Eliminating false sharing:
Abstract: "The performance and scalability of bus-based, shared memory multiprocessors is limited by the amount of bus traffic. Previous studies have shown that for machines with large caches and a write- invalidate coherency protocol, most of the bus traffic stems from coherency overhead. Some...
Gespeichert in:
| Hauptverfasser: | , |
|---|---|
| Format: | Buch |
| Sprache: | English |
| Veröffentlicht: |
Seattle, Wash.
1990
|
| Schriftenreihe: | University of Washington <Seattle, Wash.> / Department of Computer Science: Technical report
90,12,1 |
| Schlagworte: | |
| Zusammenfassung: | Abstract: "The performance and scalability of bus-based, shared memory multiprocessors is limited by the amount of bus traffic. Previous studies have shown that for machines with large caches and a write- invalidate coherency protocol, most of the bus traffic stems from coherency overhead. Some of this overhead is unavoidable, a direct consequence of the true sharing activity in the program. However, for caches with multi- word cache blocks, coherency overhead can also be caused by multiple processors accessing different words in the same block, known as false sharing. This paper measures the amount of false sharing in two types of parallel applications. For some applications the amount of false sharing increases substantially as block size increases For these programs there is a potential for improving performance by reducing the amount of false sharing. For other applications the amount of false sharing is so small as to have no significant impact on performance. For the programs that exhibit large amounts of false sharing we present two simple transformations for restructuring shared data, and apply them to the shared objects responsible for most of the false sharing. Simulations of the traces from the modified programs indicate that the transformations reduce the miss rates significantly. The false sharing miss rates decreased by 50% to 60%, the total miss rates by 20% to 40%, the coherency related bus traffic by 50% to 75%, and bus utilization by an average of more than 15% Shared misses were reduced by an order of magnitude more than in a previous study which applied a different type of transformation. |
| Beschreibung: | 21 S. |
Internformat
MARC
| LEADER | 00000nam a2200000 cb4500 | ||
|---|---|---|---|
| 001 | BV008992802 | ||
| 003 | DE-604 | ||
| 005 | 00000000000000.0 | ||
| 007 | t | ||
| 008 | 940206s1990 |||| 00||| eng d | ||
| 035 | |a (OCoLC)26797094 | ||
| 035 | |a (DE-599)BVBBV008992802 | ||
| 040 | |a DE-604 |b ger |e rakddb | ||
| 041 | 0 | |a eng | |
| 049 | |a DE-29T | ||
| 100 | 1 | |a Eggers, Susan J. |e Verfasser |4 aut | |
| 245 | 1 | 0 | |a Eliminating false sharing |c Susan J. Eggers and Tor E. Jeremiassen |
| 264 | 1 | |a Seattle, Wash. |c 1990 | |
| 300 | |a 21 S. | ||
| 336 | |b txt |2 rdacontent | ||
| 337 | |b n |2 rdamedia | ||
| 338 | |b nc |2 rdacarrier | ||
| 490 | 1 | |a University of Washington <Seattle, Wash.> / Department of Computer Science: Technical report |v 90,12,1 | |
| 520 | 3 | |a Abstract: "The performance and scalability of bus-based, shared memory multiprocessors is limited by the amount of bus traffic. Previous studies have shown that for machines with large caches and a write- invalidate coherency protocol, most of the bus traffic stems from coherency overhead. Some of this overhead is unavoidable, a direct consequence of the true sharing activity in the program. However, for caches with multi- word cache blocks, coherency overhead can also be caused by multiple processors accessing different words in the same block, known as false sharing. This paper measures the amount of false sharing in two types of parallel applications. For some applications the amount of false sharing increases substantially as block size increases | |
| 520 | 3 | |a For these programs there is a potential for improving performance by reducing the amount of false sharing. For other applications the amount of false sharing is so small as to have no significant impact on performance. For the programs that exhibit large amounts of false sharing we present two simple transformations for restructuring shared data, and apply them to the shared objects responsible for most of the false sharing. Simulations of the traces from the modified programs indicate that the transformations reduce the miss rates significantly. The false sharing miss rates decreased by 50% to 60%, the total miss rates by 20% to 40%, the coherency related bus traffic by 50% to 75%, and bus utilization by an average of more than 15% | |
| 520 | 3 | |a Shared misses were reduced by an order of magnitude more than in a previous study which applied a different type of transformation. | |
| 650 | 4 | |a Parallel programming (Computer science) | |
| 700 | 1 | |a Jeremiassen, Tor E. |e Verfasser |4 aut | |
| 810 | 2 | |a Department of Computer Science: Technical report |t University of Washington <Seattle, Wash.> |v 90,12,1 |w (DE-604)BV008930431 |9 90,12,1 | |
| 999 | |a oai:aleph.bib-bvb.de:BVB01-005941720 | ||
Datensatz im Suchindex
| _version_ | 1804123335320141824 |
|---|---|
| any_adam_object | |
| author | Eggers, Susan J. Jeremiassen, Tor E. |
| author_facet | Eggers, Susan J. Jeremiassen, Tor E. |
| author_role | aut aut |
| author_sort | Eggers, Susan J. |
| author_variant | s j e sj sje t e j te tej |
| building | Verbundindex |
| bvnumber | BV008992802 |
| ctrlnum | (OCoLC)26797094 (DE-599)BVBBV008992802 |
| format | Book |
| fullrecord | <?xml version="1.0" encoding="UTF-8"?><collection xmlns="http://www.loc.gov/MARC21/slim"><record><leader>02706nam a2200325 cb4500</leader><controlfield tag="001">BV008992802</controlfield><controlfield tag="003">DE-604</controlfield><controlfield tag="005">00000000000000.0</controlfield><controlfield tag="007">t</controlfield><controlfield tag="008">940206s1990 |||| 00||| eng d</controlfield><datafield tag="035" ind1=" " ind2=" "><subfield code="a">(OCoLC)26797094</subfield></datafield><datafield tag="035" ind1=" " ind2=" "><subfield code="a">(DE-599)BVBBV008992802</subfield></datafield><datafield tag="040" ind1=" " ind2=" "><subfield code="a">DE-604</subfield><subfield code="b">ger</subfield><subfield code="e">rakddb</subfield></datafield><datafield tag="041" ind1="0" ind2=" "><subfield code="a">eng</subfield></datafield><datafield tag="049" ind1=" " ind2=" "><subfield code="a">DE-29T</subfield></datafield><datafield tag="100" ind1="1" ind2=" "><subfield code="a">Eggers, Susan J.</subfield><subfield code="e">Verfasser</subfield><subfield code="4">aut</subfield></datafield><datafield tag="245" ind1="1" ind2="0"><subfield code="a">Eliminating false sharing</subfield><subfield code="c">Susan J. Eggers and Tor E. Jeremiassen</subfield></datafield><datafield tag="264" ind1=" " ind2="1"><subfield code="a">Seattle, Wash.</subfield><subfield code="c">1990</subfield></datafield><datafield tag="300" ind1=" " ind2=" "><subfield code="a">21 S.</subfield></datafield><datafield tag="336" ind1=" " ind2=" "><subfield code="b">txt</subfield><subfield code="2">rdacontent</subfield></datafield><datafield tag="337" ind1=" " ind2=" "><subfield code="b">n</subfield><subfield code="2">rdamedia</subfield></datafield><datafield tag="338" ind1=" " ind2=" "><subfield code="b">nc</subfield><subfield code="2">rdacarrier</subfield></datafield><datafield tag="490" ind1="1" ind2=" "><subfield code="a">University of Washington <Seattle, Wash.> / Department of Computer Science: Technical report</subfield><subfield code="v">90,12,1</subfield></datafield><datafield tag="520" ind1="3" ind2=" "><subfield code="a">Abstract: "The performance and scalability of bus-based, shared memory multiprocessors is limited by the amount of bus traffic. Previous studies have shown that for machines with large caches and a write- invalidate coherency protocol, most of the bus traffic stems from coherency overhead. Some of this overhead is unavoidable, a direct consequence of the true sharing activity in the program. However, for caches with multi- word cache blocks, coherency overhead can also be caused by multiple processors accessing different words in the same block, known as false sharing. This paper measures the amount of false sharing in two types of parallel applications. For some applications the amount of false sharing increases substantially as block size increases</subfield></datafield><datafield tag="520" ind1="3" ind2=" "><subfield code="a">For these programs there is a potential for improving performance by reducing the amount of false sharing. For other applications the amount of false sharing is so small as to have no significant impact on performance. For the programs that exhibit large amounts of false sharing we present two simple transformations for restructuring shared data, and apply them to the shared objects responsible for most of the false sharing. Simulations of the traces from the modified programs indicate that the transformations reduce the miss rates significantly. The false sharing miss rates decreased by 50% to 60%, the total miss rates by 20% to 40%, the coherency related bus traffic by 50% to 75%, and bus utilization by an average of more than 15%</subfield></datafield><datafield tag="520" ind1="3" ind2=" "><subfield code="a">Shared misses were reduced by an order of magnitude more than in a previous study which applied a different type of transformation.</subfield></datafield><datafield tag="650" ind1=" " ind2="4"><subfield code="a">Parallel programming (Computer science)</subfield></datafield><datafield tag="700" ind1="1" ind2=" "><subfield code="a">Jeremiassen, Tor E.</subfield><subfield code="e">Verfasser</subfield><subfield code="4">aut</subfield></datafield><datafield tag="810" ind1="2" ind2=" "><subfield code="a">Department of Computer Science: Technical report</subfield><subfield code="t">University of Washington <Seattle, Wash.></subfield><subfield code="v">90,12,1</subfield><subfield code="w">(DE-604)BV008930431</subfield><subfield code="9">90,12,1</subfield></datafield><datafield tag="999" ind1=" " ind2=" "><subfield code="a">oai:aleph.bib-bvb.de:BVB01-005941720</subfield></datafield></record></collection> |
| id | DE-604.BV008992802 |
| illustrated | Not Illustrated |
| indexdate | 2024-07-09T17:28:08Z |
| institution | BVB |
| language | English |
| oai_aleph_id | oai:aleph.bib-bvb.de:BVB01-005941720 |
| oclc_num | 26797094 |
| open_access_boolean | |
| owner | DE-29T |
| owner_facet | DE-29T |
| physical | 21 S. |
| publishDate | 1990 |
| publishDateSearch | 1990 |
| publishDateSort | 1990 |
| record_format | marc |
| series2 | University of Washington <Seattle, Wash.> / Department of Computer Science: Technical report |
| spelling | Eggers, Susan J. Verfasser aut Eliminating false sharing Susan J. Eggers and Tor E. Jeremiassen Seattle, Wash. 1990 21 S. txt rdacontent n rdamedia nc rdacarrier University of Washington <Seattle, Wash.> / Department of Computer Science: Technical report 90,12,1 Abstract: "The performance and scalability of bus-based, shared memory multiprocessors is limited by the amount of bus traffic. Previous studies have shown that for machines with large caches and a write- invalidate coherency protocol, most of the bus traffic stems from coherency overhead. Some of this overhead is unavoidable, a direct consequence of the true sharing activity in the program. However, for caches with multi- word cache blocks, coherency overhead can also be caused by multiple processors accessing different words in the same block, known as false sharing. This paper measures the amount of false sharing in two types of parallel applications. For some applications the amount of false sharing increases substantially as block size increases For these programs there is a potential for improving performance by reducing the amount of false sharing. For other applications the amount of false sharing is so small as to have no significant impact on performance. For the programs that exhibit large amounts of false sharing we present two simple transformations for restructuring shared data, and apply them to the shared objects responsible for most of the false sharing. Simulations of the traces from the modified programs indicate that the transformations reduce the miss rates significantly. The false sharing miss rates decreased by 50% to 60%, the total miss rates by 20% to 40%, the coherency related bus traffic by 50% to 75%, and bus utilization by an average of more than 15% Shared misses were reduced by an order of magnitude more than in a previous study which applied a different type of transformation. Parallel programming (Computer science) Jeremiassen, Tor E. Verfasser aut Department of Computer Science: Technical report University of Washington <Seattle, Wash.> 90,12,1 (DE-604)BV008930431 90,12,1 |
| spellingShingle | Eggers, Susan J. Jeremiassen, Tor E. Eliminating false sharing Parallel programming (Computer science) |
| title | Eliminating false sharing |
| title_auth | Eliminating false sharing |
| title_exact_search | Eliminating false sharing |
| title_full | Eliminating false sharing Susan J. Eggers and Tor E. Jeremiassen |
| title_fullStr | Eliminating false sharing Susan J. Eggers and Tor E. Jeremiassen |
| title_full_unstemmed | Eliminating false sharing Susan J. Eggers and Tor E. Jeremiassen |
| title_short | Eliminating false sharing |
| title_sort | eliminating false sharing |
| topic | Parallel programming (Computer science) |
| topic_facet | Parallel programming (Computer science) |
| volume_link | (DE-604)BV008930431 |
| work_keys_str_mv | AT eggerssusanj eliminatingfalsesharing AT jeremiassentore eliminatingfalsesharing |