Publications Partially Reconfigurable Platforms
Relational Memory: Native In-Memory Accesses on Rows and Columns
2023 International Conference on Extending Database Technology (EDBT), 2023, Ioannina, Greece
Abstract
Analytical database systems are typically designed to use a columnfirst data layout to access only the desired fields. On the other hand, storing data row-first works great for accessing, inserting, or updating entire rows. Transforming rows to columns at run-time is expensive, hence, many analytical systems ingest data in row-first form and transform it in the background to columns to facilitate future analytical queries. How will this design change if we can always efficiently access only the desired set of columns? To address this question, we present a radically new approach to data transformation from rows to columns. We build upon recent advancements in embedded platforms with re-programmable logic to design native in-memory access on rows and columns. Our approach, termed Relational Memory (RM), relies on an FPGA-based accelerator that sits between the CPU and main memory and transparently transforms base data to any group of columns with minimal overhead at runtime. This design allows accessing any group of columns as if it already exists in memory. We implement and deploy RM in real hardware, and we show that we can access the desired columns up to 1.63× faster compared to a row-wise layout, while matching the performance of pure columnar access for low projectivity, and outperforming it by up to 2.23× as projectivity (and tuple reconstruction cost) increases. Overall, RM allows the CPU to access the optimal data layout, radically reducing unnecessary data movement without high data transformation costs, thus, simplifying software complexity and physical design, while accelerating query execution.
Cite this paper
@inproceedings{RelMem,
title = {{Relational Memory: Native In-Memory Accesses on Rows and Columns}},
author = {Roozkhosh, Shahin and Hoornaert, Denis and Mun, Ju-Hyoung and Papon, Tarikul Islam and Drepper, Ulrich and Mancuso, Renato and Athanassoulis, Manos},
booktitle = {2023 International Conference on Extending Database Technology (EDBT)},
year = 2023,
address = {Ioannina, Greece},
doi = {10.48786/edbt.2023.06},
url = {https://doi.org/10.48786/edbt.2023.06}
}
TY - CONF AU - Roozkhosh, Shahin AU - Hoornaert, Denis AU - Mun, Ju-Hyoung AU - Papon, Tarikul Islam AU - Drepper, Ulrich AU - Mancuso, Renato AU - Athanassoulis, Manos TI - Relational Memory: Native In-Memory Accesses on Rows and Columns T2 - 2023 International Conference on Extending Database Technology (EDBT) PY - 2023 CY - Ioannina, Greece DO - 10.48786/edbt.2023.06 UR - https://doi.org/10.48786/edbt.2023.06 AB - Analytical database systems are typically designed to use a columnfirst data layout to access only the desired fields. On the other hand, storing data row-first works great for accessing, inserting, or updating entire rows. Transforming rows to columns at run-time is expensive, hence, many analytical systems ingest data in row-first form and transform it in the background to columns to facilitate future analytical queries. How will this design change if we can always efficiently access only the desired set of columns? To address this question, we present a radically new approach to data transformation from rows to columns. We build upon recent advancements in embedded platforms with re-programmable logic to design native in-memory access on rows and columns. Our approach, termed Relational Memory (RM), relies on an FPGA-based accelerator that sits between the CPU and main memory and transparently transforms base data to any group of columns with minimal overhead at runtime. This design allows accessing any group of columns as if it already exists in memory. We implement and deploy RM in real hardware, and we show that we can access the desired columns up to 1.63× faster compared to a row-wise layout, while matching the performance of pure columnar access for low projectivity, and outperforming it by up to 2.23× as projectivity (and tuple reconstruction cost) increases. Overall, RM allows the CPU to access the optimal data layout, radically reducing unnecessary data movement without high data transformation costs, thus, simplifying software complexity and physical design, while accelerating query execution. ER -
%0 Conference Paper %A Roozkhosh, Shahin %A Hoornaert, Denis %A Mun, Ju-Hyoung %A Papon, Tarikul Islam %A Drepper, Ulrich %A Mancuso, Renato %A Athanassoulis, Manos %T Relational Memory: Native In-Memory Accesses on Rows and Columns %B 2023 International Conference on Extending Database Technology (EDBT) %D 2023 %C Ioannina, Greece %R 10.48786/edbt.2023.06 %U https://doi.org/10.48786/edbt.2023.06 %X Analytical database systems are typically designed to use a columnfirst data layout to access only the desired fields. On the other hand, storing data row-first works great for accessing, inserting, or updating entire rows. Transforming rows to columns at run-time is expensive, hence, many analytical systems ingest data in row-first form and transform it in the background to columns to facilitate future analytical queries. How will this design change if we can always efficiently access only the desired set of columns? To address this question, we present a radically new approach to data transformation from rows to columns. We build upon recent advancements in embedded platforms with re-programmable logic to design native in-memory access on rows and columns. Our approach, termed Relational Memory (RM), relies on an FPGA-based accelerator that sits between the CPU and main memory and transparently transforms base data to any group of columns with minimal overhead at runtime. This design allows accessing any group of columns as if it already exists in memory. We implement and deploy RM in real hardware, and we show that we can access the desired columns up to 1.63× faster compared to a row-wise layout, while matching the performance of pure columnar access for low projectivity, and outperforming it by up to 2.23× as projectivity (and tuple reconstruction cost) increases. Overall, RM allows the CPU to access the optimal data layout, radically reducing unnecessary data movement without high data transformation costs, thus, simplifying software complexity and physical design, while accelerating query execution.
[
{
"id": "RelMem",
"type": "paper-conference",
"title": "Relational Memory: Native In-Memory Accesses on Rows and Columns",
"author": [
{
"family": "Roozkhosh",
"given": "Shahin"
},
{
"family": "Hoornaert",
"given": "Denis"
},
{
"family": "Mun",
"given": "Ju-Hyoung"
},
{
"family": "Papon",
"given": "Tarikul Islam"
},
{
"family": "Drepper",
"given": "Ulrich"
},
{
"family": "Mancuso",
"given": "Renato"
},
{
"family": "Athanassoulis",
"given": "Manos"
}
],
"container-title": "2023 International Conference on Extending Database Technology (EDBT)",
"issued": {
"date-parts": [
[
2023
]
]
},
"publisher-place": "Ioannina, Greece",
"DOI": "10.48786/edbt.2023.06",
"abstract": "Analytical database systems are typically designed to use a columnfirst data layout to access only the desired fields. On the other hand, storing data row-first works great for accessing, inserting, or updating entire rows. Transforming rows to columns at run-time is expensive, hence, many analytical systems ingest data in row-first form and transform it in the background to columns to facilitate future analytical queries. How will this design change if we can always efficiently access only the desired set of columns? To address this question, we present a radically new approach to data transformation from rows to columns. We build upon recent advancements in embedded platforms with re-programmable logic to design native in-memory access on rows and columns. Our approach, termed Relational Memory (RM), relies on an FPGA-based accelerator that sits between the CPU and main memory and transparently transforms base data to any group of columns with minimal overhead at runtime. This design allows accessing any group of columns as if it already exists in memory. We implement and deploy RM in real hardware, and we show that we can access the desired columns up to 1.63× faster compared to a row-wise layout, while matching the performance of pure columnar access for low projectivity, and outperforming it by up to 2.23× as projectivity (and tuple reconstruction cost) increases. Overall, RM allows the CPU to access the optimal data layout, radically reducing unnecessary data movement without high data transformation costs, thus, simplifying software complexity and physical design, while accelerating query execution."
}
]
S. Roozkhosh, D. Hoornaert, J. H. Mun, T. I. Papon, U. Drepper, R. Mancuso, and M. Athanassoulis, “Relational Memory: Native In-Memory Accesses on Rows and Columns,” in 2023 International Conference on Extending Database Technology (EDBT), 2023, doi: 10.48786/edbt.2023.06.
Roozkhosh, S., Hoornaert, D., Mun, J. H., Papon, T. I., Drepper, U., Mancuso, R., & Athanassoulis, M. (2023). Relational Memory: Native In-Memory Accesses on Rows and Columns. In 2023 International Conference on Extending Database Technology (EDBT). https://doi.org/10.48786/edbt.2023.06