Does anyone have any data on:
1. The average latency of a connection.
2. The average data transfer rate of a connection.
3. Average number of lost connections per time of connection.
4. Percentage of transactions in new blocks, up-to-date nodes already have.
I'm trying to calculate the benefit of the proposal here: https://en.bitcoin.it/wiki/User:MatthewLM/ImprovedBlockRelayingProposal
The proposal increases latency but increases parallelism and removes redundancy. It may increase or decrease bandwidth usage depending upon the amount of redundancy saved vs the additional overhead.
Jeff Garzik seems to think that latency is the biggest problem when downloading blocks and additional latency would make things slower. Is there data to support this? Of-course, I'm considering the future: when block sizes could grow into several MBs and beyond. Shouldn't we prepare changes for this now?
Maybe 6 months ago... latency is much less of an issue now with the increase in size of the blocks. A dial-up connection has about 120-150ms base latency (transfer speed, ~5kB/s), a satellite connection approaching 1000ms (can reach 100kB/s+ easily).... If you're transferring anything over, uh, 30kB, it seems to me like it would be preferable to connect to the person using the satellite connection. It's not something that would require a lot of ACKs (like, say, online gaming)1. The average latency of a connection.
2. The average data transfer rate of a connection.
3. Average number of lost connections per time of connection.
4. Percentage of transactions in new blocks, up-to-date nodes already have.
I'm trying to calculate the benefit of the proposal here: https://en.bitcoin.it/wiki/User:MatthewLM/ImprovedBlockRelayingProposal
The proposal increases latency but increases parallelism and removes redundancy. It may increase or decrease bandwidth usage depending upon the amount of redundancy saved vs the additional overhead.
Jeff Garzik seems to think that latency is the biggest problem when downloading blocks and additional latency would make things slower. Is there data to support this? Of-course, I'm considering the future: when block sizes could grow into several MBs and beyond. Shouldn't we prepare changes for this now?
Umm, I have some log files that may be useful, I've been running with around 900 connections