Hardware implementation of dynamic load balancing in the parallel inference machine PIM c:
Abstract: "This paper proposes and evaluates the hardware implementation required for dynamic load balancing in the prototype PIM/c of the Parallel Inference Machine (PIM). On one hand, PIM/c is configured along a hierarchical structure of loosely coupled TCMP clusters (TCMP -- Tightly-Coupled...
Gespeichert in:
Format: | Buch |
---|---|
Sprache: | English |
Veröffentlicht: |
Tokyo, Japan
1991
|
Schriftenreihe: | Shin-Sedai-Konpyūta-Gijutsu-Kaihatsu-Kikō <Tōkyō>: ICOT technical report
704 |
Schlagworte: | |
Zusammenfassung: | Abstract: "This paper proposes and evaluates the hardware implementation required for dynamic load balancing in the prototype PIM/c of the Parallel Inference Machine (PIM). On one hand, PIM/c is configured along a hierarchical structure of loosely coupled TCMP clusters (TCMP -- Tightly-Coupled Multi-Processors) in order to utilize the high locality of logic programs. On the other hand, an LCMP (Loosely-Coupled Multi- Processors) approach enables us to build a more scalable machine using a crossbar network. Load balancing algorithms and corresponding hardware suitable for each hierarchy are used. First, for dynamic load balancing in the TCMP hierarchy, we propose a register with broadcast facility to request load dispatching The evaluation determines the overhead due to memory polling in order to detect the request. The proposed hardware reduces the execution time of logic programs by 15%. Second, for dynamic load balancing in the LCMP hierarchy, we propose the use of a shortcut path to request the value of a total load within a cluster. The evaluation shows that the overhead due to the request of that value is reduced as a result of introducing the shortcut path. Consequently the proposed hardware reduces the processing time by 50%. The results obtained confirm that the use of hardware mechanisms reduces the overhead due to the dynamic load balancing. |
Beschreibung: | 13 S. |
Internformat
MARC
LEADER | 00000nam a2200000 cb4500 | ||
---|---|---|---|
001 | BV010956020 | ||
003 | DE-604 | ||
005 | 00000000000000.0 | ||
007 | t | ||
008 | 960918s1991 |||| 00||| engod | ||
035 | |a (OCoLC)26483711 | ||
035 | |a (DE-599)BVBBV010956020 | ||
040 | |a DE-604 |b ger |e rakddb | ||
041 | 0 | |a eng | |
049 | |a DE-91G | ||
245 | 1 | 0 | |a Hardware implementation of dynamic load balancing in the parallel inference machine PIM c |c by T.Nakagawa ... |
264 | 1 | |a Tokyo, Japan |c 1991 | |
300 | |a 13 S. | ||
336 | |b txt |2 rdacontent | ||
337 | |b n |2 rdamedia | ||
338 | |b nc |2 rdacarrier | ||
490 | 1 | |a Shin-Sedai-Konpyūta-Gijutsu-Kaihatsu-Kikō <Tōkyō>: ICOT technical report |v 704 | |
520 | 3 | |a Abstract: "This paper proposes and evaluates the hardware implementation required for dynamic load balancing in the prototype PIM/c of the Parallel Inference Machine (PIM). On one hand, PIM/c is configured along a hierarchical structure of loosely coupled TCMP clusters (TCMP -- Tightly-Coupled Multi-Processors) in order to utilize the high locality of logic programs. On the other hand, an LCMP (Loosely-Coupled Multi- Processors) approach enables us to build a more scalable machine using a crossbar network. Load balancing algorithms and corresponding hardware suitable for each hierarchy are used. First, for dynamic load balancing in the TCMP hierarchy, we propose a register with broadcast facility to request load dispatching | |
520 | 3 | |a The evaluation determines the overhead due to memory polling in order to detect the request. The proposed hardware reduces the execution time of logic programs by 15%. Second, for dynamic load balancing in the LCMP hierarchy, we propose the use of a shortcut path to request the value of a total load within a cluster. The evaluation shows that the overhead due to the request of that value is reduced as a result of introducing the shortcut path. Consequently the proposed hardware reduces the processing time by 50%. The results obtained confirm that the use of hardware mechanisms reduces the overhead due to the dynamic load balancing. | |
650 | 4 | |a Fifth generation computers | |
650 | 4 | |a Parallel processing (Electronic computers) | |
700 | 1 | |a Nakagawa, Toshio |e Sonstige |4 oth | |
830 | 0 | |a Shin-Sedai-Konpyūta-Gijutsu-Kaihatsu-Kikō <Tōkyō>: ICOT technical report |v 704 |w (DE-604)BV010923438 |9 704 | |
999 | |a oai:aleph.bib-bvb.de:BVB01-007328209 |
Datensatz im Suchindex
_version_ | 1804125443184394240 |
---|---|
any_adam_object | |
building | Verbundindex |
bvnumber | BV010956020 |
ctrlnum | (OCoLC)26483711 (DE-599)BVBBV010956020 |
format | Book |
fullrecord | <?xml version="1.0" encoding="UTF-8"?><collection xmlns="http://www.loc.gov/MARC21/slim"><record><leader>02419nam a2200313 cb4500</leader><controlfield tag="001">BV010956020</controlfield><controlfield tag="003">DE-604</controlfield><controlfield tag="005">00000000000000.0</controlfield><controlfield tag="007">t</controlfield><controlfield tag="008">960918s1991 |||| 00||| engod</controlfield><datafield tag="035" ind1=" " ind2=" "><subfield code="a">(OCoLC)26483711</subfield></datafield><datafield tag="035" ind1=" " ind2=" "><subfield code="a">(DE-599)BVBBV010956020</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-91G</subfield></datafield><datafield tag="245" ind1="1" ind2="0"><subfield code="a">Hardware implementation of dynamic load balancing in the parallel inference machine PIM c</subfield><subfield code="c">by T.Nakagawa ...</subfield></datafield><datafield tag="264" ind1=" " ind2="1"><subfield code="a">Tokyo, Japan</subfield><subfield code="c">1991</subfield></datafield><datafield tag="300" ind1=" " ind2=" "><subfield code="a">13 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">Shin-Sedai-Konpyūta-Gijutsu-Kaihatsu-Kikō <Tōkyō>: ICOT technical report</subfield><subfield code="v">704</subfield></datafield><datafield tag="520" ind1="3" ind2=" "><subfield code="a">Abstract: "This paper proposes and evaluates the hardware implementation required for dynamic load balancing in the prototype PIM/c of the Parallel Inference Machine (PIM). On one hand, PIM/c is configured along a hierarchical structure of loosely coupled TCMP clusters (TCMP -- Tightly-Coupled Multi-Processors) in order to utilize the high locality of logic programs. On the other hand, an LCMP (Loosely-Coupled Multi- Processors) approach enables us to build a more scalable machine using a crossbar network. Load balancing algorithms and corresponding hardware suitable for each hierarchy are used. First, for dynamic load balancing in the TCMP hierarchy, we propose a register with broadcast facility to request load dispatching</subfield></datafield><datafield tag="520" ind1="3" ind2=" "><subfield code="a">The evaluation determines the overhead due to memory polling in order to detect the request. The proposed hardware reduces the execution time of logic programs by 15%. Second, for dynamic load balancing in the LCMP hierarchy, we propose the use of a shortcut path to request the value of a total load within a cluster. The evaluation shows that the overhead due to the request of that value is reduced as a result of introducing the shortcut path. Consequently the proposed hardware reduces the processing time by 50%. The results obtained confirm that the use of hardware mechanisms reduces the overhead due to the dynamic load balancing.</subfield></datafield><datafield tag="650" ind1=" " ind2="4"><subfield code="a">Fifth generation computers</subfield></datafield><datafield tag="650" ind1=" " ind2="4"><subfield code="a">Parallel processing (Electronic computers)</subfield></datafield><datafield tag="700" ind1="1" ind2=" "><subfield code="a">Nakagawa, Toshio</subfield><subfield code="e">Sonstige</subfield><subfield code="4">oth</subfield></datafield><datafield tag="830" ind1=" " ind2="0"><subfield code="a">Shin-Sedai-Konpyūta-Gijutsu-Kaihatsu-Kikō <Tōkyō>: ICOT technical report</subfield><subfield code="v">704</subfield><subfield code="w">(DE-604)BV010923438</subfield><subfield code="9">704</subfield></datafield><datafield tag="999" ind1=" " ind2=" "><subfield code="a">oai:aleph.bib-bvb.de:BVB01-007328209</subfield></datafield></record></collection> |
id | DE-604.BV010956020 |
illustrated | Not Illustrated |
indexdate | 2024-07-09T18:01:38Z |
institution | BVB |
language | English |
oai_aleph_id | oai:aleph.bib-bvb.de:BVB01-007328209 |
oclc_num | 26483711 |
open_access_boolean | |
owner | DE-91G DE-BY-TUM |
owner_facet | DE-91G DE-BY-TUM |
physical | 13 S. |
publishDate | 1991 |
publishDateSearch | 1991 |
publishDateSort | 1991 |
record_format | marc |
series | Shin-Sedai-Konpyūta-Gijutsu-Kaihatsu-Kikō <Tōkyō>: ICOT technical report |
series2 | Shin-Sedai-Konpyūta-Gijutsu-Kaihatsu-Kikō <Tōkyō>: ICOT technical report |
spelling | Hardware implementation of dynamic load balancing in the parallel inference machine PIM c by T.Nakagawa ... Tokyo, Japan 1991 13 S. txt rdacontent n rdamedia nc rdacarrier Shin-Sedai-Konpyūta-Gijutsu-Kaihatsu-Kikō <Tōkyō>: ICOT technical report 704 Abstract: "This paper proposes and evaluates the hardware implementation required for dynamic load balancing in the prototype PIM/c of the Parallel Inference Machine (PIM). On one hand, PIM/c is configured along a hierarchical structure of loosely coupled TCMP clusters (TCMP -- Tightly-Coupled Multi-Processors) in order to utilize the high locality of logic programs. On the other hand, an LCMP (Loosely-Coupled Multi- Processors) approach enables us to build a more scalable machine using a crossbar network. Load balancing algorithms and corresponding hardware suitable for each hierarchy are used. First, for dynamic load balancing in the TCMP hierarchy, we propose a register with broadcast facility to request load dispatching The evaluation determines the overhead due to memory polling in order to detect the request. The proposed hardware reduces the execution time of logic programs by 15%. Second, for dynamic load balancing in the LCMP hierarchy, we propose the use of a shortcut path to request the value of a total load within a cluster. The evaluation shows that the overhead due to the request of that value is reduced as a result of introducing the shortcut path. Consequently the proposed hardware reduces the processing time by 50%. The results obtained confirm that the use of hardware mechanisms reduces the overhead due to the dynamic load balancing. Fifth generation computers Parallel processing (Electronic computers) Nakagawa, Toshio Sonstige oth Shin-Sedai-Konpyūta-Gijutsu-Kaihatsu-Kikō <Tōkyō>: ICOT technical report 704 (DE-604)BV010923438 704 |
spellingShingle | Hardware implementation of dynamic load balancing in the parallel inference machine PIM c Shin-Sedai-Konpyūta-Gijutsu-Kaihatsu-Kikō <Tōkyō>: ICOT technical report Fifth generation computers Parallel processing (Electronic computers) |
title | Hardware implementation of dynamic load balancing in the parallel inference machine PIM c |
title_auth | Hardware implementation of dynamic load balancing in the parallel inference machine PIM c |
title_exact_search | Hardware implementation of dynamic load balancing in the parallel inference machine PIM c |
title_full | Hardware implementation of dynamic load balancing in the parallel inference machine PIM c by T.Nakagawa ... |
title_fullStr | Hardware implementation of dynamic load balancing in the parallel inference machine PIM c by T.Nakagawa ... |
title_full_unstemmed | Hardware implementation of dynamic load balancing in the parallel inference machine PIM c by T.Nakagawa ... |
title_short | Hardware implementation of dynamic load balancing in the parallel inference machine PIM c |
title_sort | hardware implementation of dynamic load balancing in the parallel inference machine pim c |
topic | Fifth generation computers Parallel processing (Electronic computers) |
topic_facet | Fifth generation computers Parallel processing (Electronic computers) |
volume_link | (DE-604)BV010923438 |
work_keys_str_mv | AT nakagawatoshio hardwareimplementationofdynamicloadbalancingintheparallelinferencemachinepimc |