SIGMOD'23: Industry talks, and Ted Codd Innovation Award

-

Sponsored industry talks

Amazon: Innovation in AWS Database Services — Marc Brooker

Marc, who is a VP/Distinguished Engineer at AWS, talked about systems issues in AWS database services. (If you are not following Marc's blog feed, you should take a couple seconds and fix this problem now.) He talked about the enormous scale that the AWS databases deals with by mentioning that the DynamoDB processed 89 million req/sec during Prime Day. The problem is not only to scale up, but also down. There is a spectrum of scale, with billion times difference in scale. Systems support is need to enable scale up and down. Consider Aurora Serverless v2, which launched couple years ago. We had to solve problems like how to grow and shrink in place, how much cache to allocate, and how much temporary storage to use. (CPU provisioning was easier, since it is elastic.) The logic and reasoning of the 5 minute rule by Gray/Putzolu still remain relevant, although the constants changed since 1986. We still care about latency/throughput and variance in them. Workload placement is another important question. Which workload do we move? The growing one, idle one, or the busy one?

Another theme Marc talked about is the composition of systems from components. Consider MemoryDB, which is a durable fault-tolerance version of redis. We used a distributed transaction log called journal and treated the system as materialized views from the journal. Since journal is an important component that other systems build upon, it is important to invest in that component in terms of ownership, performance, and efficiency, and formal correctness (via TLA+, P, etc.).


Salesforce: Enterprise and the Cloud: Why Is It Challenging? — Pat Helland

Pat is a fount firehose of wisdom. If you haven't seen him talk, you can get a taste here or here.

Pat compared enterprise and cloud computing and then talked about how to reconcile the two. He said enterprise computing is a very tough problem, where durability and low-latency transactions are prioritized. Traditional enterprise computing means you are standing on bedrock, rock solid stuff. It is a lot like driving in an F1 track. Compare this to the cloud, which is like driving in public highways. It is cheap and good most of the time, but it is not the bedrock, rock solid stuff the traditional enterprise computing built on. The cloud philosophy is that many small servers beat fewer large servers. That is, horizontal scalability beats vertical scalability. In the cloud, it is easy to do fast and pretty good, or retriable. But fast and perfect is hard. For example, updating in place, and consensus/quorum is hard. As we move to cloud, we work to deal with this. Another area of work is tolerating gray failures and metastable failures.


Confluent: Consensus in Apache Kafka: from Theory to Production — Jason Gustafson and Guozhang Wang

OK, let's talk about Kafka's control plane needs. Kafka is a streaming platform. Data is organized as partitioned topics by brokers, then data is replicated and log-structured stored, and consumed. As the VLDB15 paper discussed, Kafka circa 2013 used Zookeeper for consensus on metadata (including leader election), and for data replication, it used leader follower replication. Zookeeper governed the control plance, and a single broker acting as a controller propagated all data to followers.

But 2013 was before cloud domination (the presenter's words, not mine!). There was an explosion on metadata, because new features meant new metadata. They then asked the question of how to scale Kafka clusters efficiently in the cloud. The answer involved getting rid of the ZooKeeper as a dependency. By using an event-based variant of the Raft consensus protocol, Kraft greatly consolidating responsibility for metadata into Kafka itself, rather than relying on ZooKeeper. KRaft mode also makes use of a new quorum controller service in Kafka which replaces the previous controller.

They model checked for correctness with TLA+. This doesn't prevent production incidents of course. An example was Kafka 15019: broker heartbeat timeouts. Kraft was put in production on 2000+ clusters. It is the default for new clusters in all regions in AWS, GCP, and Azure. Look for the VLDB'23 paper titled "Kora: the cloud native engine for Kafka" for
 more information.


Google: Data and AI at Google BigQuery Scale — Tomas Talius

Big query is the core of Googles' data cloud used by many data professionals. It provides cross cloud support, geospatial analytics, stream analytics, big query ML, and log analytics. Big query is extremely disaggregated (I am pretty sure they didn't use the word in Gustavo's sense):

  •  colossus storage
  •  shuffle (mindmeld) in memory
  •  compute via borg
  •  scalable metadata via spanner

The talk was a high level features overview, and was low on technical details. It mentioned big query serverless and autoscaling for paying what you use in small increments. It mentioned Bigquery ML, and ML for all SQL users, being run over Google TPUs. It also mentioned the potential of using  LLM models in SQL.


Alibaba: Enhancing Database Systems with AI — by Bolin Ding and Jingren Zhou


This talk discussed how AliBaba enhances their database systems with AI. Like the Google talk, all I was able to get from this talk was that I should be excited because ML is being adopted for database services. These look like they are all research prototypes at this stage. The talk didn't give me technical discussion of how these are being incorporated. So I will just list some keywords I caught, without much semantic around them. I apologize for string of incomplete sentences.

Database autonomy service(DAS) in Alibaba cloud databases collects statistics, and recommends actions. Use cases may be  anomaly detection, treatment, SQL query optimization, recommending new sizes, performance diagnosis, devops for cloud databases as a microservice system. Timeseries analysis, cardinality estimation. Learning to be a statician; prepare training data points and feed into an NN model Wu et al vldb22. NN+ regularization as an NDV estimator. https://github.com/wurenzhi/learned_ndv_estimator Learned query optimizer. Lero: learning to rank query optimizer, Zhu et al vldb23. Plan explorer, plan comparator, and model trainer. Database tuning; challenges and opportunities of configuration auto-tuning for cloud databases.

Even at this form, this sounds exciting.


Microsoft: Microsoft Fabric - Analytics in the AI Era — Raghu Ramakishnan


The talk was about united analytics in the AI era. Fabric is MS intelligent data platform. (AFAIR, Fabric was a consensus system for control plane at Microsoft. Microsoft seems to like that name and reuse it.) A use case is persona optimized experiences in office. The "Polaris: distributed sql engine in azure synapse paper" from a VLDB'20 looked at it.


Raghu mentioned an eclectic mix of features around Fabric, including lakehouses and warehouses, one copy for all computes, delta-parquet format, going from row/page orient data layout to column oriented log-structured parquet format. Related to the last item,  he mentioned a recent paper called, lst-bench, which sounds interesting. He also talked about data co-pilots: "we can produce good enough SQL, your productivity will go off the roof". Similar to the Google and Alibaba talks, this talk also advertised exciting developments, but at high level, and I wasn't able to get a good appreciation of systems challenges/insights in getting these to work.


Edgar F Codd Innovations Award: Joe Hellerstein

I am a big fan of Joe's and his students' work. I covered many of his work in this blog, including the CALM line of work. I was happy to see him recognized with this Innovation Award.

Joe had prepared a visually stunning presentation! He kept it on the theme of the award. We saw slide after slide of artwork showing Ted Codd in different context, as a robot, in the clouds etc. I thought Joe hired artists to prepare the slides. Turns out it was all prepared by him, with the help of generative AI.


Here is Joe talking about his research work.  

The work takes a broader perspective of declarativity. Ted Codd said 1/ formalize your specifications, 2/ automate your implementations, and 3/ keep them separate. Joe took a broader perspective on Codd's lesson. In his words, relational databases are an artificial limitation. Data is everywhere, so we applied the lessons everywhere. This resulted in work on :

  • Declarative networking
  • Declarative data science and machine learning
  • Declarative programming for the cloud: The cloud is the biggest computer, the humans assembled, programming it in Java is an abomination. Cloud is invented to hide computing resources and how computations are executed.

This last part got more coverage, because this is what Joe and his colleagues are working on with the Hydro project.

Of course, the talk would not be complete without talking about generative AI. Joe speculated about the application of generative AI to data systems in two parts.

Programming: The narrow waist is declarative specifications. If you generate Rust with generative AI compiler complains. If you generate Python with generative AI it runs, but you don't know what it does. This is an enormous opportunity for programming.

Interaction: Guide-decide loop, pointing at data visualization and predictive interaction (CIDR 2015).



Joe had a lengthy gratitudes section in his talk. He credited many people including Simon, Judith, Lisa, Billie, all profs, Meichun Hysu, Hamid Pirahesh, Mike Stonbraker, Jeff Naughton, Mike, Rober Wilensky, Christos, David Culler, colleagues in Berkeley db, and fellow data wranglers at trifacta. He said this
award is really an award for my students and postdocs. His wife and one of his daughters was also attending the ceremony.



Since Joe is such a mensch, he had a short talk. So, the organizers started a Q&A session, after which Joe got a standing ovation from close to 1000 SIGMOD/PODS participants.

Q: Why did you get into databases?
A: At Harvard, CS was not cool then, it was for boring people. But Mae had a special seminar on databases. Joe's sister in Berkeley became a Unix sysadmin at IBM. Laura Hass asked Joe would you like to be a pre-doc? And that got things rolling.

Q: How does one go from academia to being a CEO?
A: Running a company is people and project management. There is a big learning curve. Once a startup reaches to 20 people, a phase-transition point is hit. People stop trusting each other; you need process in place. People management skills is the hardest thing you learn. Culture of the company follows the leadership, so you need to be a great role model.

Q: How did you decide what project to take on?
A: You can't get famous by being a smart guy, they come a dozen a dime. You need long term goals, and following them. I had only three themes that I keep banging on.

Q: What is your view of industrial landscape?
A: I have concerns about big companies reducing jobs. Cloud is a big opportunity! IBM research culture was beautiful, except for MS it is not there in the industrial landscape. I hope we can see more of that.

Q: For the DEI (diversity, equity, inclusion) initiative, what else can we be doing?
A: I will be humble in answering the question, I don't think my experience is too much. I learned a lot though. Institutional buy-in is absolutely critical.


I hadn't interacted with Joe in person before. Based on our email interactions, he came across as a nice/considerate person. Seeing him in person, and seeing him acknowledge others for his work, he comes across a genuinely nice and humble, yet very creative person. This rubs on to his students as well based on my observation.

Comments

Post a Comment

Popular posts from this blog

Hints for Distributed Systems Design

-
This is with apologies to Butler Lampson, who published the " Hints for computer system design " paper 40 years ago in SOSP'83. I don't claim to match that work of course. I just thought I could draft this post to organize my thinking about designing distributed systems and get feedback from others. I start with the same  disclaimer Lampson gave. These hints are not novel, not foolproof recipes, not laws of design, not precisely formulated, and not always appropriate. They are just hints.  They are context dependent, and some of them may be controversial. That being said, I have seen these hints successfully applied in distributed systems design throughout my 25 years in the field, starting from the theory of distributed systems (98-01), immersing into the practice of wireless sensor networks (01-11), and working on cloud computing systems both in the academia and industry ever since. These heuristic principles have been applied knowingly or unknowingly and has proven
Read more >>

Learning about distributed systems: where to start?

-
This is definitely not a "learn distributed systems in 21 days" post. I recommend a principled, from the foundations-up, studying of distributed systems, which will take a good three months in the first pass, and many more months to build competence after that. If you are practical and coding oriented you may not like my advice much. You may object saying, "Shouldn't I learn distributed systems with coding and hands on? Why can I not get started by deploying a Hadoop cluster, or studying the Raft code." I think that is the wrong way to go about learning distributed systems, because seeing similar code and programming language constructs will make you think this is familiar territory, and will give you a false sense of security. But, nothing can be further from the truth. Distributed systems need radically different software than centralized systems do.  --A. Tannenbaum This quotation is literally the first sentence in my distributed systems syllabus. Inst
Read more >>

Making database systems usable

-
Image
C. J. Date's Sigmod 1983 keynote, "Database Usability", was prescient. Usability is the most important thing to the customers. They care less about impressive benchmarks or clever algorithms, and more about whether they can operate and use a database efficiently to query, update, analyze, and persist their data with minimal headache. (BTW, does anyone have a link to the contents of this Sigmod'83 talk? There is no transcript around, except for this short abstract .) The paper we cover today is from Sigmod 2007. It takes on the database usability problem raised in that 1983 keynote head-on, and calls out that the king is still naked.  Let's give some context for the year 2007. Yes, XML format was still popular then. The use-case in the paper is XQuery. The paper does not contain any  reference to json. MongoDB would be released in 2009 with the document model; and that seems to be great timing for some of the usability pains mentioned in the paper! Web 2.0 was in
Read more >>

Looming Liability Machines (LLMs)

-
As part of our zoom reading group ( wow, 4.5 years old now ), we discussed a paper that uses LLMs for automatic root cause analysis (RCA) for cloud incidents. This was a pretty straightforward application of LLMs. The proposed system employs an LLM to match incoming incidents to incident handlers based on their alert types, predicts the incident's root cause category, and provides an explanatory narrative. The only customization is through prompt-engineering. Since this is a custom domain, I think a more principled and custom-designed  machine learning system would be more appropriate rather than adopting LLMs. Anyways, the use of LLMs for RCAs spooked me vicerally. I couldn't find the exact words during the paper discussion, but I can articulate this better now. Let me explain. RCA is serious business Root cause analysis (RCA) is the process of identifying the underlying causes of a problem/incident, rather than just addressing its symptoms. One RCA heuristic is asking 5 Why&#
Read more >>

Foundational distributed systems papers

-
I talked about the importance of reading foundational papers last week. To followup, here is my compilation of foundational papers in the distributed systems area. (I focused on the core distributed systems area, and did not cover networking, security, distributed ledgers, verification work etc. I even left out distributed transactions, I hope to cover them at a later date.)  I classified the papers by subject, and listed them in chronological order. I also listed expository papers and blog posts at the end of each section. Time and State in Distributed Systems Time, Clocks, and the Ordering of Events in a Distributed System. Leslie Lamport, Commn. of the ACM,  1978. Distributed Snapshots: Determining Global States of a Distributed System. K. Mani Chandy Leslie Lamport, ACM Transactions on Computer Systems, 1985. Virtual Time and Global States of Distributed Systems.  Mattern, F. 1988. Practical uses of synchronized clocks in distributed systems. B. Liskov, 1991. Expository papers
Read more >>

Advice to the young

-
Image
I notice I haven't written any advice posts recently. Here is a collection of my advice posts pre 2020. I've been feeling all this elderly wisdom pent up in me, ready to pour at any moment. So here it goes. Get ready to quench your thirst from my fount of wisdom. No man, think for yourself, only get what works for you. It is called foundations, not theory Foundations of computer science (or rather any field of study) are the most important topics you can learn. These lay down the frame of thinking/perspective for that area of study. Yet, I am saddened to hear these called as "theory", and labeled as "unpractical". This couldn't be farther from the truth. Take a look at how I recommend studying distributed systems . Don't you dare call this "theory" and "unpractical". This lays the bedrock that you build your practice on. Don't skimp on the foundations. Don't build your home on quicksand. Keep your hands dirty, your mind cl
Read more >>

Linearizability: A Correctness Condition for Concurrent Objects

-
Image
This paper is from Herlihy and Wing appeared in ACM Transactions on Programming Languages and Systems 1990. This is the canonical reference for the linearizability definition. I had not read this paper in detail before, so I thought it would be good to go to the source to see if there are additional delightful surprises in the original text. Hence, this post. I will dive into a technical analysis of the paper first, and then discuss some of my takes toward the end. I had written an accessible explanation of linearizability earlier; you may want to read that first. I will assume an understanding of linearizability to keep this review at reasonable length. Introduction I love how the old papers just barge in with the model, without bothered by pleasantries such as motivation of the problem. These are the first two sentences of the introduction. "A concurrent system consists of a collection of sequential processes that communicate through shared typed objects . This model encompass
Read more >>

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

-
Image
This paper is from Pat Helland, the apostate philosopher of database systems, overall a superb person, and a good friend of mine. The paper appeared this week at CIDR'24. (Check out the program for other interesting papers). The motivating question behind this work is: " What are the asymptotic limits to scale for cloud OLTP (OnLine Transaction Processing) systems? " Pat says that the CIDR 2023 paper "Is Scalable OLTP in the Cloud a Solved Problem?" prompted this question.  The answer to the question? Pat says that the answer lies in the joint responsibility of database and the application. If you know of Pat's work, which I have summarized several in this blog , you would know that Pat has been advocating along these lines before. But this paper provides a very crisp, specific, concrete answer. Read on for my summary of the paper. Disclaimer: This is a wisdom and technical information/detail packed 13-page paper, so I will try my best to summarize the sa
Read more >>

Understanding the Performance Implications of Storage-Disaggregated Databases

-
Image
Storage-compute disaggregation in databases has emerged as a pivotal architecture in cloud environments, as evidenced by Amazon ( Aurora ), Microsoft ( Socrates ), Google (AlloyDB), Alibaba ( PolarDB ), and Huawei (Taurus). This approach decouples compute from storage, allowing for independent and elastic scaling of compute and storage resources. It provides fault-tolerance at the storage level. You can then share the storage for other services, such as adding read-only replicas for the databases. You can even use the storage level for easier sharding of your database. Finally, you can also use this for exporting a changelog asynchronously to feed into peripheral cloud services, such as analytics. Disaggregated architecture was the topic of Sigmod 23 panel . I think this quote summarizes the industry's thinking on the topic. "Disaggregated architecture is here, and is not going anywhere. In a disaggregated architecture, storage is fungible, and computing scales independently.
Read more >>

Designing Data Intensive Applications (DDIA) Book

-
Image
We started reading this book as part of Alex Petrov's book club . We just got started, so you can join us, by joining the discord channel above. We meet Wednesday's 11am Eastern Time.  Previously we had read transaction processing book by Grey and Reuters. This page links to my summaries of that book. Chp 1. Reliable, Scalable, and Maintainable Applications I love the diagrams opening each chapter. Beautiful! The first chapter consists of warm up stuff. It talks about the definitions of reliabilty, scalability, and maintainability. It is still engaging, because  it is written with an educator and technical blogger voice, rather than a dry academic voice. This book came out on 2017. Martin is working on the new version. So if you have comments for things to focus on for the new version, it would be helpful to collect them in a document and email it to Martin. For example, I am curious about how the below paragraph from the Preface will get revised with 8 more years of hindsight:
Read more >>