Database Reference
In-Depth Information
bin/ycsb run cassandra-10 -P /home/vivek/source/
book_source/ycsb/casssandra-ycsb.properties -threads 10
With YCSB there are two modes of running load and run. With load it will store
data into backend and throughput and latency would be measured. Whereas with run
mode we can tweak it for heavy read or read/update proportions (see
Cassandra-
ycsb.properties
).
YCSB comes with two types of distribution: zipfian and uniform. Continuous uni-
form distribution would refer to symmetric distribution of data for each column at each
interval. Whereas zipfian distribution is based on power law distribution where load
and throughput are inversely proportional to each other. For more details on zipfian
Summary
As we reach the end of this chapter, here are a few key performance checks to make if
an application's throughput is consistently degraded over time:
•
Monitor GC and examine the Cassandra server logs for the
GCIn-
spector
class.
•
If data nodes are running out of space, add new nodes. We will discuss
this further in the next chapter.
•
Monitor
bloom_filter_fp_chance
and cache configuration.
In the next chapter, we will cover more about Cassandra administration and monit-
oring.