It's HighScalability Time:
- Quotable Quotes:
- @rbranson: So turns out the EBS Provisioned IOPS volumes are not total shit. This injects the man with a few weeks of runway.
- @markzohar: It is amazing to witness the efficiency & scalability of algorithms. Given the choice, build your tech company on algorithms not code.
- Your Bottleneck is Dead. Long Live Your Bottleneck. Mailinator's Paul Tyma brings up the interesting notion that once bottlenecks have moved to the CPU because programs can nolonger hide behind IO overhead - code efficiency becomes important again.
- Facebook Moving To The JVM: The main seismic event will be nothing less than the complete removal of interpretors from main stream general purpose programming.
- Improved HTTPS Performance with Early SSL Termination. Filepicker.io has found early SSL termination cuts the time from 384ms to 146ms, over 200ms improvement. (on Hacker News)
- Hot Interconnects, which talks all things high-performance computing and high-speed networking, has made available last years videos in anticipation of this year's conference starting on Aug 24. You might find Large Data Center Fabrics: Is There a Good Answer? interesting to watch.
- Boundary: not just for clusters. A taught tale of debugging a surprising problem: the aggregate behaviour of individually fast queries can bring it all down. Slow query log isn't enough.
- A Brief Summary of the NoSQL World. Brief and to the point.
- Analysing performance data. LogNormal with a good look at distributions and the stats you need to make sense of your data.
- Performance tuning syslog-ng. Etsy was crafty and was able not buy new logging servers by tweaking their syslog configuration: make sure syslog-ng is parsing the most frequently logged lines first, not last; optimize flow control variables; reduce power usage by scaling back.
- Is 100% uptime really possible? Baron Schwartz takes on dubious marketing claims saying "Downtime is not only always possible, but always inevitable." If the Earth is destroyed will your database really survive?
- Nagios Monitoring Performance. 37signals shows how they scaled Nagios by rewriting check scripts, reducing HTTP calls, using passive services, and installing tweaks. Tuning resulted in reduced machine load, reasoanable signal latency, and increased coverage.
- On Eventual Consistency from Basho: The use of eventual consistency in well-designed systems does not lead to inconsistency. Instead, such systems may allow brief (but shortly resolved) discrepancies at precisely the moments when the other alternative would be to simply fail. Also, Basho Technologies is hosting RICON 2012, a distributed systems conference for developers in San Francisco.
- Some good Geek Reading from Rob Diana.