Hi, I wouldn't try to use MySQL for a 10TB single instance DB unless most of the data is historical data that isn't accessed frequently. Sharding generally won't work for such an application, because every table will likely need partitioned by different keys (ie you have a very large order table and invoice table and it doesn't make sense to shard the invoice table by order number when the most frequent access is by invoice number). MySQL is not good for batch processing of large amounts of data (ie ,year over year reports) because it lacks intra-query parallelism. MySQL could be a fit if the last say 90 days of data fit into the buffer pool. In such a case you could use a slave for OLAP reporting. You could partition the tables using the partitioning option and report on them using Shard-Query (http://shardquery.com) to get intra-query parallelism as it can access partitions of tables in parallel, and it knows how to efficiently join tables partitioned on different keys on a single node. --Justin On Wed, Mar 9, 2016 at 5:06 AM, Seung Hoon Yoo <sehyoo96@gmail.com> wrote:
hi I am some question, i want some opinion below question
We planning ERP solution convert oracle to mariaDB so erp application pattern and BIG DB SIZING IS ISSUE.
Our IT background & goal 1) Apppliciation Complexity & Pattern: ERP Level complexity (high , OLTP & many batch) 2) DB SIZE : 10 TB Over (Single DB , NO sharding) 3) replication Gap issue
** My Question ** 1) Any Global reference is in MariaDB (on ERP level complexity and application pattern)
2) Any some reference in BIG DATA SIZING ? (EX: 5TB DB Exists in XX COMPARY e-commerce soluntion..)
IS Only solution is DB Sharding? no big data global reference ??
3) Replication performance : Our test result 2MB/SEC binlog written in master , so slave node get replication lag glorwing. (in row binlog format) ( 1master - 1 slave ) Any Body have experience some reference max transaction in no replication gap ? ( ex : 20MB / SEC but no replication gap our application experience..)
: parellel slave thread is good performance in single domain-id and ordered method replicaiton ?
( how much better performance in parallel replication?
(ex single repliation 2MB/SEC BINLOG => Replication GAP GROW but, parellel replicaiton is good in 10MB/Sec binlog write Environment)
thanks a lot.
regards, seung hoon yoo.
_______________________________________________ Mailing list: https://launchpad.net/~maria-discuss Post to : maria-discuss@lists.launchpad.net Unsubscribe : https://launchpad.net/~maria-discuss More help : https://help.launchpad.net/ListHelp