WHATIF.txt

What if this service needed to scale to 10,000 URL generation requests per second?

The service has 2 part for URL generation. Generating id and storing
 url and id pair. They corresponds to the redis operation INCR and MSET.

For this prototype, I have used MockJedis as a big hash table. If you change it 
 to real redis, you'll achieve the goal. 
The official benchmark result (http://redis.io/topics/benchmarks) shows it can
 handle 100018 INCR reqs completed in 1.46 seconds and 100007 reqs completed
 in 0.88 seconds on a single machine (Intel(R) Xeon(R) CPU E5520 @ 2.27GHz).
 
If you like more scalable, distributed system can be considered. For id 
generation we can choose Twitter's snowflake model, the composition of:
 timestamp, worker number and sequence number. And Storing is easy. Just scale
 out by the demand of throughput. Sharding or coordination
  (e.g. Hadoop NameNode) can be considered.

How about 100,000 URL resolve requests per second?

The official benchmark shows 100000 GET requests completed in 1.23 seconds.
 So that's close. However, HTTP servers can be distributed in L4 because it's
  stateless service. And stateful service can be sharded by id. There might be
  hotspot. We can think about various caching and replication of those hotspot.