Informorph Travel by Laser

[jeff_wilson]

I said,

Originally Posted by jeff_wilson

Near-lightspeed is achieveable if you are willing to risk both creating a xox and fatal corruption. Proper due diligence will require a three phase commit (original transmission, destination acknowledgment and verification request, origin acknowledgment and verification confirm) that takes a minimum of 3x lightspeed time plus bandwidth usage, and the bandwidth usage needs to be multiply redundant if you want to avoid repeating phases 1 and 2 to correct transmission errors: I recommend 10x for calculation simplicity and complete reliability on all but the worst quality links.

and then Anthony said,

Originally Posted by Anthony

Both can be avoided without a reduction in speed. Just use enough bandwidth for a error rate of less than 1/10^12 or so, and have the sender always shut down their copy unless told that transmission failed (the Pluto receiver still wants to send an ack, but there's no need to wait for a response before activating the infomorph). This does use more bandwidth than just sending packets with checksums and resending any packets that have errors, but only by a factor of 2-3.

I don't see how activating the infomorph without the confirming handshake avoids xox liability.

[Anthony]

Originally Posted by jeff_wilson

I don't see how activating the infomorph without the confirming handshake avoids xox liability.

You only have xox liability if both copies are active at once; otherwise it's just a backup. You simply shut down the origin infomorph before transmission, and don't reactivate unless a situation comes up which would allow permit activating a backup (for example, a message that copying failed).

[ClayDowling]

Transmission isn't instantaneous. Would it make more sense to have the infomorph remain active on the origin machine until after active transfer has occurred, and perform the shutdown after intact transmission has been verified? Presumably there would be some protocol for managing this. After all, until the transmission is complete, at the receiving end it isn't an infomorph, but a collection of data waiting to become one.

[Flyndaran]

What's sensible and what's prescribed to avoid multiple active copies may not be the same.
Maybe how much verification / error protection is required depends on how much time you book on said transmitter.
"Oh, you booked the minimal - cross your digital fingers and hope special."

[johndallman]

Originally Posted by ClayDowling

Would it make more sense to have the infomorph remain active on the origin machine until after active transfer has occurred, and perform the shutdown after intact transmission has been verified? Presumably there would be some protocol for managing this. After all, until the transmission is complete, at the receiving end it isn't an infomorph, but a collection of data waiting to become one.

That depends on how all those terabytes of data are organised and modified. Does a running infomorph constantly make modifications all over its data as part of integrating stuff into its memories? If so, it makes sense to shut it down before starting to transmit. If most of it is read-only, then that can be sent first, and then you shut down before sending the part that gets modified as it experiences. THS does not go into detail on this, and the in-setting law on the subject might suggest playing safe, or be out of date.

[vicky_molokh]

Originally Posted by johndallman

That depends on how all those terabytes of data are organised and modified. Does a running infomorph constantly make modifications all over its data as part of integrating stuff into its memories? If so, it makes sense to shut it down before starting to transmit. If most of it is read-only, then that can be sent first, and then you shut down before sending the part that gets modified as it experiences. THS does not go into detail on this, and the in-setting law on the subject might suggest playing safe, or be out of date.

On one hand, I suspect that an infomorph does change content throughout its 'digital body'.

On the other, in cases where the travel occurs not for the first time, and thus there is an out-of-date copy at the destination (with a known version number!), it might be workable to send merely the Patch Data / Difference File.

Then again, I suspect that 'Solar System Travel Channels' have so much bandwidth that file size doesn't matter.

[johndallman]

Originally Posted by vicky_molokh

On one hand, I suspect that an infomorph does change content throughout its 'digital body'.

That's my instinctive feeling, but I feel I may be biased by being used to working with a kind of software - precise 3D space modelling - which really does work that way. Industrial-scale databases that are built to work fast with rotating storage look as if they work that way, but actually don't.

What it would actually depend on is details of the implementation of infomorphs which might be quite different in different manufacturers' products. But if you dig into that, you find yourself asking how it is that all THS computers seem to run the same kind of software (there are several ways of achieving that at TL8, and I would expect more at TL10) and how it is that the software all works so well (TL10!) and then an infinity of further questions.

[vicky_molokh]

Originally Posted by johndallman

That's my instinctive feeling, but I feel I may be biased by being used to working with a kind of software - precise 3D space modelling - which really does work that way. Industrial-scale databases that are built to work fast with rotating storage look as if they work that way, but actually don't.

I suppose one could split the ghost into the physical structure and linkage data, and the state of it. No idea how feasible.

Originally Posted by johndallman

What it would actually depend on is details of the implementation of infomorphs which might be quite different in different manufacturers' products. But if you dig into that, you find yourself asking how it is that all THS computers seem to run the same kind of software (there are several ways of achieving that at TL8, and I would expect more at TL10) and how it is that the software all works so well (TL10!) and then an infinity of further questions.

Hmm. Cross-platform managed code and all that?

[johndallman]

Originally Posted by vicky_molokh

Hmm. Cross-platform managed code and all that?

Possibilities include, but are not limited to: A universal architecture, used for everything; all software distributed in a semi-compiled form and JIT compiled; all software built for an efficient virtual machine implemented in software; binary cross-compilers, and combinations of all of these. It's TL10 and fictional, as well as being a gamable abstraction, so we can't expect to figure out how it "actually works".

A friend ran his homebrew "not-Transhuman Space" setting a few years ago, in which Microsoft had attained a system-wide monopoly of operating systems and development tools, and you had to have a license from them to learn to write software. Yes, this was black comedy. It also had implanted computers that didn't run off bioelectricity. They had really high-density batteries instead, which meant safety-minded people objected to implanting explosives which weren't all that insensitive.

[jeff_wilson]

Originally Posted by johndallman

That's my instinctive feeling, but I feel I may be biased by being used to working with a kind of software - precise 3D space modelling - which really does work that way.

This should be correct, as established in conversation with Phil on the incompressible nature of TS software, shortly before his accession. Phil put it down to null structures at the time, but real machine learning requires the densest possible information sets and widest spread updates of learned values for greatest performance.

Originally Posted by johndallman

What it would actually depend on is details of the implementation of infomorphs which might be quite different in different manufacturers' products. But if you dig into that, you find yourself asking how it is that all THS computers seem to run the same kind of software (there are several ways of achieving that at TL8, and I would expect more at TL10) and how it is that the software all works so well (TL10!) and then an infinity of further questions.

We are allowed to presume that they work in similar ways as today, but mostly ported to optical switching, and with more layers.