[LU-17821] LMR1c: Improve DNE Distributed Transaction Performance

• Type: Technical task
• Resolution: Unresolved
• Priority: Minor
• Assignee: WC Triage
• Reporter: Andreas Dilger

Description

DNE2 Distributed Transactions have excessive ordering/sync operations:

• Need to understand bottlenecks and eliminate them in code, protocol, formats
• Single OSP RPC in flight between each MDT pair (implementation issue, patch in Gerrit)
• Compromise from original design due to implementation issues
• Difficult to journal bit-level updates in recovery llog file
• Must order updates within a single byte to ensure other bits are not clobbered
• Batched OSP/OUT updates in a single RPC?

Improvements to DNE distributed transactions benefit up all DNE operations (especially striped directory operations, renames, etc. that may touch many MDTs), independent of LMR, so are worthwhile to implement independent of the LMR project.

Issue Links

• is related to: LU-11854 Dir operation on DNE2 are slower than DNE1 or non DNE
• is related to: LU-15528 downgrade remote PW/EX lock taken in distributed transactions to COS mode
• is related to: LU-6864 DNE3: Support multiple modify RPCs in flight for MDT-MDT connection
• is related to: LU-15527 remove dependency between transactions started from the same MDT
• is related to: LU-15529 optimize directory migration parent locking
• is related to: LU-7426 DNE3: Current llog format for remote update llog