production DBs flowed through the data pipeline into Redshift. BigQuery is also used for some types of data. It feeds data into secondary tables needed for analytics. Finally, analytics and dashboards are created with Looker.
You'll get additional resources like:
Blinkist transforms the big ideas from the world’s best nonfiction books into powerful little packs users can read or listen to in 15 minutes. At first, they started selling their services through a pretty basic website, and monitored statistics through Google Analytics. Unfortunately, visitor statistics gathered from Google Analytics didn’t match the ones engineers computed. This is one of the reasons why Blinkist decided to move to the AWS cloud.
They choose a central Redshift warehouse where data flow in from user apps, backend and web frontend (for visitors tracking). To get data to Redshift, data is streamed with Kinesis Firehose, also using Amazon Cloudfront, Lambda and Pinpoint. The engineering team at Blinkist is working on a newer pipeline where ingested data comes to Alchemist, before passing it to a central Kinesis system, and onwards to the warehouse.
We hope this post along with its 14 examples gives you the inspiration to build your own data pipelines in the cloud.
If you don’t have any data pipelines yet, it’s time to start building them. Begin with baby steps and focus on spinning up an Amazon Redshift cluster, ingest your first data set and run your first SQL queries.
After that, you can look at expanding by adding a dashboard for data visualization, and schedule a workflow, to build your first true data pipeline. And once data is flowing, it’s time to understand what’s happening in your data pipelines.
That’s why we built intermix.io. We give you a single dashboard to understand when & why data is slow, stuck, or unavailable.
With intermix.io you can:
Our customers have the confidence to handle all the raw data their companies need to be successful. What you get is a real-time analytics platform that collects metrics from your data infrastructure and transforms them into actionable insights about your data pipelines, apps, and users who touch your data.
Setting up intermix.io takes less than 10 minutes, and because you can leverage our intermix.io experts, you can say goodbye to paying for a team of experts with expensive and time-consuming consulting projects. We can help you plan your architecture, build your data lake and cloud warehouse, and verify that you’re doing the right things.
It’s easy – start now by scheduling a call with one our of experts.
At a glance, batch data processing seems simple. Pull data from a source, apply some business logic to it, and load it for later use. When done well, automating these jobs is a huge win. It saves time and empowers decision-makers with fresh and accurate data. But this kind of ETL (Extract, Transform, and Load) process gets very tricky, very quickly. A failure or bug in a single step of an ETL process has cascading effects which can require hours of manual intervention and cleanup. These kinds of issues can be a gigantic time sink for the analysts or engineers that have to clean up the mess. They negatively impact data quality. They erode end-user confidence. And they can make teams terrified of pushing out even the most innocuous changes to a data pipeline. To avoid falling into this kind of ETL hell, analysts and engineers need to make data pipelines that:
In short, they need to make data pipelines that are idempotent. Idempotent is a mathematical term describing an operation which can be applied arbitrarily many times without changing the result. Idempotent data pipelines are fault-tolerant, reliable, and easy to troubleshoot. And they’re much simpler to build than you might expect. Here are some simple steps you can take to make your data pipeline idempotent.
See 14 real-life examples of data pipelines built with Amazon Redshift
The easiest way to be sure that a buggy or failed ETL run doesn’t dirty up your data is to isolate the data that you deal with in a given run. If you’re working with time-series data this is fairly straightforward. When pulling data from your initial source, simply filter it by time. Operate on an hour, a day, or a month, depending on the volume of your data. By operating on data in time-windowed batches like this, you can be confident that any bugs or failures aren’t impacting your data quality beyond a single window. If you’re working with cross-sectional data, try to deal with some explicitly-defined subset of that data in a given run. Knowing exactly what data has been impacted by a bug or failure makes clean up and remediation of data pipeline issues relatively straightforward. Just delete the data that has been impacted from your final destination and then run again. And if you isolate your data correctly, you never have to worry about duplicated or double-counted data points.
In order to avoid time-consuming cleanup of failed or buggy data pipeline runs, it’s important to build your pipeline in such a way that it either loads everything or it loads nothing at all. That is to say, it should be atomic. If you’re loading into a SQL database, achieving atomicity for your data pipeline is as simple as wrapping all of your loading statements–inserts, copies, creates, or updates–in a transaction with a BEGIN statement at the beginning and an END or COMMIT statement at the end. If the final destination for your data is not a SQL database–say an Elasticsearch cluster or an Amazon S3 bucket–building atomicity into your pipeline is a little more tricky. But it’s not impossible. You’ll simply need to implement your own “rollback” logic within the load step. For instance, if you’re writing CSV files to an Amazon S3 bucket from a python script, maintain a list of the resource identifiers for the files you’ve written. Wrap your loading logic in a try-except-finally statement which deletes all of the objects you’ve written to if the script errors out. By making your data pipeline atomic, you completely obviate the need for time-consuming cleanup after failed runs.
Now that you’ve explicitly isolated the data your pipeline deals with in a given run and implemented transactions, having your pipeline clean up after itself is a breeze. You just need to delete before you insert. This is the most efficient way to avoid loading duplicate data. Say your data pipeline is loading all of your company’s sales data from the previous month into a SQL database. Before you load it, delete any data from last month that’s already in the database. Be sure to do this within the same transaction as your load statement. This may seem scary. But if you’ve implemented your transaction correctly, your pipeline will never delete data without also replacing it. Many SQL dialects have native support for this type of operation, which is typically referred to as an upsert. For instance in PostgreSQL (which is also the SQL dialect used by Amazon Redshift) you could prevent loading duplicate data with the following:
INSERT INTO sales (employee, customer, contract_number, annual_revenue, date)
VALUES ('Janet', 5439, 50000, '2018-07-20')
ON CONFLICT (contract_number)
DO UPDATE SET employee = excluded.employee;
This SQL statement will attempt to insert a new row into the database. If a row already exists with contract number 5439, the existing row will be updated with the new employee name. In this way, duplicate rows for the same contract will never be loaded into the database. By having your pipeline clean up after itself and prevent the loading of duplicate data, backfilling data becomes as simple as rerunning. If you discover a bug in your business logic or you want to add a new dimension to your data, you can push out a change and backfill your data without any manual intervention whatsoever.
ETL pipelines can be fragile, difficult to troubleshoot, and labor-intensive to maintain. They’re inevitably going to have bugs. They’re going to fail every once in a while. The key to building a stable, reliable, fault-tolerant data pipeline is not to build it so that it always runs perfectly. The key is to build it in such a way that recovering from bugs and failures is quick and easy. The best data pipelines can recover from bugs and failures without ever requiring manual intervention. By applying the three simple steps outlined above, you’re well on your way to achieving a more fault-tolerant, idempotent data pipeline.
One of the core challenges of using any data warehouse is the process of moving data to a place where the data can be queried. Amazon Redshift COPY command provides two methods to access data:
1- copy data into Redshift local storage by using the COPY command
2- use Amazon Redshift Spectrum to query S3 data directly (no need to copy it in)
This post highlights an optimization that can be made when copying data into Amazon Redshift.
The Amazon Redshift COPY command loads data into a table. The files can be located in an S3 bucket, an Amazon EMR cluster, or a remote host that is accessed using SSH. The maximum size of a single input row from any source is 4 MB. Amazon Redshift Spectrum external tables are read-only. You can’t COPY to an external table.
The COPY command appends the new data to the table. In Amazon Redshift, primary keys are not enforced. This means that deduplication must be handled by your application.
The recommended way of de-duplicating records in Amazon Redshift is to use the UPSERT process.
UPSERT is a method of de-duplicating data when copying into Amazon Redshift. The UPSERT operation merges new records with existing records using primary keys. While some RDBMSs support a single “UPSERT” statement, Amazon Redshift does not support it. Instead, you should use a staging table for merging records.
Here is an example of an “UPSERT” statement for Amazon Redshift.
-- Start transaction BEGIN;
-- Create a staging table
CREATE TABLE my_staging (LIKE cities);
-- Load data into the staging table
COPY my_staging (name, zipcode, state)
-- Update records
SET name = s.name, zipcode = s.zipcode
FROM my_staging s
WHERE name.id = s.id;
-- Insert records
INSERT INTO cities
SELECT s.* FROM my_staging s LEFT JOIN cities
ON s.id = cities.id
WHERE cities.id IS NULL;
-- Drop the staging table
DROP TABLE my_staging;
-- End transaction
The default behavior of Redshift COPY command is to automatically run two commands as part of the COPY transaction:
Amazon Redshift runs these commands to determine the correct encoding for the data being copied. This may be useful when a table is empty. But in the following cases the extra queries are useless and thus should be eliminated:
In the below example, a single COPY command generates 18 ‘analyze compression’ commands and a single ‘copy analyze’ commands.
Extra queries can create performance issues for other queries running on Amazon Redshift. Extra queries may saturate the number of slots in a WLM queue, thus causing all other queries to have queue wait times.
The solution is to adjust the COPY command parameters to add “COMPUPDATE OFF” and “STATUPDATE OFF”. These parameters will disable these features during “UPSERT”s. Here is an example of a “COPY” command with these options set.
-- Load data into the staging table
COPY users_staging (id, name, city)
COMPUPDATE OFF STATUPDATE OFF;
App Tracing surfaces important information about how apps & users interact with your data. It can help answer questions like:
There are three categories of data apps:
App Tracing requires the data app to annotate the executed SQL with a comment. The comment encodes metadata about the application which submitted this query.
Intermix.io will automatically index all data contained in the annotation, and make it accessible as first-class labels in our system. I.e. for Discover searches, Saved Searches, and aggregations in the Throughput Analysis page.
Out of the box, we support:
In the below example, a query spike in WLM 3 causes a bottleneck in query latency. The result is that queries which would otherwise take 13-14 seconds to execute, are stuck in the queue for > 3 minutes.
App Tracing detects that the majority of these queries are from Looker. How do you know which user is causing this?
Click on the chart, and a widget will pinpoint the specific Looker user(s) who ran those queries. In this example, we see that user 248 is responsible.
Armed with this information, you can now:
See all the activity for this user by heading to Discover and use the new ‘App’ filter to search for Looker user 248.
To set up an alarm to get email notifications, save that search and stream the following metrics to CloudWatch:
We soft-launched app tracing on the morning of this blog post. It didn’t take our customer long to notice. See this screenshot of Slack conversation (each one of our customers has a direct line to our team) we had today.
If you’re using Amazon Redshift in combination with Apache Airflow, and you’re trying to monitor your DAGs – we’d love to talk! We’re running a private beta for a new Airflow plug-in with a few select customers. Go ahead and and click on the chat widget on the bottom right of this window. Answer three simple questions, schedule a call, and then mention “Airflow” at the end and we’ll get you set up! As a bonus, we’ll throw in an extended trial of 4 weeks instead of 2!
Photo by Denise Johnson
Since Amazon Redshift launched in 2013, customer keep asking us “what are some of the uses cases for Amazon Redshift?”. It’s 2017, and with the four years since launch, that’s a long time in technology. Key things that have changed since then:
Each bullet sort of feeds the other, and it’s somewhat of a positive feedback loop. When you can process more data faster, and do more stuff with it – your use cases will evolve.
Redshift started out as a simpler, cheaper and faster alternative to legacy on-premise warehouses. If you read Jeff Barr’s blog post announcing Redshift (“Amazon Redshift – The New AWS Data Warehouse”), the pitch was all about simplicity and price.
Fast forward to the end of 2017, and the use cases are way more sophisticated than just running a data warehouse in the cloud. I recommend looking at the slides from ReInvent 2016, “What’s New With Redshift” (also see full video of the talk on YouTube).
You will find 4 uses cases:
See 14 real-life examples of data pipelines built with Amazon Redshift
Data warehousing has been around since Kimball and Inmon. What changed with Amazon Redshift was the price at which you can get it – about 20x less than what you had to carve out going with the legacy vendors like Oracle and Teradata.
The use case for data warehousing is to unify disparate data sources in a single place and run custom analytics for your business.
Let’s say you‘re the head of business intelligence for a web property that also has a mobile app. The typical categories of data sources are:
With a rich ecosystem of data integration vendors, it’s easy to build pipelines to those sources and feed data into Redshift. Put a powerful BI / dashboard tool on top, and you have a full-blown BI stack.
A key advantage of Redshift that I think a lot of people are not aware of is simplicity. It used to take months if not quarters to get a data warehouse up and running. And you’d need the help of an Accenture or IBM. None of that anymore. You can spin up a Redshift cluster in less than 15 minutes, and build a whole business intelligence stack in a weekend.
Some examples by teams who built an early analytics stack on Redshift:
The combination of price / speed / simplicity expanded the addressable market for data warehousing from large corporations to SMBs. However, because it is so easy to get going, data engineers must make sure to follow the best practices when setting up their cluster and avoid any performance issues they might have data volume and pipeline complexity grows.
Because of the cost of storage, in previous generations of data warehouses you had to aggregate data. It was too expensive to store raw data. That changed with Amazon Redshift. Because Redshift is cheap, it’s possible to store raw, event-level data without getting killed on storage cost.
Event-level data comes with three key benefits.
And so that made Amazon Redshift a perfect fit for analyzing machine-generated data like web logs or clickstream data. Massive amounts of data that come in at high velocity.
Because Redshift is fast and cheap, processing machine data data is cost-effective, and you can drive the time required for “ingest-to-insight” (i.e. the time between pushing data into Redshift and the final analysis output) below the 5 minute mark. Not just for basic aggregations, but complex 10-way joins, across billions (billions with a “b”) of rows. That’s remarkable.
The business value comes with exposing that data back into the business. For decision making, data-driven services, etc. Here are a few links of tech talks I recommend reading up on, they describe in detail how Yelp, Lyft and Pinterest use Amazon Redshift to process data and then expose it to services that need it.
The business value here goes beyond mere cost savings by migrating your warehouse to the cloud. Rather, you’re enabling new services, informed by data. These “data-driven services” are the foundation for better / faster decision making, and also new revenue-generating products.
That distinction is key. In previous days, with the use of basic reporting, companies would look at a data warehouse as a “cost center”. Yes, data is important, but also expensive. And so the goal was to keep that cost down as much as possible. Limited exposure of data to a limited set of people, etc.
Now, it makes sense to increase spend into your data infrastructure. That’s because an incremental $ spent on analyzing data can lead to a larger incremental increase in $ revenue generated.
That leads us to the next use case, where Redshift drives new revenue as the core engine behind an analytics product, i.e. business applications.
Not all companies have the technical abilities and budget to build and run a custom streaming pipeline with near real-time analytics.
But analytical use cases can be pretty similar across a single industry or vertical. That has given rise to “analytics-as-a-service” vendors. They use Redshift under the covers, and then offer analytics in a SaaS model to their customers.
These vendors either run a single cluster in a multi-tenant model, or offer a single cluster to customers in a premium model. Take Acquia Lift as an example. The charging model then is a subscription fee to the analytics service.
To give you some back-of-the-envelope math: In a multi-tenant model, you can cram data from 10s of customers onto a single node cluster, which costs you ~$200 / month. Price out the actual analytics service at $500 / month / subscriber, and you have a business with some pretty good gross margins.
Using Redshift for mission-critical workloads has emerged in the past few years. Here, data sitting in Redshift feeds into time-sensitive apps. It’s key that the the database stays up, because otherwise the business goes down (quite literally).
In some cases, e.g. the NASDAQ, that is daily reporting. That reporting can’t be late or wrong, otherwise somebody might quite literally go to jail.
Other cases include building predictive models on top of Redshift, and then embed the results programmatically into another app, via a data API. An example is automated ad-bidding, where bids across certain ad networks are adjusted on a near real-time basis. The adjustments are calculated on ROI and what ad types performed best in the last week, day and even hour.
Redshift has driven down the cost of running a data warehouse, and as a result expanded the addressable market. Because Redshift is cheap, it allows to store event-level data, which opens up a whole new world of use cases. Some of these use cases include data-driven services that create new revenue streams for companies.
And when data in Amazon Redshift becomes critical for business success, it’s important to make sure your cluster is not a black box. If you’re part of a data team that’s building mission-critical data pipelines, sign-up for a free trial. We’re sure you’ll be surprised by the amount of visibility you’ll get, and how much faster that allows you to move.