Status as of Wednesday, July 8
Posted: Wed Jul 08, 2015 9:26 am
A few notes for today:
- Thanks for all the input yesterday. After some research and discussion, Chris and I came to agree with kires that we aren't willing to use a RAID 0 for the database server, or even a JBOD setup. Additionally, we determined that the size of the database will be around 0.6 TB after we finish reloading the block explorer data. We determined that we can't afford the $1500 required to buy the hardware necessary for an all-SSD solution at this time. Also, because the change would require a downtime of 12-24 hours, additional lost profit would add to our costs.
- Instead, we are going to continue with software optimizations over the weekend. I identified some code that is unnecessarily deleting and then inserting identical rows into some tables, and we can reduce the size of transactions to reduce writes. There are also some postgres configuration parameters we are investigating. We'll see what effect these changes have on performance, especially after block explorer reindexing finishes, and then reevaluate next week. Unless hashrate quadruples during that period, we shouldn't have any performance issues for normal load.
- We determined the cause of the outages that were occurring at midnight. It turns out that the query that computes the percentage of time mining coins every day locks tables and takes far too long. For now, we'll just disable the query. Over the weekend, I will see if there is an alternative way to compute that chart, and if we can't do it, we'll just remove the chart from the website. Note that the data in the chart right now is incorrect and displays a hashrate that is too high; we have a fix for that but can't run it. Tonight, we should not see this issue recur.