Source: Goetz Graefe CMU 15-721 (Spring 2018)
copy from: http://15721.courses.cs.cmu.edu/spring2018/notes/02-inmemory.pdf
1. Disk-Oriented Database Management Systems
For a disk oriented DBMS, the system architecture is predicated on the assumption that data is stored in non-volatile memory. The database is organized as a set of fixed-length blocks called slotted pages. The system uses an in-memory (volatile) buffer pool to cache the blocks cached from disk.
Buffer Pool:
—index stores page id and slots，search it in the page table ,which knows the data actually in.
– When a query accesses a page, the DBMS checks to see if that page is already in memory.
– If not, the DBMS retrieves the memory from disk and copies it into a frame in its buffer pool.
– Once the page is in memory, the DBMS translates any on-disk addresses to their in-memory addresses.
– Every tuple access has to go through the buffer pool manager regardless of whether that data will always be in memory
Concurrency Control:
– In a disk oriented DBMS, the system assumes that a transaction could stall at any time when it tries to access data that is not in memory.
– The system’s concurrency control protocol allows the DBMS to execute other transactions at the same time to improve performance while still preserving atomicity and isolation guarantees.
LOCKS VS. LATCHES
Locks
→ Protects the database's logical contents from other txns.
→ Held for txn duration.→ Need to be able to rollback changes.
Latches
→ Protects the critical sections of the DBMS's internal data structures from other threads.
→ Held for operation duration.
→ Do not need to be able to rollback changes.
|
|
Locks |
Latches |
|
Separate... |
User transactions |
Threads |
|
Protect... |
Database Contents |
In-Memory Data Structures |
|
During... |
Entire Transactions |
Critical Sections |
|
Modes... |
Shared, Exclusive, Update, Intention |
Read, Write |
|
Deadlock |
Detection & Resolution |
Avoidance |
|
...by... |
Waits-for, Timeout, Aborts |
Coding Discipline |
|
Kept in... |
Lock Manager |
Protected Data Structure |
Logging and Recovery:
– Most DBMS use STEAL + NO-FORCE buffer pool policies so all modifications have to be flushed to the WAL before a transaction can commit [3].
STEAL: the system is allowed to evict dirty pages from the buffer pool out the desk for transactions that haven’t committed yet
NO-FORCE: database is required to flush all the dirty pages of a transaction when it commits.
– Log entries contain before and after image of record modified.
What make database slow?
meet a modified record before and after image, you need to before so that you can use undo rollback any changes and you need to ask image so that you can redo it, and then it’s up to database to decide whether your transaction actually commit or not.
keep track of LSNs(log sequence number) all throughout the DBMS.
In a disk-based system, only approximately 7% of instructions are done on actual work.
34% for buffer pool,14% for latching,16% for locking,12% for logging,26% for b-tree
2.In-Memory Database Management Systems
The system architecture assumes that the primary storage location of the database is in memory. This means that the DBMS does not need to perform extra steps during execution to handle the case where it has to retrieve data from disk. If disk I/O is no longer the slowest resource, much of the DBMS architecture will have to change to account for other bottlenecks:
– Locking/latching
– Cache-line misses
– Pointer chasing
– Predicate evaluation
– Data movement and copying
– Networking (between application and DBMS)
Data Organization:
– In-memory DBMS will organize tuples in blocks/pages.
– Direct memory pointers vs record ids.
– Fixed-length vs variable-length data pools.
– Use checksums to detect software errors from trashing the database.
Pg toast: pg stores data in the same row at single page, for a really large value like a text field then split up into a separate page, called toast
Why not mmap
Memory-map (mmap) a database file into DRAM and let the OS be in charge of swapping data in and out as needed.
Use madvise and msync to give hints to the OS about what data is safe to flush.
Notable mmap DBMSs:→ MongoDB (pre WiredTiger)
→ MonetDB: column store → LMDB
Using mmap gives up fine-grained control on the contents of memory.→ Cannot perform non-blocking memory access.→ The "on-disk" representation has to be the same as the “in-memory" representation.→ The DBMS has no way of knowing what pages are in memory or not.→ Various mmap-related syscalls are not portable.
Mmap if find a record not in memory, will have a page_fault, and load the data into memory.we cannot do everything, give up complete control to the authoring system.
WiredTiger is a multi-thread storage engine, if not in memory, we can ask other thread to fetch it, and return the location. The thread work for all the transactions for other queries.
Concurrency Control:
– Fine-grained locking allows for better concurrency but requires more locks.
– Coarse-grained locking requires fewer locks but limits the amount of concurrency.– The DBMS can store locking information about each tuple together with its data.
–The new bottle neck is contention caused from transactions trying to access data at the same time.
Indexes:
– Indexes are usually rebuilt in an in-memory DBMS after restart to avoid logging overhead. – Modern databases typically do not log index updates.
Query Processing:
– The best strategy for executing a query plan in a DBMS changes when all the data is already in memory. Sequential scans are no longer significantly faster than random access. – The traditional tuple-at-a-time iterator model is too slow because of function calls.
Logging and Recovery:
– The DBMS still needs WAL on non-volatile storage since the system could halt at anytime.
– May be possible to use more lightweight logging schemes (e.g., only store redo information).
– Since there are no “dirty pages”, we do not need to maintain LSNs throughout the systems.– System also still takes checkpoints, however, there are different methods for checkpointing.
Non-Volatile Memory: Emerging hardware that is similar to speed of DRAM with the persistence guarantees of an SSD [1].
Notable Early In-memory DBMSs:
– TimesTen: Originally Smallbase [5] from HP Labs. Multi-process, shared memory DBMS. Bought by Oracle in 2005 [7].
– Dali: Multi-process shared memory storage manager using memory mapped files [6].
– P*TIME: Korean in-memory DBMS from the 2000s [2]. Lots of interesting features (e.g., hybrid storage layouts, support for larger-than-memory databases). Sold to SAP in 2005 and is now part of HANA.
References
[1] J. Arulraj, A. Pavlo, and S. Dulloor. Let’s Talk About Storage & Recovery Methods for Non-Volatile Memory Database Systems. In Proceedings of the 2015 International Conference on Management of Data, SIGMOD ’15, pages 707–722, 2015.
[2] S. K. Cha and C. Song. P*time: Highly scalable oltp dbms for managing update-intensive stream workload. In Proceedings of the Thirtieth International Conference on Very Large Data Bases - Vol- ume 30, VLDB ’04, pages 1033–1044, 2004. URL http://dl.acm.org/citation.cfm?id=1316689. 1316778.
[3] M. J. Franklin. Concurrency control and recovery. In Computing Handbook, Third Edition: Information Systems and Information Technology, pages 12: 1–21. 2014. URL http://db.lcs.mit.edu/6.893/ F04/ccandr.pdf.
[4] S. Harizopoulos, D. J. Abadi, S. Madden, and M. Stonebraker. OLTP through the looking glass, and what we found there. In SIGMOD ’08: Proceedings of the 2008 ACM SIGMOD International Confer- ence on Management of Data, pages 981–992, 2008. doi: http://doi.acm.org/10.1145/1376616.1376713.
[5] M. Heytens, S. Listgarten, M.-A. Neimat, and K. Wilkinson. Smallbase: A main-memory dbms for high-performance applications. Technical report, Hewlett-Packard Laboratories, 1995.
[6] H. Jagadish, D. Lieuwen, R. Rastogi, and A. Silberschatz. Dali: A high performance main memory storage manager. VLDB, pages 48–59. URL http://www.vldb.org/conf/1994/P048.PDF.
[7] T. Lahiri, M.-A. Neimat, and S. Folkman. Oracle timesten: An in-memory database for enterprise applications. IEEE Data Eng. Bull., 36(2):6–13, 2013. URL http://sites.computer.org/debull/ A13june/TimesTen1.pdf.
[8] M. Stonebraker, S. Madden, D. J. Abadi, S. Harizopoulos, N. Hachem, and P. Helland. The end of an architectural era: (it’s time for a complete rewrite). In VLDB ’07: Proceedings of the 33rd international conference on Very large data bases, pages 1150–1160, 2007. URL http://hstore.cs.brown.edu/ papers/hstore-endofera.pdf.