NoSQL for the Real-Time User Data Problem

dataversity.net

12 points·skempe··8 comments

From the architecture overview; they need 7 GB of memory to index 100 million objects. That's over 75 bytes per object, which is quite a lot.

To be able to guarantee maximum 1 disk access, they would need to store the hash, the address, and the size. You can do that naively by using 160 bits for SHA1, 64 bits for address and 32 bits for size, which is only 32 bytes. Maybe they use the rest for metadata like generation count...

http://citrusleaf.net/_docs/Architecture_Overview.pdf

Does anyone have a TL;DW version of this? This speaker takes an incredibly long time to get to the point, but I am interested in the topic.

This was more of a teaser for a future talk and contains very little data that's relevant from a development / design perspective. The only interesting stats are on a slide at 4:30 where he talks a little about the data model and shows a few benchmarks.

Meh.

Thanks. :)

Are you a human or a bot? The video is a 4m47s lightning talk which is basically a TL;DW for a real talk.

I'm probably a bot.

If you're going to post a video, post a transcript.

Who has time to watch these things?

Why would anyone use this as opposed to the proven open source alternatives?

Additionally, how does this actually actually solve any realtime requirements?

Reply on news.ycombinator.com