SQL Server, with Windows and Office, helped make Microsoft. The database sits in a $13bn business unit and while Microsoft’s server products grew 11 per cent – $1.7bn last year – that was driven primarily by sales of SQL Server.
The man who paved the path was Bob Muglia. A 24-year Redmond veteran, he was SQL Server’s first product manager and led the server and tools business.
That was until 2011, when he was elbowed out by ex-CEO Steve Ballmer in a disagreement over strategy for the server and tools business.
Also, the 2008 DATAllegro acquisition, to build a Netezza-like appliance, had proved to little and too late for Microsoft as the market had moved.
Now Muglia is back, and he is heading up a cloud data warehouse firm. And, rather than run with the Hadoop or NoSQL crowd, Muglia’s back with his first love: SQL. Only it’s SQL on the cloud – yes, AWS.
Snowflake Computing, Muglia’s company since June 2014, launched a product-grade version of its elastic SQL for the cloud on Tuesday: Elastic Data Warehouse. Having been in limited availability, Snowflake Elastic Data Warehouse features data encryption and strong back up with other features designed to tempt serious business users.
Snowflake also announced $45m in series C funding, brining its total to $65m – the company was founded in 2012 by ex-Oracle RAC engineer Beniott Degeville ex-Oracle parallelization expert Thierry Cruanes, and former Vectorwise CEO Marcin Zukowski.
Muglia is now coming after both is ex-employer Microsoft and the NoSQL kids that have flourished in his absence, including Amazon’s Redshift. Why is Muglia backing the SQL pony again rather than swim with the tide on NoSQL?
“For data analysis, SQL has won,” Muglia said. “SQL is the lingua franca of data analysis. All these guys three years ago were talking about NoSQL for data analysis are now falling over themselves to put SQL on their sys.”
He picks on Hadoop, famous for its complexity of set-up and lack of ready tools for the analytics community in business – who, as a result, would normally turn to SQL.
“It’s a flexible system that doesn’t do anything well,” he says. “The vast majority of Hadoop is taking semi structured machine data and munging that into a semi structured form that they spit into existing tools for analysis. It’s a very unsatisfying process.”
He reckons SQL “won” the analytics war because it provides what Muglia considers a mathematically complete model for creating data. So complete that, since it emerged in the 1980s, it’s become the basis for most modern databases and the foundation of a multi-billion dollar industry.
According to Muglia, NoSQL database makers and those flogging Hadoop have come to accept that and are now busy with SQL. And, according to Muglia, they will fail.
“They are trying to put SQL on a different mathematical model and that doesn’t work. Now the vendors are trying to create relational engines on top of Hadoop but Hadoop was never designed for that and it makes it harder for them.”
His proof? Just developing Snowflake to run on AWS NoSQL infrastructure.
Snowflake splits the data part of the database architecture from the compute: data lives as storage blobs on AWS S3 with compute performed on EC2. This de-coupling of storage and compute, according to Muglia, is Snowflake’s secret.
“It lets our customers elastically and infinitely scale the storage and data they have and increase the compute capacity and servers working on the system because we can have multiple warehouses on the same data,” Muglia said.
He denies Snowflake has done anything funky or that it’s some kind of SQL fork or niche. Snowflake is SQL and supports existing SQL databases and apps – it’s compatible with Oracle, SQL Server and Teradata while working with tools like Informatica and Talend. Existing database apps should work on Snowflake, he said, but won’t if your app is – say – optimised to work on Oracle’s OLTP.
As you’d suspect, Muglia is dismissive of Hadoop. It has its place crunching large volumes of data, he says, but not in data warehousing or analytics. Snowflake’s pitch is that it’s suited to structured and “semi-structured” data but won’t handle unstructured data like video files – just like SQL of old.
The problem with that is semi-structured data is a moving target: it starts unstructured and might end up back as unstructured, so you will end up needing Snowflake plus a NoSQL pure play like Hadoop or database.
Target customers are those with lots of data already on the cloud. Snowflake claims 80 early customers, including Adobe, along with unnamed ads, media and gaming firms. It claims to be “engaged” with enterprises concerned their infrastructure “can’t keep up.”
It’s here that Muglia and Snowflake are turning their fire on host Amazon and its Redshift data warehouse. Amazon reckons Redshift is its fastest growing business; Muglia says Snowflake is faster and more flexible than Redshift.
Muglia claims to have competed against Redshift in a “very large percentage” of deals and won “time and time again.” He claims to have “pulled out” installations of Hortonworks and stopped rollouts of Cloudera on the Hadoop front.
Does he run into his old employer in the cloud, with Microsoft Azure? Yes, but only because “Microsoft has a very active salesforce." In general, he reckons, Microsoft is behind in the cloud race. ®