Status as of Tuesday, September 5, 2017
Posted: Tue Sep 05, 2017 8:33 am
A few comments for today:
- The code to reduce data usage upon disconnect was deployed yesterday and again reduced bandwidth for live data. This was the second success after we had done the same for connects as well.
- Chris was able to make a discovery last night about coin assignment. It turns out that part or all of the 100% CPU spikes might be caused by what happens when a coin like litecoin or bitconnectcoin gets a new block that we didn't find, and every person in the pool is reevaluated for a new assignment. During this time, disconnects from WAMP and other services occur because the server can't respond in time. The next step is to profile that time period and determine if most of the time is spent sending data to miners or if most of the time is spent in coin assignment itself.
- We are aware that some users are experiencing low efficiencies. The issue is caused by how the coin assignment algorithm was made less exact as a temporary measure a few months ago. Once the system is parallelized, then we can devote more CPU power to coin assignment and send fewer work restarts to miners, improving efficiency.
- Some users are reporting that they are earning less than previous days, but since there isn't extra profit left over, it's unclear what the cause of the problem is. Undoubtedly, there are a few people who misunderstand how mining works and that one should generally expect to earn less over time. However, we also know of a few issues with share insertion that theoretically could use bad pricing data in some instances. We won't have the manpower to investigate these until Thursday because we are prioritizing the CPU usage issue, which is a definite problem we now have a concrete path forward on.