#448 closed enhancement (fixed)
download: speak to as few servers as possible
| Reported by: | warner | Owned by: | warner |
|---|---|---|---|
| Priority: | major | Milestone: | 1.8.0 |
| Component: | code-peerselection | Version: | 1.0.0 |
| Keywords: | download performance | Cc: | |
| Launchpad Bug: |
Description
Our old roadmap.txt pointed out, in "Download Peer Selection step 2", that the current simplistic download code asks all known peers for the shares that they hold, whereas we really only need to ask enough peers to get "k" shares (plus a few extra in case we lose some during download, see #287).
This will probably occur about the same time we do a large-scale refactoring of the download code.
Change History (9)
comment:1 Changed at 2010-02-27T06:42:15Z by zooko
- Milestone changed from undecided to 1.7.0
comment:2 Changed at 2010-02-27T06:44:32Z by zooko
- Owner set to warner
comment:3 Changed at 2010-02-27T17:48:44Z by zooko
- Summary changed from download: only speak to minimal number of peers to download: speak to as few servers as possible
comment:4 Changed at 2010-05-08T19:43:50Z by zooko
comment:5 Changed at 2010-05-08T19:45:44Z by zooko
If you like this ticket, you might also like the "Brian's New Downloader" bundle of tickets: #800 (improve alacrity by downloading only the part of the Merkle Tree that you need), #798 (improve random-access download to retrieve/decrypt less data), #809 (Measure how segment size affects upload/download speed.), #287 (download: tolerate lost or missing servers), and #605 (two-hour delay to connect to a grid from Win32, if there are many storage servers unreachable).
comment:6 Changed at 2010-05-08T22:48:46Z by zooko
- Milestone changed from 1.7.0 to 1.8.0
Brian's New Downloader is now planned for v1.8.0.
comment:7 Changed at 2010-08-10T04:17:32Z by davidsarah
To what extent does the New Downloader address this ticket?
comment:8 Changed at 2010-08-14T18:24:05Z by warner
- Resolution set to fixed
- Status changed from new to closed
#798 resolves this ticket. The ShareFinder logic will send out multiple queries at once, so we'll wind up with more shares (and talk to more servers) that we strictly need, but it's limited to a window of 10 outstanding requests, which felt reasonable. (using something like 3*k might be more adaptive). And it will ask for more if it loses servers during the download. But it won't talk to the whole grid, which was the concern about this ticket.
So I'm closing this one.
comment:9 Changed at 2010-08-15T06:14:40Z by zooko
I created #1180 (put more DYHBs into flight at once when K is larger).
#605 (two-hour delay to connect to a grid from Win32, if there are many storage servers unreachable) is blocked on this ticket because I want to see if fixing this ticket (#448) improves (or otherwise changes) the behavior of the #605 issue.