Nicolas Liochon

zk, big data, realtime systems and machine learning. Apache HBase committer & PMC.

[Jay Kreps @jaykreps]

Great article on stream processing by @takidau, one of people behind Google's stream processing engine Millwheel http://t.co/JBK1XBoJNA

[Martin Thompson @mjpt777]

It can now take longer in the Java NIO code than time on the physical network. We need to think differently about core libraries.

[Martin Thompson @mjpt777]

Aeron 0.1 is available on Maven Central!
Adds < 1μs on top of raw network latency and can do 20+ million 40 byte messages per second.

[Qarnot @Qarnot]

Straight from NYC, Time Warner praises Qarnot after Cloud Innovation World Cup award, by @AdamBalkin >>
http://t.co/XX9JkJfcki

[Martin Kleppmann @martinkl]

My @PapersWeLoveLdn talk on Attita&Welch paper:
Slides: https://t.co/dACo5pFAQm
Video: https://t.co/aM9lIMQx04
Paper: http://t.co/nBD6WTXZB8

[Tino Tereshko 🇺🇦 @thetinot]

#Google #BigQuery - $80,000 per month. #AWS #Redshift - $300k+ per month. How? Cloud-native consumption pricing. http://t.co/JDkghpBiIE

[Nicolas Liochon @nkeywal]

Congrats! And good luck. https://t.co/F8VkOPVnZg

[Jay Kreps @jaykreps]

Excited to announce that @ConfluentInc has raised a $24m series B round led by Index Ventures http://t.co/HgRtzKkuLx

[Nicolas Liochon @nkeywal]

@johnhugg @eric_brewer We differ here :-). I like my example on the machine learning calculations. I would not choose CA for a database.

[Nicolas Liochon @nkeywal]

@johnhugg @eric_brewer It's "less than that of other systemic failures": sometimes the most probable failure scenario is not a partition.

[Nicolas Liochon @nkeywal]

@johnhugg @eric_brewer Oh it was our bugs :-). Or the error is not clear so the admin decides to restart the whole system (15 mins downtime)

[Nicolas Liochon @nkeywal]

@johnhugg @eric_brewer With more nodes, true partition tolerance (i.e. minimal impact on latency) is important but hard & very expensive.

[Nicolas Liochon @nkeywal]

@johnhugg @eric_brewer for functionally complex systems w/ less than 10 nodes I had much more issues with software bugs and human errors.

[Nicolas Liochon @nkeywal]

@jambay I totally agree. (even if I'm not sure that the vast majority of cases deployed in production are partition tolerant :-).

[John Evdemon @jevdemon]

Being partition tolerant is comfortable. You have to be Stonebraker to claim partition intolerance. Worth reading. http://t.co/q22ik1j3F0

[Nicolas Liochon @nkeywal]

My new post on the CA category in the CAP theorem. With some extra details from @eric_brewer himself! http://t.co/WGPzLd9f0c

[William Vambenepe @vambenepe]

Thank you @cpswan for giving me a chance to explain the value of Big Data as a service in this @infoq interview: http://t.co/2NAYN7FDPc

[Nicolas Liochon @nkeywal]

A must read on @Hadapt strategy on /Presto and Open Source... http://t.co/YCI12cyKxT https://t.co/C7k0qXgdXi

[Daniel Abadi @daniel_abadi]

New post from me on why we love #presto at @Hadapt / @Teradata http://t.co/YF4xO9fP1s

[Jesse Yates @jesse_yates]

Over a year since the originally coded, HDFS-6440 has been committed! 3+ NNs now possible on Hadoop 3.0.0+

[Nicolas Liochon @nkeywal]

@herberts 2/3 thinking, 1/3 stealth :-)

[Nicolas Liochon @nkeywal]

@herberts thanks