The right read optimization is actually write optimizationjames tong
?
Fractal Tree indexes provide a way to optimize both reads and writes for large, growing datasets. They achieve this by combining aspects of log-structured merge trees (LSM trees) and B-trees - buffering data during writes like LSM trees to batch inserts, but maintaining a B-tree structure for efficient queries. This allows fractal tree indexes to have very fast insertion performance like LSM trees while also supporting fast queries like B-tree indexes. However, fractal tree indexes do introduce more complexity in the tree structure that can make concurrency more difficult.
The right read optimization is actually write optimizationjames tong
?
Fractal Tree indexes provide a way to optimize both reads and writes for large, growing datasets. They achieve this by combining aspects of log-structured merge trees (LSM trees) and B-trees - buffering data during writes like LSM trees to batch inserts, but maintaining a B-tree structure for efficient queries. This allows fractal tree indexes to have very fast insertion performance like LSM trees while also supporting fast queries like B-tree indexes. However, fractal tree indexes do introduce more complexity in the tree structure that can make concurrency more difficult.
This document summarizes Joshua Hoffman's talk on scalable system operations at Tumblr. The talk outlines Tumblr's management stack for automating server provisioning including iPXE, Invisible Touch, Collins, Phil, Kickstart, and Puppet. It describes how the tools are used together in workflows for server intake, provisioning, and addressing challenges like configuring networking and storage during installation. The talk emphasizes principles like modularity, simplicity, and avoiding breaking the operating system.
Benchmarks, performance, scalability, and capacity what s behind the numbers...james tong
?
Baron Schwartz gave a presentation on analyzing database performance beyond surface-level metrics and benchmarks. He discussed how ideal benchmarks provide full system specifications and metrics over time to understand response times and throughput. Little's Law and queueing theory can predict concurrency, response times, and capacity given arrival rates and service times. While tools like Erlang C model queues, the assumptions must be validated. True scalability is nonlinear due to bottlenecks, and debunking performance claims requires examining raw data.
The document discusses various stability antipatterns that can cause systems to fail. It describes issues that can arise from integration points, database calls that hang, unexpected failures from external systems, traffic surges overwhelming a system, attacks from users, unbalanced capacities across systems, unbounded result sets, and more. It provides examples of each antipattern and emphasizes the importance of monitoring dependencies, using timeouts, testing with realistic data volumes, and implementing circuit breakers and other proven patterns to prevent failures from cascading across systems and spreading.
6. Enqueue 模式兼容性 请求 / 占用 Null SS SX S SSX X Null Yes Yes Yes Yes Yes Yes SS Yes Yes Yes Yes Yes No SX Yes Yes Yes No No No S Yes Yes No Yes No No SSX Yes Yes No No No No X Yes No No No No No
24. TX 锁 -itl contention 当 block 上的活动事务槽用完 , 又没有足够的空间来扩展事务槽或者是达到 maxtrans 设定的事务槽数量时 . 此 block 上新请求的事务会等到 mode=4 的 tx lock session 1 Update james_itl set value = ‘test 1’ where id =1; Session 2 Update james_itl set value = ‘test 2’ where id = 2; Session 3 Update james_itl set value = ‘test 3’ where id = 3;
#16: 我自己测试一般都是 Alter session set events ‘10704 trace name context forever , level 10’;
#21: 内容摘自 www.dbthink.com/?p=452 相关过程我自己做了测试 . 下面是持有锁的时间顺序 Lock table in SubShare mode Lock table in Share mode (very short time if not been blocked by others) Lock table in subShare mode Processing index rebuild Lock table in Share mode (very short time if not been blocked by others) Lock journal table in Exclusive mode (drop the journal table)