greg/stasis-aries-wal
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

one full pass

Browse source
This commit is contained in:
Eric Brewer 2005-03-31 15:28:27 +00:00
parent dbef511fbc
commit 09e018f72b
1 changed files with 62 additions and 60 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

120
doc/position-paper/LLADD.txt
View file

@ -1,24 +1,24 @@
Russell Sears Russell Sears
Eric Brewer Eric Brewer
UC Berkeley
Automated Verification and Optimization of Extensions to Transactional A Flexible, Extensible Transaction Framework
Storage Systems.
Existing transactional systems are designed to handle specific Existing transactional systems are designed to handle specific
workloads well. Unfortunately, these systems' implementations are workloads well. Unfortunately, these systems' implementations are
geared toward specific workloads and data layouts such as those mononolithic and hide the transactional infrastructure underneath a
traditionally associated with SQL. Lower level implementations such SQL interface. Lower-level implementations such as Berkeley DB handle
as Berkeley DB handle a wider variety of workloads and are built in a a wider variety of workloads and are built in a more modular fashion.
modular fashion. However, they do not provide APIs to allow However, they do not provide APIs to allow applications to build upon
applications to build upon and modify low level policies such as and modify low-level policies such as allocation strategies, page
allocation strategies, page layout or details of recovery semantics. layout or details of recovery semantics. Furthermore, data structure
Furthermore, data structure implementations are typically not broken implementations are typically not broken into separable, public APIs,
into separable, public APIs, encouraging a "from scratch" approach to which discourages the implementation of new transactional data
the implementation of new transactional data structures. structures.
Contrast this to the handling of data structures within modern object Contrast this to the handling of data structures within modern
oriented programming languages such as C++ or Java. Such languages object-oriented programming languages such as C++ or Java. Such
typically provide a large number of data storage algorithm languages typically provide a large number of data storage algorithm
implementations. These structures may be used interchangeably with implementations. These structures may be used interchangeably with
application-specific data collections, and collection implementations application-specific data collections, and collection implementations
may be composed into more sophisticated data structures. may be composed into more sophisticated data structures.
@ -37,7 +37,7 @@ developers to implement sophisticated cross-layer optimizations easily.
Overview of the LLADD Architecture Overview of the LLADD Architecture
---------------------------------- ----------------------------------
General purpose transactional storage systems are extremely complex General-purpose transactional storage systems are extremely complex
and only handle certain types of workloads efficiently. However, new and only handle certain types of workloads efficiently. However, new
types of applications and workloads are introduced on a regular basis. types of applications and workloads are introduced on a regular basis.
This results in the implementation of specialized, ad-hoc data storage This results in the implementation of specialized, ad-hoc data storage
@ -45,25 +45,27 @@ systems from scratch, wasting resources and preventing code reuse.
Instead of developing a set of general purpose data structures that Instead of developing a set of general purpose data structures that
attempt to behave well across many workloads, we have implemented a attempt to behave well across many workloads, we have implemented a
lower level API that makes it easy for application designers to lower-level API that makes it easy for application designers to
implement specialized data structures. Essentially, we have implement specialized data structures. Essentially, we have
implemented an extensible navigational database system. We implemented an extensible navigational database system. We
believe that this system will support modern development practices and believe that this system will support modern development practices and
address rapidly evolving applications before allows transactions to be used in a wider range of applications.
appropriate general-purpose solutions have been developed.
In cases *** This paragraph doesn't make sense to me:
where the development of a general-purpose solution is not economical,
our approach should lead to maintainable and efficient long-term
solutions. Semi-structured data stores provide good examples of both
types of scenarios. General XML storage technologies are improving
rapidly, but still fail to handle many types of applications.
In cases where the development of a general-purpose solution is not
economical, our approach should lead to maintainable and efficient
long-term solutions. Semi-structured data stores provide good
examples of both types of scenarios. General XML storage technologies
are improving rapidly, but still fail to handle many types of
applications.
*** this is risky: there are many people working on XML databases
For instance, For instance,
we know of no general purpose solution that seriously addresses we know of no general-purpose solution that seriously addresses
semi-structured scientific information, such as the large repositories semi-structured scientific information, such as the large repositories
typical of bioinformatics research efforts[PDB, NCBI, Gene Ontology]. typical of bioinformatics research efforts[PDB, NCBI, Gene Ontology].
While many scientific projects are moving toward XML for their data Although many scientific projects are moving toward XML for their data
representation, we have found that XML is used primarily as a data representation, we have found that XML is used primarily as a data
interchange format, and that existing XML tools fail to address the interchange format, and that existing XML tools fail to address the
needs of automated data mining, scientific computing and interactive needs of automated data mining, scientific computing and interactive
@ -89,7 +91,7 @@ These structures were developed with reusability in mind, encouraging
developers to compose existing operations into application-specific data developers to compose existing operations into application-specific data
structures. For example, the hashtable is structures. For example, the hashtable is
implemented on top of reusable modules that implement a resizable array implemented on top of reusable modules that implement a resizable array
and two exchangeable linked list variants. and two exchangeable linked-list variants.
In other work, we show that the system is competitive with In other work, we show that the system is competitive with
Berkeley DB on traditional (hashtable based) workloads, and have shown Berkeley DB on traditional (hashtable based) workloads, and have shown
@ -113,25 +115,24 @@ page cache in order to efficiently service write requests. We also
leveraged our customizable log format to log differences to objects leveraged our customizable log format to log differences to objects
instead of entire copies of objects. instead of entire copies of objects.
With these optimizations, we showed a 2-3x performance improvement over Berkeley DB on object With these optimizations, we showed a 2-3x performance improvement
persistence across our benchmarks, and a 3-4x improvement over an over Berkeley DB on object persistence across our benchmarks, and a
in-process version of MySQL with the InnoDB backend. (A traditional 3-4x improvement over an in-process version of MySQL with the InnoDB
MySQL setup that made use of a separate server process was prohibitively backend. (A traditional MySQL setup that made use of a separate
slow. InnoDB provided the best performance among MySQL's durable storage managers.) server process was prohibitively slow. InnoDB provided the best
Furthermore, our system uses memory more efficiently, performance among MySQL's durable storage managers.) Furthermore, our
increasing its performance advantage in situations where the size of system uses memory more efficiently, increasing its performance
system memory is a bottleneck. advantage in situations where the size of system memory is a
bottleneck.
We We leave systematic performance tuning of LLADD to future work, and
leave systematic performance tuning of LLADD to future work, and
believe that further optimizations will improve our performance on believe that further optimizations will improve our performance on
these benchmarks significantly. these benchmarks significantly. In general, LLADD's customizability
enables many optimizations that are difficult for other systems.
LLADD's customizability provides superior performance over existing, Because of its natural integration into standard
complicated systems. Because of its natural integration into standard
system software development practices, we think that LLADD can be system software development practices, we think that LLADD can be
naturally extended into networked and distributed domains. naturally extended into networked and distributed domains.
For example, typical write-ahead-logging protocols implicitly For example, typical write-ahead-logging protocols implicitly
implement machine independent, reorderable log entries in order to implement machine independent, reorderable log entries in order to
implement logical undo. These two properties have been crucial in implement logical undo. These two properties have been crucial in
@ -144,7 +145,7 @@ Current Research Focus
---------------------- ----------------------
LLADD's design assumes that application developers will LLADD's design assumes that application developers will
implement high performance transactional data structures. This is a implement high-performance transactional data structures. This is a
big assumption, as these data structures are notoriously difficult to big assumption, as these data structures are notoriously difficult to
implement correctly. Our current research attempts to address these implement correctly. Our current research attempts to address these
concerns. concerns.
@ -161,7 +162,7 @@ Also, while data structure algorithms tend to be simple and easily
understood, performance tuning and verification of implementation understood, performance tuning and verification of implementation
correctness is extremely difficult. correctness is extremely difficult.
Recovery based algorithms must behave correctly during forward Recovery-based algorithms must behave correctly during forward
operation and also under arbitrary recovery scenarios. The latter operation and also under arbitrary recovery scenarios. The latter
requirement is particularly difficult to verify due to the large requirement is particularly difficult to verify due to the large
number of materialized page file states that could occur after a number of materialized page file states that could occur after a
@ -169,7 +170,7 @@ crash.
Fortunately, write-ahead-logging schemes such as ARIES make use of Fortunately, write-ahead-logging schemes such as ARIES make use of
nested-top-actions to vastly simplify the problem. Given the nested-top-actions to vastly simplify the problem. Given the
correctness of page based physical undo and redo, logical undo may correctness of page-based physical undo and redo, logical undo may
assume that page spanning operations are applied to the data store assume that page spanning operations are applied to the data store
atomically. atomically.
@ -182,11 +183,12 @@ behavior during recovery is equivalent to the behavior that would
result if an abort() was issued on each prefix of the log that is result if an abort() was issued on each prefix of the log that is
generated during normal forward operation. generated during normal forward operation.
By using coarse (one latch per logical operation) latching, we can *** below implies that two operations have two latches and can thus run in parallel ***
By using coarse latching (one latch per logical operation), we can
drastically reduce the size of this space, allowing conventional drastically reduce the size of this space, allowing conventional
state-state based search techniques (such as randomized or exhaustive state-state based search techniques (such as randomized or exhaustive
state-space searches, or unit testing techniques) to be state-space searches, or unit testing techniques) to be
practical. It has been shown that such coarse grained latching can practical. It has been shown that such coarse-grained latching can
yield high-performance concurrent data structures if yield high-performance concurrent data structures if
semantics-preserving optimizations such as page prefetching are semantics-preserving optimizations such as page prefetching are
applied [ARIES/IM]. applied [ARIES/IM].
@ -205,25 +207,25 @@ latching/unlatching behavior, but this would greatly complicate the
API that application developers must work with, and complicate any API that application developers must work with, and complicate any
application code that made use of such optimizations. application code that made use of such optimizations.
*** code hoisting might be a better example
Compiler optimization techniques such as partial common subexpression Compiler optimization techniques such as partial common subexpression
elimination solve an analogous problem to remove redundant algebraic elimination solve an analogous problem to remove redundant algebraic
computations. We hope to extend such techniques to reduce the number computations. We hope to extend such techniques to reduce the number
of buffer manager and locking calls made by existing code at runtime. of buffer manager and locking calls made by existing code at runtime.
Anecdotal evidence and personal experience suggest Anecdotal evidence and personal experience suggest that similar
that similar optimization techniques are applicable to optimization techniques are applicable to application code. Because
application code. Because local LLADD calls are simply normal local LLADD calls are simply normal function calls, it may even be
function calls, it may even be possible to apply the transformations that these optimizations possible to apply the transformations that these optimizations perform
perform to application code that is unaware of the underlying storage implementation. to application code that is unaware of the underlying storage
This class of implementation. This class of optimizations would be very difficult
optimizations would be very difficult to implement with existing to implement with existing transactional storage systems but should
transactional storage systems but should significantly improve application performance. significantly improve application performance.
Our implementation of LLADD is still unstable and inappropriate for *** no reason to say this: Our implementation of LLADD is still unstable and inappropriate for use on important data.
use on important data. We hope to validate our ideas about static analysis We hope to validate our ideas about static analysis by incorporating
by incorporating them into the development process as we increase them into the development process as we increase the reliability and
the reliability and overall quality of LLADD's implementation and its overall quality of LLADD's implementation and its APIs.
APIs.
Our architecture provides a set of tools that allow applications to implement Our architecture provides a set of tools that allow applications to implement
custom transactional data structures and page layouts. This avoids custom transactional data structures and page layouts. This avoids