hybrid stm/htm for nested transactions on openjdk

Hybrid STM/HTM for Nested Transactions on OpenJDK

Keith ChapmanPurdue U

Tony HoskingANU/Data61, Purdue U

Eliot MossUMass

MotivationSTM has been around for ages

• But STM is slow

Commodity hardware for transactions available now

• But HTM approaches are only best effort

Out goal: Accelerate STM with HTM when possible

Nested TransactionsAllow composition of transactions

Two flavors

• Closed Nested Transactions

• HTM inside STM not useful; must do all the same work

• Open Nested Transactions

• HTM inside STM avoids most TM work – HTM well-suited!

XJ: Transactional JavaXJ Language

• Supports flat and closed/open/boosted nested transactions

The implementation supports hybrid transactions

• HTM support via Intel TSX

• HTM and STM can co-exist

Uses an Optimistic-reads / Pessimistic-writes protocol

TM Metadata

Txn Log

0 0Version # 0 0 0Txn ID 1

One metadata word for every object

STM Transaction Protocol (Read)

T1 Log

0 00 0


T1 Log

0 00 0

Check Mode


T1 Log

0 0

0 0

T1 Read


T1 Log

0 0

0 0

T1 Validate


T1 Log

0 0

T1 Commit

0 0

STM Transaction Protocol (Write)

T1 Log

0 0


T1 Log

Check Mode

T1 1

0 0


T1 Log

T1 Write

T1 11 0


T1 Log

T1 Commit

STM Conflict Detection

T1 Log

0 00 0

T2 Log

0 0


T1 Log

0 0

0 0

T2 Log

0 0

T1 Read


T1 Log

0 0

0 0

T2 Log

0 0

T2 Write

T2 1


T1 Log

0 0

0 0

T2 Log

0 0

T1 Validate

T2 1


T1 Log

0 0

T2 Log

0 0

T1 Abort

T2 1

Hybrid Transaction ProtocolSTM – HTM conflicts detected by lock word accesses

• Explicit XABORT if locked by another transaction

• HTM reads – Read the metadata word

• STM writes modify the metadata word

• Causes HTM to abort

• HTM writes – Increment version number

• Causes STM read invalidation / HTM abort

Abstract Locking &

Undo Operations

Abstract Locks for STM

Open / Boosted Atomic Method Body

Acquire abstract locks

Release abstract locks

If top level transaction



Log abstract locks

If nested transaction

Log undo operations

Abstract Locks for Hybrid TM


Validate abstract locks



Log abstract locks

If top level is HTM

Log undo operations

Why Validation Works• HTM vs STM

• If abstract locks conflict they must touch some same physical words in the abstract locking data structure — otherwise they could not detect the conflict



Why Validation Works• HTM vs STM

• If abstract locks conflict they must touch some same physical words in the abstract locking data structure — otherwise they could not detect the conflict

• HTM vs HTM

• No conflict in the locking data structure because all accesses to it are reads

• Any real conflicts that exist will occur on the actual data structure



STM and HTM Methods

STM needs logging HTM doesn't

Different actions during read/write

Different actions for abstract locks

HTM should fall back to STM

STM and HTM Methods

STM needs logging HTM doesn't

Different actions during read/write

Different actions for abstract locks

HTM should fall back to STM

Maintain separate HTM and STM versions of methods

STM Method Variants

Original(Non-txnal)

STM Method Variants

Original(Non-txnal)

Can callA B

STM Method Variants

Original(Non-txnal)

Top-level Txn STM

Can callA B

STM Method Variants

STM Transactionalized

Original(Non-txnal)

Top-level Txn STM

Can callA B

STM Method Variants

STM Transactionalized

Original(Non-txnal)

New nested Txn STM

Top-level Txn STM

Can callA B

Hybrid TM Method VariantsOriginal

(Non-txnal)

Hybrid TM Method VariantsOriginal

(Non-txnal)Can callA B

Hybrid TM Method Variants

Router

Original(Non-txnal)Can call

A B


Top-level Txn STM

Router


A B


Top-level Txn HTM

Top-level Txn STM

Router


A B


HTM Transactionalized

Top-level Txn HTM

Top-level Txn STM

Router


A B


Optimized New nested Txn

HTMHTM

Transactionalized

Top-level Txn HTM

Top-level Txn STM

Router


A B



HTMHTM

Transactionalized

Top-level Txn HTM

STMTransactionalized

Top-level Txn STM

Router


A B



HTMHTM

Transactionalized

Top-level Txn HTM


Top-level Txn STM

New nestedTxn STM

Router


A BParent is closed

A B



HTMHTM

Transactionalized

Top-level Txn HTM


Top-level Txn STM

New nestedTxn STM

Router Nested Router


A BParent is closed

A B

Parent is openA B



HTMHTM

Transactionalized

Top-level Txn HTM

New nestedTxn HTM


Top-level Txn STM

New nestedTxn STM

Router Nested Router


A BParent is closed

A B

Parent is openA B

XJ System Architecture


XJ source code


XJ source code XJ Compilerstandard Java

bytecode

compile


XJ source code XJ Compilerstandard Java

bytecode XJ Rewriterbytecode

+ run-time calls

compile load


XJ source code

XJ run-time library

XJ Compilerstandard Java


+ run-time callsHTM-enabled

JVM

compile load run


HTM 4-5 times faster than STM

XJ source code

XJ run-time library

XJ Compilerstandard Java


+ run-time callsHTM-enabled

JVM

compile load run

VM ModificationsKept to a minimum

Modifications done on OpenJDK:

• Native methods to begin, end, and abort a HW transaction

• Made them intrinsic to the HotSpot C1/C2 compilers

Had to go through several hoops to get HTM to work with HotSpot’s optimizing compilers

Results

SynchrobenchMicro-benchmarks to evaluate synchronization performance on various data structures

Added ability to run multiple operations within a single transaction (group size)

Included XJ versions of the benchmarks

• TransactionalFriendlyTreeSet

48-way, Intel Xeon E5-2690 v3 machine with 2 sockets of 12 hyperthreaded cores

Thro

ughp

ut (N

orm

aliz

ed)

0

0.2

0.4

0.6

0.8

1

Threads1 2 4 8 12 16 20 24 28 32 36 40 44 48

Thro

ughp

ut (N

orm

aliz

ed)

0

0.2

0.4

0.6

0.8

1

Threads1 2 4 8 12 16 20 24 28 32 36 40 44 48

Open nestedClosed nested

Group size 1

5% Updates

Thro

ughp

ut (N

orm

aliz

ed)

0

0.2

0.4

0.6

0.8

1

Threads1 2 4 8 12 16 20 24 28 32 36 40 44 48

Thro

ughp

ut (N

orm

aliz

ed)

0

0.2

0.4

0.6

0.8

1

Threads1 2 4 8 12 16 20 24 28 32 36 40 44 48

Group size 1Group size 2

Open nestedClosed nested 5% Updates

Thro

ughp

ut (N

orm

aliz

ed)

0

0.2

0.4

0.6

0.8

1

Threads1 2 4 8 12 16 20 24 28 32 36 40 44 48

Thro

ughp

ut (N

orm

aliz

ed)

0

0.2

0.4

0.6

0.8

1

Threads1 2 4 8 12 16 20 24 28 32 36 40 44 48

Group size 1Group size 2Group size 4


Thro

ughp

ut (N

orm

aliz

ed)

0

0.2

0.4

0.6

0.8

1

Threads1 2 4 8 12 16 20 24 28 32 36 40 44 48

Thro

ughp

ut (N

orm

aliz

ed)

0

0.2

0.4

0.6

0.8

1

Threads1 2 4 8 12 16 20 24 28 32 36 40 44 48


Group size 8


Thro

ughp

ut (N

orm

aliz

ed)

0

0.2

0.4

0.6

0.8

1

Threads1 2 4 8 12 16 20 24 28 32 36 40 44 48

Thro

ughp

ut (N

orm

aliz

ed)

0

0.2

0.4

0.6

0.8

1

Threads1 2 4 8 12 16 20 24 28 32 36 40 44 48


Group size 8Group size 16


Thro

ughp

ut (N

orm

aliz

ed)

0

0.2

0.4

0.6

0.8

1

Threads1 2 4 8 12 16 20 24 28 32 36 40 44 48

Thro

ughp

ut (N

orm

aliz

ed)

0

0.2

0.4

0.6

0.8

1

Threads1 2 4 8 12 16 20 24 28 32 36 40 44 48




0

50

100

150

200

1 2 4 8 12 16 20 24 28 32 36 40 44 48 1 2 4 8 12 16 20 24 28 32 36 40 44 48 1 2 4 8 12 16 20 24 28 32 36 40 44 48

Group size 1 Group size 2 Group size 4

com

mitt

ed o

ps a

nd a

borte

d tx

ns (1

06 )

threads

open htm commitsclosed htm commits

open stm commitsclosed stm commits

htm aborts

ConclusionsSTM and HTM can co-exist for nested transactions in Java

• Closed nesting — Similar to previous schemes

• Open nesting — Novel validation mechanism

• Implemented in OpenJDK on Intel TSX — Artifact evaluated





When it works, HTM is ~4-5× faster than STM






Open nesting increases the envelope of effectiveness for HTM






Open nesting increases the envelope of effectiveness for HTM

Production VM would need deeper modification

hybrid stm/htm for nested transactions on openjdk

Software