BL
B.D.A. Luppes
info
Please Note
<p>This page displays the records of the person named above and is not linked to a unique person identifier. This record may need to be merged to a profile.</p>
2 records found
1
FPGAs in Big Data
On the transparent and efficient acceleration of big data frameworks
The increasing volume and latency requirements of big data impose challenges on the processing capacity of existing computing systems. FPGA accelerators can be leveraged to overcome these challenges, but questions remain as to how these accelerators are best deployed to accelerate big data frameworks. This work investigates how future big data frameworks should be designed such that they facilitate transparent and efficient integration with FPGA accelerators in the context of SQL workloads. Three big data frameworks are accelerated to answer this question. These implementations offload the evaluation of a regular expression filter to Tidre, an FPGA-based regular expression matcher. First, Dremio is accelerated to obtain a 1750x speedup of the filter operator. Second, Dask is accelerated and a speedup of 92x is achieved for the same operator. Lastly, an accelerated version of Dask distributed is implemented. This implementation is deployed in a cluster environment to attain an end-to-end speedup of 3.6x as well as a reduction of the total cost per query by 23%. It is identified that query exploration phases could be used to increase the impact of existing FPGA kernels. In addition, it is found that the batch size of batch-processing frameworks has a significant impact on the performance of FPGA accelerated big data frameworks in distributed setups. Tuning this batch size can increase end-to-end query throughput up to 2.1x. Finally, future big data frameworks should make use of hardware-friendly and language-independent in-memory formats such as Apache Arrow. Correct alignment of the memory buffers of this data format can further increase the throughput of the system by 2.5x.
...
The increasing volume and latency requirements of big data impose challenges on the processing capacity of existing computing systems. FPGA accelerators can be leveraged to overcome these challenges, but questions remain as to how these accelerators are best deployed to accelerate big data frameworks. This work investigates how future big data frameworks should be designed such that they facilitate transparent and efficient integration with FPGA accelerators in the context of SQL workloads. Three big data frameworks are accelerated to answer this question. These implementations offload the evaluation of a regular expression filter to Tidre, an FPGA-based regular expression matcher. First, Dremio is accelerated to obtain a 1750x speedup of the filter operator. Second, Dask is accelerated and a speedup of 92x is achieved for the same operator. Lastly, an accelerated version of Dask distributed is implemented. This implementation is deployed in a cluster environment to attain an end-to-end speedup of 3.6x as well as a reduction of the total cost per query by 23%. It is identified that query exploration phases could be used to increase the impact of existing FPGA kernels. In addition, it is found that the batch size of batch-processing frameworks has a significant impact on the performance of FPGA accelerated big data frameworks in distributed setups. Tuning this batch size can increase end-to-end query throughput up to 2.1x. Finally, future big data frameworks should make use of hardware-friendly and language-independent in-memory formats such as Apache Arrow. Correct alignment of the memory buffers of this data format can further increase the throughput of the system by 2.5x.
On the Enhancement of Intelligibility
Investigating the influence of different speech modifications on the intelligibility of speech in near-end noise
Several algorithms to enhance the intelligibility of speech in near-end noise were analyzed and implemented. The algorithms considered were assessed based on the intrusive instrumental intelligibility metric SIIB_Gauss. An implementation based on the direct optimization for this metric is assessed, as well as an implementation based on human induced speech modifications, including increased sound intensity, flattening of the spectral tilt, increased vowel duration and increased consonant-vowel ratio. Another implemented algorithm is the amplification of the transient component of speech. Results show that for increased vowel duration a decrease in intelligibility was found in SIIB_Gauss value as well as in informal listening tests. The other implementations did show an increase in intelligibility according to SIIB_Gauss at SNRs between -4 dB and 6 dB in both stationary and fluctuating noise, under a power constraint. Finally, the implementations were combined into a system that automatically selects the optimal algorithm to use under the given noise conditions. It is shown that this combined system is able to increase intelligibility of speech in the presence of non-fluctuating noise, fluctuating noise, speech shaped noise, and competing speaker noise.
...
Several algorithms to enhance the intelligibility of speech in near-end noise were analyzed and implemented. The algorithms considered were assessed based on the intrusive instrumental intelligibility metric SIIB_Gauss. An implementation based on the direct optimization for this metric is assessed, as well as an implementation based on human induced speech modifications, including increased sound intensity, flattening of the spectral tilt, increased vowel duration and increased consonant-vowel ratio. Another implemented algorithm is the amplification of the transient component of speech. Results show that for increased vowel duration a decrease in intelligibility was found in SIIB_Gauss value as well as in informal listening tests. The other implementations did show an increase in intelligibility according to SIIB_Gauss at SNRs between -4 dB and 6 dB in both stationary and fluctuating noise, under a power constraint. Finally, the implementations were combined into a system that automatically selects the optimal algorithm to use under the given noise conditions. It is shown that this combined system is able to increase intelligibility of speech in the presence of non-fluctuating noise, fluctuating noise, speech shaped noise, and competing speaker noise.