Oct 17, 2018 · 23m · mad
Ultra-Resilient SQL for Global Business // Spencer Kimball, Cockroach (FirstMark's Data Driven NYC)
0:00 / 0:00
▶ Watch on YouTube →
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
In this Data Driven NYC presentation and fireside chat, Cockroach Labs CEO Spencer Kimball demonstrates how CockroachDB combines SQL transactional consistency with NoSQL horizontal scalability, explaining how consensus replication and geo-partitioning solve enterprise resilience, multi-region latency, and data sovereignty challenges.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. Matt holds 2.3% of the talking time here. How this is scored →
Matt as informed peer 0.5 Guest teaching 4.2 Guest disagreement 0.2 Matt pushing back 0.2
speaking balance: gold is Matt, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 21:09 Spencer defends mapping oversight on Africa
When an audience member points out that Africa is marked red on his latency map, Spencer politely defuses the implicit criticism by acknowledging it was a facile example rather than a literal deployment limit.
Hardest push from Matt ▶ 21:00 Audience member challenges regional coverageAn audience member presses Spencer on why Africa appeared severely latency-restricted on his global architecture diagram.
Biggest teaching moment ▶ 22:00 Explaining mandatory serializable transaction isolationSpencer educates the audience on PostgreSQL wire protocol compatibility and explains why CockroachDB mandates strict serializable isolation to prevent concurrency vulnerabilities like acid rain attacks.
Matt holds his own ▶ 17:40 Host smoothly transitions to business model Q&AMatt Turck steps in at the end of the presentation to steer the conversation from deep database internals to business growth and commercialization strategy.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Matt as informed peer | Guest teaching | Guest disagreement | Matt pushing back | Why |
|---|---|---|---|---|---|---|
| What is CockroachDB? Bridging SQL and NoSQL | 0 | 3 | 0 | 0 | In this opening presentation segment, Spencer Kimball introduces CockroachDB's mission to bridge SQL transactional guarantees with NoSQL elastic scale. The host does not speak, making host expertise and pushback scores 0. | |
| A Refresher on RTO and RPO Concepts | 0 | 4 | 0 | 0 | Spencer defines Recovery Time Objective (RTO) and Recovery Point Objective (RPO), citing industry trends regarding system recovery degradation. This is a solo presentation without host participation. | |
| A Better Way: Consensus Replication Mechanics | 0 | 5 | 0 | 0 | Spencer details how consensus replication achieves zero RPO and multi-active availability across three regional data centers. The host remains off-mic throughout this technical explanation. | |
| US Expansion: Five Data Center Architecture | 0 | 4 | 0 | 0 | Spencer walks through a five-data-center architecture across the US, demonstrating consensus majority latency and survival through multiple node failures. The host does not speak. | |
| Global Data Architecture and Geo-Partitioning | 0 | 5 | 0 | 0 | Spencer presents global scale topologies, addressing data sovereignty laws and separating global control planes from regional data partitioning. The monologue continues without host intervention. | |
| Summary and Architectural Growth Framework | 3 | 4 | 1 | 1 | Host Matt Turck joins for Q&A, asking high-level questions about commercial growth, open-core licensing, and target customer profiles, while audience members ask technical and mapping questions. The tone remains collaborative and conversational. |
Statements from this episode (19)
Early NoSQL database versions abandoned strong transactional consistency
“Strong consistency in transactions, and those are things that were dropped in the early versions of NoSQL.”
CockroachDB combines NoSQL elastic scalability with SQL strong consistency
“So, you know, Cockroach bridges the gap. It's both of those things. It has both of those things, and it doesn't compromise.”
CockroachDB ran a 50,000-warehouse TPC-C test across 100+ nodes
“We recently ran a TPCC 50 K, which is 50,000 warehouses. It's a pretty big cluster and that was over a hundred plus nodes so, you know, it does scale.”
Under 3% of IT failures were resolved within an hour by 2013
“According to the Forrester research. It's gone from, I think in 2000, ah, seven, it was, ah, something like 13, or 30% of failures were resolved in less than one hour. In 2010, it was, ah, 10%, something like that, and in 2013 it was less than three percent.”
Data center failovers rarely happen in seconds due to manual intervention
“You also expect that the recovery time objective, when you actually have a major failure in a data center, like a replication site can be, you hope it's in the minutes. It's rarely in the seconds, because these things can be slow if they're automated, and some…”
Active-passive is the most common database setup due to conceptual simplicity
“You typically see active-passive. That's the most common because it's probably the, this, the sort of simplest conceptually, and people can make it work.”
CockroachDB's default configuration achieves a 4.5-second recovery time objective
“Our expected time for the way cockroach comes out of the box configured is 4.5 seconds.”
Three-site consensus replication yields 43% higher non-storage resource utilization
“It does require three copies, so you're using three times, ah, three X storage as opposed to two X. But you actually get 43% higher utilization On your actual, ah, you know, resources other than storage.”
Cockroach Labs ran 10,000-warehouse TPC-C benchmarks under active chaos testing
“We've actually run TPCC, 10 K, we probably could run 50 K, but it's actually expensive for us to run 50 K, but 10 K with chaos, and what chaos is, it's kind of like the Netflix model, where you actually bring nodes down.”
CockroachDB's regional cluster architecture closely mirrors Amazon Aurora
“This is a pretty interesting and common way to run Cockroach, and it's similar to what Amazon's doing with Aurora.”
CockroachDB collapses entire SQL transactions into single consensus latency round trips
“When you think of a SQL transaction from the begin to, you know, the various selects to the updates and inserts to the end transaction, we collapse that all into a single consensus latency.”
Google runs AdWords across five data centers to guarantee fault tolerance
“This isn't just sort of a, you know, a random example. This is how Google runs their AdWords system. With five data centers, and the reason is that you can actually have a planned maintenance on a data center, and actually have an unplanned, ah, outage on anot…”
CockroachDB hits full TPC-C throughput globally despite 100-millisecond latency increases
“We've run TPC-C-T-K at this, and we get full throughput. The latencies are just higher, and they're higher by, like, a hundred milliseconds.”
Kimball: 95% of data in global software applications is inherently local
“When you're building a global service, most data, like 95% of it, actually happens to be very local.”
Relying on asynchronous database replication for performance is outdated thinking
“This idea that asynchronous replication is the correct way, maybe the only way to get performant database operation it should be sort of thinking from the past.”
Proper geo-partitioning enables zero RPO and low global database latency
“It's quite possible to have zero RPO and low latencies, even in a global case if you do this kind of partitioning correctly.”
CockroachDB publishes all source code openly without a closed enterprise edition
“We don't have a, an enterprise edition that's closed source. All of our source code is, is available. It's all in the same GitHub repo.”
Comcast uses CockroachDB's geo-partitioning feature within the United States
“Well, one interesting one is Comcast. And, ah, they're using it only in the United States, but they're actually using this geo-partitioning feature.”
CockroachDB exclusively enforces serializable isolation without offering weaker alternatives
“We actually use serializable and it's not even an option. It's everything is serializable, which means your application developers never have to know When they have to do a for select update or something like that in order to lock safely if they're just using …”