Daniel Abadi

New post where I discuss some of Chujun Song's benchmarking experiments (done in my lab at UMD) showing the overhead of large JDBC reads in Trino: starburst.io/blog/benchmark…

I wrote 2 blog posts recently on that discuss using Trino to access data stored in SQL databases and related performance issues. How to Avoid Data Access Bottlenecks When Using Trino lnkd.in/emj_i6SA The dangers of the JDBC bottleneck in Trino lnkd.in/exd4gNY3

Congrats to newly minted PhDs from my lab: Cuong (Ethan) Nguyen and Chujun Song (2nd and 4th from the right). Can't wait to see the incredible things you both will accomplish in your next steps ...

@andy_pavlo @pateljm @CMUDB @sriniseshan Big pick up for CMU. And now a total transformation from where CMU was a decade ago after Natassa and Chris left.

Blow out news! @pateljm is joining me at @CMUDB this fall semester. We will hold a live press conference in August once Jignesh gets settled. CMU's DB squad is now primed to rival the Germans at TUM and the Dutch at CWI! Props to @sriniseshan for turning this one out!

@ScyllaDB Thank you to @daniel_abadi and @kostja_osipov for a great distributed consensus discussion. We've come a long way since #CAP. Do you know about #PACELC?

I'll be speaking at the #ScyllaDB summit (scylladb.com/summit) on the throughput vs. consistency tradeoff in distributed systems on Feb 15.

@__pragma__ @muratdemirbas @PatHelland @danrkports Agreed. Serializability scalability got a bad reputation from systems that did it wrong. Well-designed systems can scale serializability to enormous scales, even for geographically distributed transactions.

@PatHelland @muratdemirbas @danrkports For distributed databases, sure. For single–node in-memory systems, well, plenty have cracked 1M TPS with full serializability. Snapshot isolation is great if all your writes are simple RMW ops, dangerous otherwise (requires devs to understand isolation semantics in detail).

The grammar police at Starburst (I won't name names) tried to nix my phrase: "It is no longer acceptable to “code, growed, and offload” a dataset" from the post I tweeted earlier today, but I got it in via a special poetry exception.

New post from me on Starburst's blog: starburst.io/blog/can-the-d…

@kartick_vad Depends on the mysql storage engine. The only difference between repeatable read and serializable in theory is phantom protection, and does not cause 10X slowdown in traditional DBMSs. Both systems that bolt on serializability after-the-fact do run into performance problems.

@daniel_abadi When I increased the consistency level of MySQL from the default repeatable read to serializable, the performance dropped 10x. Is that a problem with our code, a fundamental theoretical constraint, or a MySQL limitation?

@johnhugg @andy_pavlo Thanks John. My 4 year old promised me that she's going to teach him how to smile, but she hasn't succeeded yet. So I think we'll work on that before full sentences :)

@daniel_abadi @andy_pavlo In about 3 months he’s just gonna start talking in full sentences. BTW Congrats Dan!

Thank you @andy_pavlo for the database-themed newborn clothing for our new addition.

Very proud of my student @gangliao101 who recently graduated from my lab at @umdcs and just accepted a full time Senior Researcher position. To find out where --- you will have to follow him 😀

Maryland's primary is tomorrow. Of the three main candidates running on the democratic side, was surprised to see that only @peterfranchot seems to discuss importance of investment in *all* public universities on his Website. Maybe I'm looking in the wrong place for the others?

Agree with much of this advice. Especially that you don't need to network with famous people. Interactions with peers are more likely to lead to long term relationships and many will go on to become famous over time. Further, it's a way for your work to get a grass roots impact.

On my way to SIGMOD --- my first in-person conference since COVID. Can't wait to catch up will all the people in the database community that I haven't seen in way too long ...

@daniel_abadi and, speaking on PACELC with us tomorrow, I might add. Looking forward to it lp.scylladb.com/wbn-scylla-clo…

Found out this week that there is a wikipedia page about me that somehow I didn't know existed until now. Thank you to whoever contributed to it. en.wikipedia.org/wiki/Daniel_Ab…

In light of @VoltronData's 100M series A funding last week, it might be worth revisiting my 2017 and 2018 posts on Apache Arrow: dbmsmusings.blogspot.com/2018/03/an-ana… and dbmsmusings.blogspot.com/2017/10/apache…

Nice post on isolation levels and consistency from Ben Darnell at @CockroachDB: cockroachlabs.com/blog/db-consis…. The basic approach of thinking about isolation and consistency separately was also discussed in my 2019 post: dbmsmusings.blogspot.com/2019/08/an-exp….