Status as of Monday, November 17

[Steve Sokolowski]

I'm back to being hard at work on this project after a frustrating few weeks dealing with rejected shares. Concurrency errors have been the cause of most of the bugs that we have encountered over the past month or two, but I think that we are close to resolving all known issues. First, here is a list of the new things in this release from yesterday:

• I adjusted the algorithm which is used to calculate dynamic difficulties. Previously, the interval in which the number of shares was looked at was too short, so that if a miner had very bad luck, then he could be assigned a difficulty of 1024 even if the previous difficulty was 65536. This change should reduce the number of difficulty adjustments for dynamic difficulty miners.
• Difficulties are now sent out on a per-connection basis, rather than a per-worker basis, which resolves many of the spikes encountered when people connect lots of miners to the pool with the same worker.
• The "current status" chart was moved to the bottom of the charts page and was significantly expanded with additional information. This information will allow people who want to mine single coins to make an informed decision based upon the information in the chart.
• Daemon-network-management, which throttles the daemons that are not being mined, was adjusted. Before, we unthrottled the daemon that was currently being mined, but that doesn't work anymore since many coins can be mined simultaneously. Now, we only unthrottle daemons when a block is found. The "found block" daemon has unmetered upload bandwidth for 15s, while the rest of them are limited to 15Mbps combined. Because we aren't using up bandwidth on unnecessary daemons, this might provide a site performance increase if we ever become bandwidth constrained.
• The reasons for rejected shares are now more clear to the users. Instead of errors like "job not found," you will now see "stale share," "merge mining block was stale," "miner did not change difficulty upon request," and so on.
• I performed some research into mining equipment and determined that some miners do not respond correctly to difficulty change requests. The Stratum protocol specification states that a difficulty change request is effective on the next block, while some miners incorrectly implement the protocol by assuming that the difficulty change occurs immediately. I tried to reduce this issue as much as possible, but if your miners are getting rejected shares because of the "miner not changing difficulty upon request," that is a problem with your manufacturer's firmware and you can resolve it by using a static difficulty.
• A new URL, https://prohashing.com/exchange/instantProfitability, provides the current instantaneous profitability for the pool. Developers of "meta-pools" who want to switch miners between pools can get the current profitability from this URL.
• The number of coin switches has been reduced. Now, we do not send out new work when a transaction is received in the dogecoin network, because the transaction fees are minimal and not worth the effort. Also, we do not send out work when a transaction on the primary coin increases profitability by less than 1%.
• There were some changes to the payout algorithms to make payouts more reliable. Every day, there seems to be another error with coins that have small balances due, requiring manual intervention. Now, however, we have errors down to 15%, and I would imagine that we will be able to have an error-free run across all coins by the end of the week. Most of these errors are caused by issues with specific coins that require workarounds.
• Finally, there is one outstanding issue that can cause some shares to be incorrectly recorded as duplicates, and it appears rarely, about 0.5% of the time. Even so, we have a fix that will eliminate those (and increase profitability by 0.5%), and we'll deploy it after things are tested. There is also an error that has been fixed that sends double work restarts when a dogecoin block is received, which will reduce work restarts by 33% after tested.

In other news, Cryptsy had another outage this morning. Multiple exchanges continues to remain a top priority after the remainder of the code is proven rock solid.

[Steve Sokolowski]

I issued the release this evening to reduce the number of work restarts and to correct the duplicate shares issue.