1 Trillion Web Pages Archived | Hacker News Reader

1 Trillion Web Pages Archived

blog.archive.org

703 pointspabs394 comments

Something I wish we could have is some kind of peer mirror of archive.org. The main IA web application gets angry pretty quickly if you're trying to click through a few different dates. If there were some kind of way to slowly mirror (torrent-style) and offer pages as a peer from archive.org that would be neat. It would be cool to show up as an alternative source for the data and the archive.org app could fetch it out of there on a user's choice and validate the checksum if required.

In the end, I've ended up just keeping my own ArchiveBox and it's an all right experience. In the end, it's only useful for things I know I wanted to archive. For almost everything I go to the IA - which has so much.

Is there such thing as "versioned" torrents? Assuming you have the right PGP key you could mix bittorrent and packaging systems to get an update-able distribution

A Torrent would probably suffocate under the small file distribution. I’m not sure how the romset torrents work but I thought they were versioned.

But torrent is probably the wrong tech. I’m sure there would be many players willing to host a few TB or more each, which could be fronted via something so it’s transparent to the user.

But a better option might be a subscription model, anything else will be slammed by crawlers.

Couple of BEPs related to updating torrents:

https://www.bittorrent.org/beps/bep_0039.html https://www.bittorrent.org/beps/bep_0046.html

trere is the bittorrent v2 standard: https://blog.libtorrent.org/2020/09/bittorrent-v2/

but unfortunately most foss torrent clients do not support it, partly because at release libtorrent 2.0.x had poor io performance in some cases so torrent clients reverted to the 1.2.x branch

I think SciOp is doing something in that area, with a catalog site and webseeds. https://sciop.net/

- I can confirm that the web archive can be really slow

- I think I have seen that AI scrapers create bottleneck in the bandwidth

- To some digital archives you need to create scientific accounts (I think Common Crawl works like that)

- Data quite easily can be very big. The goal is to store many things. We not only store Internet, but with additional dimension of time

- Since there is a lot of data, it is difficult to navigate it, search it, so it easily can become unusable

- For example that is why I created my own meta data link, I needed some information about domains

Link:

https://github.com/rumca-js/Internet-Places-Database

The Archive Team - not part of the Internet Archive - worked on a distributed backup of a portion of the Internet Archive - https://wiki.archiveteam.org/index.php/INTERNETARCHIVE.BAK

It's been dormant / on hiatus for a few years now.

That can only cover other collections though, because the WARC files from the Wayback Machine web scrapes are not public.

I do wonder why IA does not maintain a IPFS instance, or if they do, why they're not more popular? There's tons of IPFS mirror services out there that operate at reasonable speeds. One issue I've run into with IA is old enough websites that there's JS or CSS that just wont render, what I'm not sure about is, can we retroactively fix such things? Would be nice to be able to un-ruin the code somehow if they exported everything possible at the time.

Edit:

Would be really neat if you could click on a domain while on IA, and a desktop client downloads as many WAR files in a slower priority download queue, as many as you're interested in, with higher priority pages first, and then you can view it fully offline.

Because nobody pins on IPFS. It's basically http with extra steps, at this point.

They do torrents. I was looking into this recently as well, considering building an Activity Pub alternative to IA. I came to what I assume is the same conclusion that IA came to.

No one uses IPFS. For the average user, it is significantly more difficult to get started. For the experienced user, the ecosystem of tools around IPFS is extremely small.

All in all, IPFS offers very little benefit over torrents in practice and has a much smaller user pool.

IPFS is a great idea poorly executed. Content addressable storage is a great idea, but it is so difficult to use in practice for real world scaled scenarios (larger than one hard disk drive).

The problems with the torrents is that they can be updated if the file changes (sometimes small metadata changes) and now your seeders can't be found. Maybe if they also kept a list of old hashes so that you could at least manually try to recover data from the older torrent?

This is outdated information. These issues have been solved by various BitTorrent Enhancement Proposals. You do create a new torrent, but you distribute it in a way that to a swarm member is functionally equivalent to updating an old torrent. Check out BEP-0039 and BEP-0046 which respectively cover the HTTP and DHT mechanisms for updating torrents:

https://www.bittorrent.org/beps/bep_0039.html

https://www.bittorrent.org/beps/bep_0046.html

If that updated torrent is a BEP-0052 (v2) torrent it will hash per-file, and so the updated v2 torrent will have identical hashes for files which aren't changed: https://www.bittorrent.org/beps/bep_0052.html

This combines with BEP-0038 so the updated torrent can refer to the infohash of the older torrents with which it shares files, so if you already have an old one you only have to download files that have changed: https://www.bittorrent.org/beps/bep_0038.html

Have any of these even started to be implemented in any client/library? It's been years.

I spent a bit of time trying to find it just now but I swear I read a super long blog or comment or something by someone at archive.org where they concluded essentially that IPFS just "isn't ready" or wasn't feasible for their needs because it's super slow and they didn't see how that couldn't be the case when they consider the volume of transactions they need to do (they didn't see an optimization path).

I wish I could find that article!

edit: https://github.com/internetarchive/dweb-archive/blob/master/...

I have a design for a system where you can "donate" your disk space to a provider. Basically, you run the client, you say you want to make 1TB available to archive.org, and their server can push the rarest content to your computer.

It's based on torrents, and you can easily make a content delivery system on top of this (so people can fetch data from this network).

I emailed a few archiving teams but nobody seemed interested, so I never made it.

It's a hard problem to solve, because its easy to temporarily donate resources to archiving ops via the ArchiveTeam warrior, but a long term commitment to run persistent compute and storage to mirror a chunk of the internet archive. It's why I think Filecoin isn't going to work either; very little overlap between the people who feel its important to keep these archives alive versus people who would run distributed storage to collect financial compensation for doing so.

Easier to send fiat to IA for them to invest (~$2/GB) and to pay to keep the disks spinning somewhere safe across the world.

(ia volunteer, no affiliation otherwise)

The system I have in mind is strictly volunteer-run, and it automatically balances the files so that it minimises rare copies.

You're right, though, long-term commitment is rare from volunteers. That's why the idea is to make short-term commitment so easy that you have a good enough pool of short-termers that it works out in the aggregate.

Appreciate your work on this.

Eh I didn't really do any work, it's just a design right now, but I think it's a nice one. If any archive team wants to work with me on this, I'd be happy to make it a reality so we have a nice FOSS system for distributed, volunteer-led backups.

I suggest emailing textfiles, he'll know who to connect you with in ArchiveTeam, and if there is an opportunity to connect with the decentralized web folks at ia. Strongly believe your architecture is superior to filecoin and IPFS due to relying on torrent primitives.

(ia source of truth, storage system of last resort -> item index -> torrent index -> global torrent swarm)

Thanks, I will!

I emailed him but haven't received a reply. In case you were curious for a bit more detail, here's a short design doc I wrote:

https://gist.github.com/skorokithakis/68984ef699437c5129660d...

Anna's Archive has this system. This also sounds like Freenet.

Freenet has a bunch of encryption, which is out of scope for this. What does Anna's Archive have, besides torrents?

I'm a bit confused. Isn't this such a system where people can volunteer disk space?

https://annas-archive.org/torrents

I think I'm misunderstanding you.

My system is more "I want to donate X GB" and it handles everything, filling that space up, getting the rarest torrents, getting updates, etc. Think of it as a central server managing a globally-distributed, unreliable JBOD in a "push" manner, rather than just downloading a torrent and being done.

Sounds a bit like Wuala https://www.youtube.com/watch?v=3xKZ4KGkQY8

Hmm, maybe, I don't remember exactly how it worked. I'll watch the video, thanks!

Yeah, I did a scraping project a while back where I wanted to look back at historical snapshots. Getting the info out of Internet Archive was surprisingly difficult. I ended up using https://pypi.org/project/pywaybackup/, which helped quite a bit.

That kinda sounds like ipfs

https://ipfs.tech/

Hi, I run the datacenter/infrastructure team at the Internet Archive! We would love to see you at our various events this fall but if paying for the ticket is difficult for you, please email me (in bio) and we'll get you in (if possible).

What events are we talking about here?

Probably these: https://blog.archive.org/events/

I would love to work for IA but openings are rare

If you are in Europe, consider Software Heritage (similar to IA but for source code) too:

https://www.softwareheritage.org/jobs/

Internet Archive now have a presence in Amsterdam

Hey, Q., so what's the size of the internet archive?

it is large enough that I am wondering if the data captured by the actual physical magnetic charges has a heft, that a person could feel. obviously the hardware would fill a house or something, but at what point does the worlds data become a discernable physical reality, at least in theory

I'm betting exabyte or close maybe

For the purposes of ballpark, between 150-200 petabytes of unique data, probably on the lowish end of that last I checked.

Are they distributed events all around the world of just in wherever the team is gathered (San Francisco I guess?)

By the way, thank you all the teams in IA, what you provide is such an important thing for humanity.

would love technical details around this feat. ex: how you even crawl to begin with, storage, etc

We all know the NSA has access to servers hosted in the U.S. How are you protecting the archive from malicious tampering? Are you using any form of immutable storage? Is it post-quantum secure?

Why would they do that? Have you previously seen a case where they "maliciously tampered" with anyone's website?

I just question the integrity and immutability of the data IA is archiving, that's all

You want to know why they'd tamper data?

https://seclab.cs.washington.edu/2017/10/30/rewriting-histor...

https://blog.archive.org/2018/04/24/addressing-recent-claims...

NSA already paid to back-door RSA, got caught shiping pre-hacked routers, can rewrite pages mid-flight with QUANTUM, penetrate and siphon data from remote infected machines.. what else could they do?

https://www.amnesty.org/en/latest/news/2022/09/myanmar-faceb...

IA themselves could tamper with the data, no? It was never meant to be an official historical snapshot to be pulled up for any serious or official purposes. Although it has been used that way for high profile internet drama. It's just a matter of time (maybe during an election) before it's surreptitiously altered and referenced for nefarious purposes.

Most of all, i'm curious about how you reliably and securely store or host so many archived pages. Would you mind briefly explaining such a huge undertaking? Also, total congratulations on the fantastic achievement of this. You guys are my go-to for so much information.

Edit: And how many terabytes it all amounts to.

Thanks for helping to run my favorite library on earth.

If anyone wants to help feed in more stuff, ArchiveTeam is a related volunteer group that sends data to IA:

https://archiveteam.org/

Presumably there needs to be some human to decide something is worth archiving to stop someone just using it as a free way to store all their holiday snaps?

ArchiveTeam members are the ones with access to start crawls of websites, everyone can request they start a crawl, usually they ask for a reason for the crawl, and most reasons mean a crawl will happen.

1 trillion web pages archived is quite an achievement. But...there's no way to search them? You have to know what url your want to pull from the archive, which reduces the usefulness of the service. I'd like to search through all those trillion pages for, say, the name of an artist, or for a filename, or for image content.

That would be hell to index

If it was a commercial problem, e.g. from Google, it would be solved.

The reality is that many things don't exist simply because someone isn't paid to do it.

How much AI companies have benefited by leeching off of IA and Common Crawl, it's a shame there's no at least some money flowing back in.

I imagine it would be no different than current indexing strategies with a temporal aspect baked in... it would act almost like a different site, and maybe roll up the results after the fact by domain

I use GPT web search, and I ask it usually to find textbooks from IA. It works really well for textbooks, but not sure about web pages.

Consider the privacy implications of that. It would effectively create a parallel web where `robots.txt` counts for nothing and where it becomes - retroactively - impossible to delete one's site. Yes, there's ultimately no way to prevent it happening, given that the data is public. But to make the existing IA searchable is IMO just a terrible idea.

Actually, I believe the IA respects robots.txt retroactively, eg. putting something on the disallow list now removes the same page scrapes from a yeaer ago from public access in teh Wayback Machine, but I'd love to be corrected on that.

It may do. I remember looking into it and not getting a definitive answer. The issue here is that taking a site offline has surely been widely understood as the ultimate robots.txt `Disallow` instruction to search engines. IMO we should respect that.

IIRC the IA no longer cares about robots.txt after it kept getting abused [1] to take down older pages. You can still request to take down pages, but it needs a form and a reason. [2]

(Remember, robots.txt is not a privacy measure, it's supposed to be something that prevents crawlers from getting stuck in tar pits!)

[1] https://blog.archive.org/2017/04/17/robots-txt-meant-for-sea...

[2] https://help.archive.org/help/how-do-i-request-to-remove-som...

Useful to know. My more general position, which apparently is not much shared here, is that removing one's site from the internet has historically meant that the site stops being accessible, stops being indexed, and stops being findable with a simple search. If, going forward, we're going to revise that norm, IMO it would be polite at least to respect it retroactively.

That seems in conflict with the idea that once something's been released, it can't ever truly be unreleased.

(Also, consider that when you forbid such functionality, the only thing that happens is that its development becomes private. It's like DRM: it only hurts legitimate customers.)

I remember this functionality existing on Kagi or something. But I can't find it.