GTK , Harman/Becker Automotive Systems GmbH, HP, Hyperion, Inktomi, Innodata Isogen, Informatica, Intel, Interlogics, IONA, IXOS, Jikes, jQuery, Library of Congress, Mathworks, Mozilla, Netezza, OpenOffice, Oracle (Solaris, Java), Lawson Software, Leica Geosystems GIS & Mapping.

The IBM one really annoys me. Their banking mainframe systems were already annoying, 15-20 years ago, before the Indian enshittification. Then they bought Netezza, which I loved using, and shit all over that, too

Indeed! Notably Oracle Exadata/Exascale (and Netezza) push various query processing operators like projection, (join) filtering, column decompression down from the DB layer to their MPP "smart storage" layer containing the disks, to push down some compute and reduce net traffic.

ちょっと前のNetezza論争を思い出すな。あの頃もNetezzaすごいと言われていたが、結局、汎用性の前に消えてしまったな。潰しが効かないんだよな。

NVIDIA の GPU に汎用性があっても AI 以外じゃゲームとレンダリング位じゃない？あと汎用型ゆえに価格でTPU勝つのは容易じゃなさそう。 昔の Exadata(GPU) vs Netezza(TPU) vs Greenplum(Groq)を見てるようだね。

This trend in disaggregated databases mirrors biological evolution, progressing from simple cells to specialized organs working in concert. Consider Oracle's Exadata: storage and compute were separated because I/O was the primary bottleneck in its MPP-imitating, shared-disk RAC systems. Over time, Exadata I/O cells became highly specialized, handling compression, indexing, filtering, encryption, and more to enhance efficiency. Snowflake later adopted this disaggregation, implementing a heterogeneously elastic separation of storage and compute. Similarly, in the data lake ecosystem, we first saw specialized file formats like Parquet, designed for efficient columnar storage, compression, and fast, filtered reads. Next, a separate table format layer evolved on top, featuring tools like Iceberg, providing the crucial ACID transactions, time travel, and schema management that Parquet alone lacks. We also see a pattern of engines being repurposed for specific storage or workload tiers, rather than existing merely as standalone components. For example, Netezza was used as an "analytics accelerator" for DB2 to handle complex queries. In a "downward" tiering example, queries were pushed down to Hadoop through connectors for cheaper and broader data access. Today, different engines work in layers, with services like Dremio and Trino acting as a specialized federation layer to connect disparate data sources. Clearly, this disaggregation trend isn't simplifying architecture. Instead, it's creating highly specialized, distinct parts that, when combined, build more complex, resilient, and powerful systems.

#SWTTokyo25 一時テーブルでのクエリ分割、NetezzaでもRedshiftでも有効だったし、MPPなOLAP DBでは常套手段の印象

放置状態だったESXiサーバからProxmoxへ移行中。サポート切れOSの墓場だったけど、改めて見ると仮想マシンにUnityVSAとかSoftware Netezzaとかあってめっちゃ謎。

Netezza Performance server ?

This is a classic IBM move: buy a successful business, alienate its loyal customer base, and then utterly destroy what’s left from within. See Lotus, Red Hat, Informix, Netezza, Phytel, Explorys Truven, Cognos, SPSS, Unica, etc

SQL was there before you were born and SQL will be there after you die: Somebody invents a "SQL replacement" every decade. It then fails, gets ignored, or worse, its good bits get unceremoniously absorbed into the SQL standard, leaving the original "revolutionary" system to wither on the vine, much like Sisyphus is doomed to push his boulder. Remember the NoSQL craze of the late 2000s? "SQL is dead! Documents are the future!" Fast forward a bit, and what do we see? MongoDB, the poster child of that movement, now has a SQL interface. Cassandra has CQL. The supposed usurpers have, in many ways, bent the knee to the established king. The funny thing is, while these "next big things" often flame out, the old stuff? It sticks around. There are IBM IMS databases, relics from the 1960s Apollo moon missions, still chugging away in major banks, processing your ATM transactions. Why? Because ripping out a core system that (mostly) works is like performing open-heart surgery with a rusty spork – risky and expensive. And it’s not just the systems; the ideas from these ancient behemoths are still surprisingly relevant. Think query compilation – turning your SQL into efficient machine code. IBM was doing that in assembly in the 1970s with System R. We use LLVM now, but the core challenge? Still the same. This is why a stroll through database history isn't just an academic exercise; it’s a critical lesson in not reinventing the flat tire. If we don't learn from the ghosts of databases past, we're just setting ourselves up to repeat their costly mistakes. So, let's take a quick, unfiltered dive into how we got here. The Pre-Relational Dark Ages: Pointers, Hierarchies, and Programmer Pain (1960s) Imagine it’s the 1960s. You’re NASA, literally trying to shoot for the moon. You have an astronomical number of parts, suppliers, and intricate dependencies. Your data is a tangled mess. Enter systems like IDS (Integrated Data Store). GE originally cooked this up for a timber company with a massive inventory problem. Think of your data as a giant spiderweb, where every piece of information was connected to others by physical pointers on disk or in memory. If one of those pointers got corrupted? Your entire database was toast. Programmers, armed with languages like COBOL, had to write complex, nested loops to navigate this maze, one agonizing record (or "tuple") at a time. In a classic corporate blunder, GE, deciding they weren't #1 in the computer biz, sold off their entire computing division to Honeywell. So much for that moonshot. Then there was IMS (Information Management System) from IBM, the powerhouse behind the actual Apollo program's parts tracking. IMS went for a hierarchical model – think of a rigid, top-down organizational chart for your data. A part could only be supplied by one vendor in its defined hierarchy. If "Battery Model Z" was available from three different suppliers, you had three complete, redundant copies of "Battery Model Z's" details. Need to update the battery's specs? Good luck hunting down every single instance. Even worse, how the data was physically stored – as a hash table or a B-tree – was hardcoded into the application. If you realized a B-tree would be better for range queries after initially choosing a hash table, you weren't just changing a setting; you were dumping data, reloading it, and rewriting your application code because the API itself changed. Yet, this beast, or its descendants, still processes your banking transactions today. The lesson? Inertia is a powerful force. Around the same time, Charles Bachman, who'd worked on IDS, pushed for a standard way for COBOL programmers to talk to databases. This led to CODASYL (Conference on Data Systems Languages), which championed the "network data model." It was like IDS on steroids – more pointers, more sets, more ways for your data to become an unmanageable spaghetti monster. Bachman even won a Turing Award for this. The Enlightenment: Ted Codd & The Relational Revolution (1970s) Working at IBM, a mathematician named Ted Codd saw the IMS programmers tearing their hair out. He realized the insanity of coupling the logical view of data with its physical storage and the inefficiency of tuple-at-a-time processing. His 1970 paper proposed a radical new approach: Simple Data Structures: Store data in simple tables (relations). High-Level Language: Let users declare what data they want, not how to get it, step-by-step. This crucial insight sparked the development of powerful new query languages. At UC Berkeley, the Ingres project, led by Michael Stonebraker, developed QUEL (QUEry Language) around 1974, directly based on Codd's relational algebra. Concurrently, at IBM, researchers on the System R project were developing SQL (Structured Query Language), with its initial designs also appearing in the mid-1970s. Physical Independence: Separate the logical data view from the physical storage. Let the database figure out the best way to store and retrieve. When Codd published his paper in 1970, the old guard scoffed. 'A machine write queries better than a human? Preposterous!' Yet, both QUEL and SQL aimed to do just that, providing high-level ways to interact with relational data. While Ingres with QUEL gained early traction in academic and research circles, and Stonebraker still insists QUEL was superior, it was SQL that eventually won the broader industry adoption. This was partly due to IBM's significant market influence when they later commercialized SQL with DB2 in the early 1980s, effectively making it the de facto standard. The path to SQL's dominance also involved the famous story of Larry Ellison at Oracle closely following and implementing System R's ideas. The standards bodies, too, ultimately leaned towards SQL, a decision Stonebraker famously attributes to his own disdain for such committees, leading him not to push QUEL as aggressively in those forums. Meanwhile, a sharp character named Larry Ellison saw an opportunity. Legend has it he’d get IBM’s research papers (sometimes by just calling up researchers who were happy to share their "academic" work) and implement the ideas in his own fledgling system: Oracle. When IBM finally launched DB2 in the early 1980s, it was a signal: the relational model was here to stay. And because IBM chose SQL (originally SEQUEL – Structured English QUEry Language – later shortened to SQL due to a trademark dispute) as its language, SQL became the de facto standard. Oracle, having already bet on SQL, was perfectly positioned to ride the wave. Ingres, with QUEL, eventually added SQL support, but by then, the race was largely decided. The "SQL is Dead" Cycles: A Recurring Theme The 1980s solidified relational dominance, but the "SQL is flawed/dead" narrative was just getting started. Object-Oriented Databases (Late 1980s/Early 1990s): C was hot. Programmers grumbled about the "impedance mismatch" – relational tables didn't map cleanly to their beloved objects. "Why can't we just store objects directly?" they asked. Thus, OODBs like ObjectStore and Versant were born. Their Achilles' heel? No standard query language (OQL arrived too late and never caught on), and applications were tightly coupled to specific OODB APIs. The good ideas (like richer data types) were eventually, you guessed it, absorbed into the relational model, leading to "object-relational" systems like PostgreSQL (which, amusingly, was initially developed in LISP by Stonebraker's team post-Ingres). The "Boring" 90s & The Internet Explosion (2000s): The 90s were a period of refinement. Microsoft forked Sybase to create SQL Server. A Finn named Michael "Monty" Widenius created MySQL (My was his daughter's name; he later created MariaDB for his other daughter, Maria, and MaxDB for his son Max). PostgreSQL, having shed its LISP and QUEL origins, embraced SQL. Then the internet hit, and suddenly, even small outfits could generate (and drown in) massive datasets. The Rise of Analytical Databases & MapReduce: People weren't just transacting; they wanted to analyze their growing data piles. Row-oriented databases choked on analytical queries. Early attempts like "data cubes" (pre-computed aggregations) were a stopgap. Then came specialized analytical databases, many of them forks of PostgreSQL, pioneering columnar storage (Netezza, Vertica, Greenplum). This was a game-changer for analytics. Simultaneously, Google, needing to process its colossal web crawl, invented MapReduce. Yahoo! quickly cloned it as Hadoop. Instead of SQL, you wrote custom Java functions for map and reduce stages. Programmers were back to defining data parsing logic from scratch, a huge step backward in terms of abstraction. While initially hyped, the inefficiencies of MapReduce for general-purpose analytics became apparent, and SQL layers (like Hive) were awkwardly bolted on top. NoSQL (The Big One - 2000s): The internet scale also revived the "SQL is too slow/rigid" argument with unprecedented force. The NoSQL movement (MongoDB, Cassandra, DynamoDB) championed "schema-less" designs, horizontal scalability, and often, "eventual consistency" over strict ACID transactions. "We need to be always on, even if the data is a bit weird sometimes!" was the mantra. The irony, as we've seen, is that most of these systems have since added SQL-like interfaces and stronger consistency options. The Modern Whirlwind: Clouds, Lakes, and More "Revolutions" (2010s - Present) The cycles continue, now accelerated by the cloud: NewSQL & Distributed SQL: The goal was NoSQL's scalability but with SQL and ACID compliance. Early NewSQL systems had mixed success. Their spiritual successors, often branded "Distributed SQL" (CockroachDB, TiDB, YugabyteDB), are learning from past mistakes and gaining more serious traction. The Cloud Transformation & Data Lakes: Cloud platforms (AWS, Azure, GCP) fundamentally changed database architecture. The dominant model shifted from shared-nothing to shared-disk (think Snowflake), where compute and storage are separated, often using object stores like S3. This gave rise to Data Lakes, where raw data in open formats (Parquet, ORC) is dumped into cheap cloud storage, accessible by various query engines. Graph Databases: "Your data is inherently connected, so store it as a native graph!" (Neo4j, TigerGraph). It’s the OODB argument in a new hat. Unsurprisingly, SQL is now incorporating graph query capabilities (SQL/PGQ). Recent research even shows general-purpose analytical databases like DuckDB, with graph-specific optimizations (like advanced join algorithms), can outperform specialized graph systems. My bet? Graph databases will be a valuable niche, and the relational model will swallow the best ideas. Time Series Databases: A specialized flavor of relational, optimized for the firehose of data from IoT devices, application metrics, and financial tickers (TimescaleDB, InfluxDB). They handle time-windowed queries and high-throughput append-only workloads very well. Vector Databases (The Current Darling): Fueled by the AI/ML boom, these are designed to store and search high-dimensional vector embeddings (the "thoughts" of AI models). Most are essentially document databases with an Approximate Nearest Neighbor (ANN) index grafted on. Guess what? Mainstream relational databases are already adding vector data types and ANN indexing capabilities. This will likely become just another feature in the ever-expanding relational toolkit. Blockchain Databases (The Emperor's New Clothes): "Decentralized! Trustless! Immutable Ledger!" The pitch is seductive. The reality, for most database use cases, is a performance nightmare and a solution desperately searching for a problem that a traditional database with proper auditing and access control can't solve more efficiently and cheaply. Amazon even built QLDB, which offers the immutable ledger part without the decentralized blockchain overhead, implicitly acknowledging where the real (limited) value lies. Still Spinning: The Enduring Power of Relational So, after decades of supposed revolutions, where are we? The relational model, with SQL as its lingua franca, isn't just surviving; it's thriving. It's a cockroach in the best sense – adaptable, resilient, and capable of absorbing almost anything thrown at it. The core principles Ted Codd laid out – data independence, simple structures, a high-level way to ask for what you want – are timeless. The next time a charismatic founder on a conference stage tells you their shiny new database will make SQL obsolete, remember this journey. History doesn't just repeat itself in the database world; it often comes around with a new marketing budget and a fresh set of buzzwords, only to find the old ideas were pretty good after all.

أرخم ايرور في العالم بجد ويجيلك بعد ٩٠٠ ثانية يارب عافيني من Netezza بجد

長く一線で開発を行ってきた経験豊富な人材となります。 即日可能、男性、62歳、日本籍、弊社プロパー 【経　験】39年 【スキル】Java、javascript、 React、 Next.js、PostgreSQL、VB.Net、Python、AWS、Aurora、OpenSearch、PostgreSQL、Netezza、PanaSearch、XMLBeans、JBoss、Oracleなど 【工　程】要件定義～総合テスト 【稼動日】即日可能 【最寄駅】JR　湘南新宿ライン添い 【面接日】事前調整可 【並　行】並行営業中 【希　望】都内出勤可 【備　考】長く一線で開発を行ってきた経験豊富な人材となります。 要件定義、基本設計から、詳細設計、製造、テストまで一貫して対応できるエンジニアです。 物腰もやわらかく、なんでも柔軟に対応できますので、是非一度面談お願いできればと思います。 #SES #SES営業

弊社プロパーです 【スキル】Java、javascript、 React、 Next.js、PostgreSQL、VB.Net、Python、AWS、Aurora、OpenSearch、PostgreSQL、Netezza、PanaSearch、XMLBeans、JBoss、Oracleなど 【工　程】要件定義～総合テスト 【稼動日】即日可能 #SES #SES営業

データセンターはあまり入ったことないんだよね。一度は自社DCのでかいAIXサーバーでの性能検証、もう一度はNetezza検証時にデータ持ち込み。最近はクラウドが多いというのもあるけど。

【弊社プロパーの要員のご紹介】です。 男性、62歳、日本籍、弊社プロパー 【経　験】39年 【スキル】Java、javascript、 React、 Next.js、PostgreSQL、VB.Net、Python、AWS、Aurora、OpenSearch、PostgreSQL、Netezza、PanaSearch、XMLBeans、JBoss、Oracleなど 【工　程】要件定義～総合テスト 【稼動日】4月～ 【最寄駅】JR　湘南新宿ライン添い 【面接日】事前調整可 【並　行】並行営業中 【希　望】都内出勤可 【備　考】長く一線で開発を行ってきた経験豊富な人材となります。 要件定義、基本設計から、詳細設計、製造、テストまで一貫して対応できるエンジニアです。 物腰もやわらかく、なんでも柔軟に対応できますので、是非一度面談お願いできればと思います。 #SES #SES営業

【弊社プロパーの要員のご紹介】です。 日本籍、弊社プロパー、男性、62歳 【経　験】39年 【スキル】Java、javascript、 React、 Next.js、PostgreSQL、VB.Net、Python、AWS、Aurora、OpenSearch、PostgreSQL、Netezza、PanaSearch、XMLBeans、JBoss、Oracleなど #SES #SES営業

弊社プロパーの要員のご紹介です！ 男性、62歳、日本籍、 弊社プロパー 【経　験】39年 【スキル】Java、javascript、 React、 Next.js、PostgreSQL、VB.Net、Python、AWS、Aurora、OpenSearch、PostgreSQL、Netezza、PanaSearch、XMLBeans、JBoss、Oracleなど #SES #SES営業

弊社プロパーの要員のご紹介です。 男性、62歳、日本籍、弊社プロパー 【経　験】39年 【スキル】Java、javascript、 React、 Next.js、PostgreSQL、VB.Net、Python、AWS、Aurora、OpenSearch、PostgreSQL、Netezza、PanaSearch、XMLBeans、JBoss、Oracleなど 【工　程】要件定義～総合テスト 【稼動日】4月～ 【最寄駅】JR　湘南新宿ライン添い 【面接日】事前調整可 【並　行】並行営業中 【希　望】都内出勤可 【備　考】長く一線で開発を行ってきた経験豊富な人材となります。 要件定義、基本設計から、詳細設計、製造、テストまで一貫して対応できるエンジニアです。 物腰もやわらかく、なんでも柔軟に対応できますので、是非一度面談お願いできればと思います。 #SES #SES営業