j********x
发帖数: 2330 | 1 http://www.eecs.berkeley.edu/~hzhao/papers/BD.pdf
We made a case for the importance of single-node enhancements
for cost and power efficiency in large-scale behavioral
data analysis. Single node performance of the BID
Data suite on several common benchmarks is faster than
generic cluster systems (Hadoop, Spark) with 50-100 nodes
and competitive with custom cluster implementations for
specific algorithms. We did not describe any MCMC algorithms,
but these also are an important part of the BID
Data roadmap in future. The toolkit will also include a substantial
subset of tools for causal analysis, and disk-scale
GPU-assisted sorting for data indexing and search tasks. | g*****g
发帖数: 34805 | 2 All costs should take into development/maintenance efforts into account.
Also a single node model can be efficient in many ways but once it hits
hardware limit, it's super expensive to scale up and it's not tolerant to
hardware failures. We all know that too well.
【在 j********x 的大作中提到】
: http://www.eecs.berkeley.edu/~hzhao/papers/BD.pdf
: We made a case for the importance of single-node enhancements
: for cost and power efficiency in large-scale behavioral
: data analysis. Single node performance of the BID
: Data suite on several common benchmarks is faster than
: generic cluster systems (Hadoop, Spark) with 50-100 nodes
: and competitive with custom cluster implementations for
: specific algorithms. We did not describe any MCMC algorithms,
: but these also are an important part of the BID
: Data roadmap in future. The toolkit will also include a substantial
| j********x
发帖数: 2330 | 3 其实都不是too well 技术红卫兵在本版是常态 理性讨论水平较低
【在 g*****g 的大作中提到】
: All costs should take into development/maintenance efforts into account.
: Also a single node model can be efficient in many ways but once it hits
: hardware limit, it's super expensive to scale up and it's not tolerant to
: hardware failures. We all know that too well.
|
|
