Investigate performance regression in development

[domodwyer]

NOTE: This issue is for the development branch, and does not effect the production release.

---

Commit 860240e (for #116) introduces a performance regression of ~5k/s reads (and probably overall, though mongo is the bottleneck for write operations).

A comparison between 91cf46c (the previous commit) and 860240e shows that the average throughput on the same instance dropped from ~64,000 reads/s to ~59,500 reads/s - this is due to an increase in request latency variance (99 percentile increased from 7ms to 12.5ms).

This might not be an issue, the feature added is pretty useful however a change of ~6% should be looked at - there might be a way to optimise the fast path.

Throughput:

x => 91cf46c3ef1013857ec915403d99a9007f84d06b-select-zipfian-throughput.log
y => 860240ea96bd411b6f6165bb634ac6209981f2c5-select-zipfian-throughput.log

     n   min   max  median  average   stddev      p99
x 3600 52778 70224 63914.0 63945.67 2320.809 68475.48
y 3600 49235 64286 59562.5 59583.09 2117.059 63637.03

       55000                 60000                 65000                  70000
 |-------+---------------------+---------------------+----------------------+--|
                                          +------+-----+
1                        -----------------|      |     |------------------
                                          +------+-----+
                       +-----+-----+
2    ------------------|     |     |---------------
                       +-----+-----+
Legend: 1=data$x, 2=data$y

At 95% probablitiy:
===> average is statistically significant     (p=0.000000, diff ~4362.578056)
===> variance is statistically significant    (p=0.000000)

Latency:

x => 91cf46c3ef1013857ec915403d99a9007f84d06b-select-zipfian-latency.log
y => 860240ea96bd411b6f6165bb634ac6209981f2c5-select-zipfian-latency.log

          n min    max median average   stddev   p99
x 214499384 0.5   7.05      4 3.89542 1.353147  7.05
y 214499384 0.5 351.65      4 4.57718 3.109341 12.50

   0                        5                        10
 |-+------------------------+-------------------------+------------------------|
                  +----+----+
1    -------------|    |    |-----------
                  +----+----+
             +---------+---------------+
2    --------|         |               |----------------------------------
             +---------+---------------+
Legend: 1=data$x, 2=data$y

At 95% probablitiy:
===> average is statistically significant     (p=0.000000, diff ~0.681760)
===> variance is statistically significant    (p=0.000000)

[domodwyer]

See #149