Database Reference
In-Depth Information
First Time P >= 99%
Last Time P < 100%
5000
4500
4000
3500
3000
2500
2000
1500
1000
500
0
Fig. 5.3
Time to see freshness with eventual consistent read
Table 5.2
Successive eventual consistent reads
First read/Second read
Stale
Fresh
Stale
39:94% (189, 926)
21:08% (100, 1949)
Fresh
23:36% (111, 118)
15:63% (74, 337)
be, since each read should always see the most recent value. However, eventual
consistent read is not monotonic and indeed the freshness of a successive operation
seems essentially independent of what was seen before. Thus, eventual consistent
read also does not meet stronger consistency options such as causal consistency.
Table
5.2
shows the probability of observing fresh or stale values in each pair
of successive eventual consistent reads performed during the range from 0 ms
to 450 ms after the time of a write. The table also shows the actual number of
observations out of 475;575 of two subsequent reads performed in this measurement
study. The monotonic read condition is violated, that is the first read returns a fresh
value but the second read returns a stale value, in 23:36% of the pairs. This is
reasonably close to what one would expect of independent operations, since the
probability of seeing a fresh value in the first read is about 33% and the probability
of seeing a stale value in the second read is about 67%. The Pearson correlation
between the outcomes of two successive reads is 0:0281, which is very low, and it
is concluded that eventual consistent reads are independent from each other.
