Dan Luu

I'm haven't debugged this enough to be sure that multi-threading is implicated, but it seems plausible. Here's the output: Run 1: ~~~ 5: 5: Thread Thread 1 exited. ~~~ Run...

The gtest code warns: > Google Test is designed to be thread-safe. The implementation is thread-safe on systems where the pthreads library is available. It is currently unsafe to use...

BitFunnelTool prints to std::cout regardless of output stream argument

1

One way to fix #370 while still keeping the end-to-end nature of the test intact would be to put the result into a stream and then look at the stream....

we have configDirectory, statisticsDirectory, and indexDirectory. Just for example, shardDefinition lives in statisticsDirectory. In BitFunnelToolTest, this means that the file is in ~~~ config/ShardDefinition.csv ~~~ FileManager appears to take a...

Thread per core vs. thread per hyper-core vs. something else?

1

The current system runs on thread per hyper-core. But this hasn't been benchmarked for a long time and the system has changed substantially since then. We should benchmark this when...

BitFunnel slows down when we consecutively add more threads

1

Case 1: Set threads to 16 and run a query log Case 2: Set threads to 1 run query log Set threads to 2 run query log ... Set threads...

There are a number of known issues with the current ingestion pipeline. Let's say we fix those but continue to use `LOCK BTS`. Suppose the `LOCK` costs us 200 cycles...

I could have sworn we had an issue on this but I can't find it. I'm not actually trying to solve this question right now, but I ended up looking...

We currently support up to 6. If we index all of wikipedia (English articles only), we see terms with frequency 1e-7, i.e., idfX10 7. I'm not going to do this...

In each section of the table below, we ask for 10, 100, and 1000 SNR, respectively. Instead of getting an order of magnitude increase in SNR, we get something that's...