Question

I’m using the following code to discard unsupported physical interfaces / subinterfaces from routers that connects to a big ISP network (by big I mean tens of thousands of routers):

private final static Pattern INTERFACES_TO_FILTER = 
   Pattern.compile("unrouted VLAN|GigabitEthernet.+-mpls layer|FastEthernet.+-802\\.1Q vLAN subif"); 

// Simplification
List<String> interfaces;
// lots of irrelevant code to query the routers 

for (String intf : interfaces) {
   if (INTERFACES_TO_FILTER.matcher(intf).find()) {
      // code to prevent the interface from being used
   } 
}

The idea is discarding entries such as:

• unrouted VLAN 2000 for GigabitEthernet2/11.2000
• GigabitEthernet1/2-mpls layer
• FastEthernet6/0/3.2000-802.1Q vLAN subif

This code is hit often enough (several times per minute) over huge sets of interfaces (some routers have 50k+ subintefaces), cache doesn’t really help much either because new subinterfaces are being configured / discarded very often. The plan is to optimize the regex so that the procedure completes a tad faster (every nanosecond counts). Can you guys enlighten me?

Note: mpls layer and 802.1Q are supported for other kinds of interfaces, unrouted VLANs isn’t.

Answer 1

I’m answering my own question for further reference, although the credits goes to @piotrekkr since he was the one that pointed the way. Also my Kudos to @JB and @ratchet. I ended up using matches(), and the logic using indexOf and several contains was almost as fast (that’s news to me, I always assumed that a single regex would be faster than several calls to contains).

Here’s a solution that is several times faster (according to the profiler, about 7 times less time is spent at Matcher class methods):

^(?:unrouted VLAN.++|GigabitEthernet.+?-mpls layer|FastEthernet.+?-802\\.1Q vLAN subif)$