Question

I need to store IP addresses in a redis hash.

Will there be considerable memory savings if the IP is stored as an integer instead of a string?

I would be using Ruby’s IPAddr to convert the IP to an int.

Answer 1

It depends on how you do it. In Redis keys and (leaf) values are strings. If you would convert an IP address to an int and send it to Redis like the following code you wouldn’t save much:

redis.hset("xyz", "ip", IPAddr.new(ip).to_i)

The IP “255.255.255.255”, for example, is 15 bytes in dotted quad form, its integer representation “4294967295” is ten bytes when saved as a string, which is what the code above will do.

To get down to just four bytes stored in Redis you would have to send the raw bytes “\xFF\xFF\xFF\xFF”.

In Ruby you would do it this way:

packed_ip = IPAddr.new(ip).hton
redis.hset("xyz", "ip", packed_ip)

And then when you read it back

packed_ip = redis.hget("xyz", "ip")
ip = IPAddr.ntop(packed_ip)

What IPAddr.hton and IPAddr.ntop do is this:

packed_ip = ip.split('.').map(&:to_i).pack('C4') # hton
ip = packed_ip.unpack('C4').join('.') # ntop

Then there’s the whole thing about IPv6 and whatnot, but I think IPAddr has you covered there.