When using wasabi you should have an eye on your data usage. Outgoing traffic must be under your active storage. So for each TB stored you have one TB outgoing traffic. If you go above that limit your account will be suspended.
Does anyone know how much traffic is caused by farming a plot for one month? If it is above 100GB you have a problem. You have to take into account that s3fs and goofys transfer data in blocks, so even if you just request a single byte, much more data gets transferred.
Plotting is about $8 per TB, and cheapest S3 storage class is around $11.3 per TB per month(minimum 30 days) . The data retrieval and api requests don’t appear to be much(cost should be considerably less than storage cost) based on my observation of farming for the past a couple of hours.
However there are too many variables here, and I strongly recommend against using AWS for plotting and farming because there are many things that can go wrong and you may end up with a huge AWS bill. All things considered, I wish I hadn’t set it up on AWS and will be pleased even if it breaks even in the end…
I reckon that it might be profitable at beginning because public cloud offers you an opportunity to build something up at a larger scale than individuals. However given current network space and chia coin price, I don’t think it’s valid option anymore.
Good call out on the data usage. I’ve talked directly with Wasabi on this. Bandwidth used by goofys was acceptable, (~1Mbps per 50TB stored) - s3fs (at least in my configuration) was not (over 1Gbps per 50TB stored).
Won’t mind providing the goofys binaries with a correctly optimized read ahead size for farming, im also currently trying to figure out the exact same issue.
Goofys usually reads up to 20mb for a random access, from debugging the fuse layer they actually been all lower than 192k bytes reads, but it’s still to slow
Even with a max of 250k chunk size
I’m using backblaze right now and havent been charged for more than 20mb bandwirh (which is nothing) in 24h
At 20 plots, so it may not be as bad as we I been thinking.
I’ve won 3 times in the cloud already. I didn’t upload my plots (@chiaappliance did ), but I’ve moved large files to s3 before. I’d expect the s3-cli with multipart support to work just fine, and be able to resume if there’s network interruption.
99th percentile on the Raspberry Pi harvesting wassabi from my home internet is under 25 seconds. From Digital Ocean the 99th percentile is under 20 seconds. My last win was turned in in 26 seconds (And this was prior to my change to chiapos 1.0.3 which is even faster on proofs). It’s a real thing that’s really making XCH
chiapos was improved to read from the plot files concurrently, which can make a big difference in a networked setting where requests can be randomly delayed and/or potentially serviced by multiple backend hosts & disks.
@luckidog Hi! I have my plots hosted on google drive, chia does recognize my plots, but the response times are very slow, could you please help me to fix it?