[ASTERIXDB-2365] Various LSM IO Operations Shouldn't Simply Share the Same Write Queue

• Type: Improvement
• Status: Open
• Priority: Major
• Resolution: Unresolved
• Affects Version/s: 0.9.4
• Component/s: STO - Storage
• Assignee: Unassigned
• Reporter: Chen Luo
• Created: 13/Apr/18 02:39
• Updated: 10/Apr/19 18:31

Description

Currently the BufferCache has a single writer queue for all write operations (mainly for LSM component bulkloading). However, without a smart scheduling algorithm, this causes unnecessary stalls for short duration operations (e.g., flush) if some heavy operation (e.g., large merge) is ongoing. It is necessary to develop some sort of I/O scheduling algorithm to ensure short duration operations are not blocked by those longer ones.