Showing posts from August, 2022

Socrates: The New SQL Server in the Cloud (Sigmod 2019)

This paper (Sigmod 2019) presents Socrates, the database-as-a-service (DBaaS) architecture of the  Azure SQL DB Hyperscale. Deploying a DBaaS in the cloud requires an architecture that is cost-effective yet performant. An idea that works well is to decompose/disaggregate the functionality of a database into two as compute services (e.g., transaction processing) and storage services (e.g., checkpointing and recovery). The first commercial system that adopted this idea is Amazon Aurora. The Socrates design adopts the separation of compute from storage as it has been proven useful. In addition, Socrates separates database log from storage and treats the log as a first-class citizen. Separating the log and storage tiers disentangles durability (implemented by the log) and availability (implemented by the storage tier). This separation yields significant benefits: in contrast to availability, durability does not require copies in fast storage, in contrast to durability, availability does n

Fast Serializable Multi-Version Concurrency Control for Main-Memory Database Systems

This paper from Sigmod 2015 describes the addition of Multi-Version Concurrency Control (MVCC) to the Hyper database, and discusses at length how they implement serializability efficiently with little overhead compared to snapshot isolation. Coming only two years later, this paper seems like a response/one-up-manship to the Hekaton paper . In the paper, you can see at many places statements like "in constrast to/ unlike/ as in/ Hekaton" phrase. In a way, that is the biggest compliment a paper can pay to another paper. Hyper is also an in-memory database as in Hekaton. The paper says that Hyper is like Hekaton in that, in order to preserve scan performance, new transactions are not allocated a separate place. The most uptodate version of the database is kept all in the same place. But, you can reach older versions of a record (that are still in use by active transactions) by working backwards from the record using the delta versions maintained. The biggest improvement in Hyp

Hekaton: SQL Server’s Memory-Optimized OLTP Engine

This paper (Sigmod 2013) gives an overview of the design of the Hekaton engine which is optimized for memory resident data and OLTP workloads and is fully integrated into Microsoft SQL Server. The paper starts with a strongly opinionated paragraph: SQL Server and other major database management systems were designed assuming that main memory is expensive and data resides on disk. This assumption is no longer valid; over the last 30 years memory prices have dropped by a factor of 10 every 5 years. Today, one can buy a server with 32 cores and 1TB of memory for about $50K and both core counts and memory sizes are still increasing. The majority of OLTP databases fit entirely in 1TB and even the largest OLTP databases can keep the active working set in memory. I really like this paper. The paper is self-contained and is easy to follow. It is very well written. It has the old-school touch: it is succinct and opinionated, it explains principles/tenets behind design, and it does not oversell

The Seattle Report on Database Research (2022)

Every 5 years, researchers from academia and industry gather to write a state-of-the-union (SOTU) report on database research. This one was released recently . It is a very readable report, and my summary consists of important paragraphs clipped from the report. Emphasis mine in bolded sentences. I use square brackets for when I paraphrase a long text with a more direct statement. TL;DR: The SOTU is strong (the relational database market alone has revenue upwards of $50B) and growing stronger thanks to boom in cloud computing and machine learning (ML). For the next 5 years, research should scale up to address emerging challenges in data science, data governance, cloud services, and database engines.   What has changed in the last 5 years Over the last decade, our research community pioneered the use of columnar storage, which is used in all commercial data analytic platforms. Database systems offered as cloud services have witnessed explosive growth. Hybrid transactional/analytical pro

Strict-serializability, but at what cost, for what purpose?

Strict-serializability guarantees that transactions appear to occur in an order consistent with the "real-time" ordering of those transactions: If transaction T1 commits before transaction T2 is invoked, then the commit timestamp of T1 precedes the commit timestamp of T2. This is, in fact, the real-time constraint from linearizability , but applied across transactions not just per-key. A strict-serializability system satisfies both serializability (transactions appear to occur as if they are executed one at a time in isolation) and linearizability per key (after all single-key reads/writes are transactions over one item). Below figure is from . However, this is a one-way implication, the other direction does not hold. You can satisfy both serializability per transactions and linearizability per key, but fail to satisfy strict-serializability. (Below I give an example accompanied with a TLA+ specification to check it.) This is because, in strict-

Popular posts from this blog

Learning about distributed systems: where to start?

Hints for Distributed Systems Design

Foundational distributed systems papers

Metastable failures in the wild

Scalable OLTP in the Cloud: What’s the BIG DEAL?

The end of a myth: Distributed transactions can scale

Always Measure One Level Deeper

Dude, where's my Emacs?

There is plenty of room at the bottom

Know Yourself