Gavin Andresen - 2010-12-19 17:16:35

As I understand it, every client receives every transaction in the system, so it can check later if transactions are valid.  What happens when the number of transactions gets very large?
When that happens, I think most people using bitcoin will not be running always-connected-to-the-bitcoin-network software.  I think there will be at least three different sets of people:

1. People who trust a web site to keep their wallet safe more than they trust themselves.  They'll use sites like MyBitcoin or MtGox, which give them an online wallet.

2. People who don't or won't trust anybody to keep their wallets safe, but don't have a high-speed, always-on Internet connection.  They will use a 'payment gateway', where they submit transactions that are signed by their own computer (they don't have to trust the payment gateway to keep their wallet, the gateway just forwards relevant transactions onto the bitcoin network for the user).

This doesn't exist yet, but writing the code to support this isn't terribly hard.

3. People who DO have always-on high-speed network connections; they'll run bitcoin just like they do today.  But, assuming the volume
of bitcoin transactions continues to increase faster than the cost of network bandwidth declines, that will become increasingly expensive,
so I think more and more users will choose option 1 or 2.

Quote
Does the client have to keep them all in the memory at once?  If not, how can the client check if somebody actually owns the coin?

The current bitcoin doesn't keep all transactions in memory at once--  the blkindex.dat and blk0001.dat files in your bitcoin data directory stores them on disk.