by Yu Li, Chair of the HBaseConAsia2018 Conference Committee and member of the HBase PMC and Michael Stack, HBase PMC-er.
A crowd of us gathered at the Gehua New Century Hotel, in North-Eastern Beijing to attend a lively HBaseConAsia2018. It was an all-day community-run event held on August 17th. Admission was free thanks to our generous host, the Alibaba Group. This was the second HBaseConAsia conference. The first was hosted by Huawei in Shenzhen, in 2017.
The majority of the talks were in Chinese  but slides were in English. Below are links that point into slideshare where you’ll find slides and video. There were three tracks -- internals, ops, and ecosystem -- of heavy-duty hbasing.
Your authors led off the conference with a keynote on the “State of Apache HBase”. It covered general state (“stable”), the effort getting hbase-2.0.0 out the door, and toward the end, a couple of features that would sit pretty in any hbase-3.0.0.
The second (dense) keynote was by Chunhui Shen (Long-time HBase PMC member) and Long Cao. They talked of “Recent Developments” around HBase at Alibaba and at Alibaba Cloud (“Aliyun”). Alibaba has been using HBase since 0.20.6! and keep up their own version internally (“AliHB”) where they dev features that later get pushed upstream to Apache. The node and cluster counts were ahem, impressive, as is the ecosystem built up over time to serve different workload types. Near-future projects are figuring separation of compute from storage and hbase deploy on pangu, the Alibaba filesystem. This talk also served as overview for some of the talks that came later in the day where topics mentioned in the morning were given a deep dive in later sessions.
One such deep-dive was the first talk in track one, on “Using CCSMap to improve HBase YGC time” by Chance Li and Lijin Bin. CCSMap is a stripped-down ConcurrentSkipListMap for use in place of the native java implementation HBase currently uses. Look for it in an upcoming Apache HBase release.
Ramkrishna Vasudevan and Anoop Sam John, HBase PMCers from Intel described their recent development work, WALLess HBase with persistent memory devices.
A bunch of talks were focused on hbase deploys at various companies. The Xiaomi folks, heavyweight contributors to Apache HBase with lots of HBase deployed at scale, gave two talks, one by Xie Gang on “HDFS optimizations for HBase at Xiaomi“, and “HBase at Xiaomi” by HBase Committer, Guanghao Zhang. JingYi Yao talked on “HBase at Didi” (GeoMesa, JanusGraph, and Phoenix). Other interesting deploy-types were described in “HBase at China Telecom”, “@ China Life Insurance”, “@ Meituan”, and “@ Lianjia”.
The boys from PInterest, Chenji Pan and Lianghong Xu, did “Improving HBase reliability at Pinterest with geo-‐replication and efficient backup” and our WenLong (Allan) Yang described an interesting scheme for treating hot and cold data differently in “Separating hot-cold data into heterogeneous storage based on layered compaction”.
Alibaba described how they do security, phoenix, spark, in their cloud HBase offering, AsparaDB, and an interesting backup solution they use internally that requires “zero-modification” to HBase. “The Application of HBase in New Energy Vehicle Monitoring System” was by Yan Yu, Chetankumar Jyestaram Khatri talked on “Scaling 30 TB's of Data lake with Apache HBase and Scala DSL at Production”, Biju Nair of Bloomberg described “Serving billions of queries in millisecond latency”, and Pankaj Kumar, Wei Zhi, and Chaoqiang Zhong of last year’s hosts Huawei presented on “HBase and OpenTSDB practice at Huawei” (Huawei also offer HBase in the cloud).
At the end of all sessions, a worn-out PMC took questions from a worn-out audience.
On the day after, HBase contributors came together for a fruitful dev meeting. Rough Notes were posted to the dev@hbase mailing list.
Let us finish this short report with a picture of all the HBaseConAsia2018 speakers taken at the end of the day after all was done.