Interesting Links for 20-09-2015

[andrewducker]

• What happens when you get your period in space?
  • (tags: space menstruation women history )
• Doctor Who companions - a data story
  • (tags: data drwho )
• Lady Gaga's anti-rape video
  • (tags: music rape viaBartCalendar )
• How you tell children they have HIV
  • (tags: hiv children )
• Facebook will now be able to show you ads based on the porn you watch
  • Those of you

Comments

[steer]

Incidentally, a quick look down this:
http://quantumfrontiers.com/2013/10/23/the-10-biggest-breakthroughs-in-physics-over-the-past-25-years-according-to-us/

I think the question here is which of these will actually lead to fundamental changes. Two of them relate to quantum computing which could be practical relatively soon. I don't actually know about topological insulators, sounds potentially useful. Then there are some revisions to standard model of physics -- when I did my physics degree the neutrino mass question was still very much open (indeed I did a long essay on the subject) now it's fixed and with the unexpected answer. These could lead to new breakthroughs over the next 25 years.

Oh, I missed photonic computing off my list of potential revolutions. Some of the guys are UCL (and elsewhere in the world of course) are directly pumping data between the memory of different machines using modulated light beams with nothing going up and down the networking stack (would bloody put all my TCP/IP knowledge out of the window). That's pretty radical.

[andrewducker]

Is DMA between machines sensible? Definitely fascinating.

And yes, lots of fundamental stuff going on. Which raises the question of what date you count things from (as you mentioned above). And I think the original article definitely cheats there in where it allocates items to.

[steer]

In essence the entire networking stack is a way of getting bits from the memory of my computer to the memory of your computer -- it's just very convoluted -- packetise the bits, encapsulate in a later 4 (TCP) header a layer 3 (IP) header a layer 2 (MAC) header then on receipt strip the headers and join the packets. In essence, however, something leaves my TCP buffer and arrives in the same form and the same order in yours... so a standard networking stack already takes data from my computer's memory and puts it in your computer's memory.

Obviously you can put firewalls in between to filter out undesirable stuff -- but then just switch the photonic model to be DMA to the firewall that filters and does DMA to your machine if it is OK with what is sent.

Obviously, its undesirable to simply trust whatever lands in your TCP buffer -- but there's already a number of ways that computers deal with the problem of untrusted and potentially malicious information incoming to your machine.

[andrewducker]

True, yes.

In which case, how does the photonic method make TCP/IP irrelevant? Do we not need to route light to the correct computer?

[steer]

To be honest, I don't know the answer to that. I think the experiments I'm thinking of link only two computers but I don't know enough to answer beyond that.

There is optical switching but it's weird and unrelated (still using standard networking stacks with layers on top).

[andrewducker]

Gotcha.

I think that's where I got confused - if we were doing without a network stack at all, I assumed we'd be piping stuff directly into memory locations in the other PC, which is clearly very dangerous. If you find out more, and can explain it in words I'd understand, I'd be interested.

[steer]

To be honest, it was probably mistaken of me to introduce the idea starting with transfer between computers -- that's the bit I'd been to a presentation on. In general the idea is replacing electronic components with photonic ones and that is already practical and working for many uses. I'll keep my eye out for a readable up-to-date "where are we with photonic computers" guide.

[fanf]

RDMA is used for iSCSI.

I wonder how to do something like that when the other endpoint is not trusted...