Closed lifepuzzlefun closed 2 months ago
when single node in high produce qps ( 8w qps for 8c16g single node ) the thread may not enough cause high produce latency.
Could we automatically make them scaling by CPU_CORES likes
public StreamReaders(long size, ObjectManager objectManager, ObjectStorage objectStorage,
ObjectReaderFactory objectReaderFactory) {
this(size, objectManager, objectStorage, objectReaderFactory, Systems.CPU_CORES);
}
Currently, the thread pool threads number is based on 2 Core.
when single node in high produce qps ( 8w qps for 8c16g single node ) the thread may not enough cause high produce latency.
Could we automatically make them scaling by CPU_CORES likes
public StreamReaders(long size, ObjectManager objectManager, ObjectStorage objectStorage, ObjectReaderFactory objectReaderFactory) { this(size, objectManager, objectStorage, objectReaderFactory, Systems.CPU_CORES); }
Currently, the thread pool threads number is based on 2 Core.
agree. and do you think we should introduce the thread number or just automatically set to 2 * Systems.CPU_CROES.
for 2cpu deploy the thread number will be drop from 8 to 4.
or keep the config and change the default to 2 * CPU
when single node in high produce qps ( 8w qps for 8c16g single node ) the thread may not enough cause high produce latency.
Could we automatically make them scaling by CPU_CORES likes
public StreamReaders(long size, ObjectManager objectManager, ObjectStorage objectStorage, ObjectReaderFactory objectReaderFactory) { this(size, objectManager, objectStorage, objectReaderFactory, Systems.CPU_CORES); }
Currently, the thread pool threads number is based on 2 Core.
agree. and do you think we should introduce the thread number or just automatically set to 2 * Systems.CPU_CROES.
for 2cpu deploy the thread number will be drop from 8 to 4.
or keep the config and change the default to 2 * CPU
I think auto-set is enough
when single node in high produce qps ( 8w qps for 8c16g single node ) the thread may not enough cause high produce latency.