Status as of Thursday, December 8
Posted: Thu Dec 08, 2016 8:27 am
Good morning! Here's a report of where we're at:
- There has been a change in the behavior of many A4s. It seems that many A4s now reconnect to the server frequently and hold the old connection open without submitting any shares on it. Someone pointed out that the cause may be that firmware causes the miner to restart every time a board crashes and to create a new connection, instead of just sending shares over the same connection. That's why the number of "connected workers" is always in the 1000s now, even though the number of actual miners is around 400.
- If you have this problem, then your miners' hashrate cannot be calculated. You will see "calculating hashrate" indefinitely, because the miners reconnect before they submit enough shares to compute hashrate. To resolve the problem, hold just one connection open and submit shares through it, instead of creating a new connection when the miner is restarted. We would like comments from users as to what caused this change and whether Innosilicon should be E-Mailed to assist. Alternatively, perhaps the problem can be resolved with some configuration changes from the miners.
- The mining server has been tested for 4000 workers, which was the maximum number we could even get to run on our desktop systems, so the server can deal with many more than the number of connections the pool is currently holding. To prevent future problems, we already tried two solutions. First, we tried disconnecting the latest connection if a worker connected twice. Since the new connection was the one that submitted shares, that led to miners being disconnected completely and we reverted. Second, we tried disconnecting the existing connection when a duplicate miner connected. That triggered cgminer-3.7.2 to reconnect the disconnected connection, which would then automatically try to reconnect, which would then disconnect what used to be the new connection, and so on, resulting in an infinite loop. Our next try will be to hold connections to these miners, and then save CPU power calculating the most profitable coin by not sending any work to miners who haven't submitted any shares in 2h. After that, they will either disconnect or they will eventually find a (stale) share, after which they will be sent up-to-date work. Comments on this new strategy are welcome.
- Meanwhile, work continues on multiple algorithm support, which is on schedule for release in February. We will start with X11 in February, add support for Ethereum in March and April, and then work on how to test miners to automatically switch to the most profitable algorithm during the summer. This weekend, we achieved the ability for miners to connect to the multiple-algorithm server and mine scrypt coins. Our goal is to have the mining server passed to Chris for testing by the end of the holidays. We may post a job to convert website charts to handle multiple algorithms, the work for which would be done during the holidays. If you are interested, contact us.
- We thank everyone for their great suggestions on improving the site and the mining server. Right now, we are managing two separate forks and any improvements made to the current site result in double work needing to merge them to the multiple algorithm fork. We appreciate and record everything that is suggested to us, and we plan to get back to addressing these improvements in February.