[erlang-questions] Mnesia vs When Databases Lie: Consistency vs Availability in Distributed Systems
Mon Dec 17 14:58:19 CET 2007
That option costs min. one million dollars and the Facebook case is just
lame. You will need 1Gbps link that is redundant and costs min. 40$/Mbps
i.e. 45K$US/month let alone duplicate servers. Globally dis pared system is
ok for doc's but not ok for HA/HPC DB. Also Amazon AWS does not cut well for
"stock exchange" because you are talking about 10< microseconds and AWS is
talking about 10< miliseconds in db transaction time, that is why Google has
Joel Reymont wrote:
> How does Erlang change or improve this situation?
> I can't imagine a fragmented Mnesia table would help here, not when
> one chunk lives in Europe and another in the US.
> I also imagine there would be significant costs in Transatlantic
> replication (in terms of transaction time, not dollars) if a regular
> distributed Mnesia table is used.
> I want to dig deep into Mnesia for the corresponding chapter of my
> book. I plan to try to figure out and write up the distributed commit
> protocol, for example.
> I think I'll also try to set up exactly the scenario that the above
> article describes (two machines, multiple Mnesia nodes) and simulate
> denial of service attacks, tripping over power cords, etc.
> I want to know exactly how much network bandwidth is taken by
> replication among other things and what exactly happens when I bring
> up a Mnesia node that went down. There was a discussion of this
> recently but nothing beats a step by step explanation.
> Thanks, Joel
> erlang-questions mailing list
View this message in context: http://www.nabble.com/Mnesia-vs-When-Databases-Lie%3A-Consistency-vs-Availability-in-Distributed-Systems-tp14336495p14370301.html
Sent from the Erlang Questions mailing list archive at Nabble.com.
More information about the erlang-questions