Database Reference
In-Depth Information
(a) 1st Query Performance
(b) 1st Query CPU Load
Fig. 5.
1st Experiment results for the system's LB configuration
Concerning the CPU load, we certainly see some improvement, as CPU load
initially drops to 95 % and then to almost 92 %. However, this means that still
the VMs exploited reach high CPU load values and there is a risk for VMs health
deterioration. Unfortunately, the budget restrictions do not allow us to scale out
for reserving more resources. Furthermore, this is a sign that the query issued
is quite complex and even if the load on a particular VM is reduced, the CPU
processing load is not reduced in a high degree as Virtuoso seems to attempt
to exploit as much resources as possible in order to address the reduced load
incurred. In addition, the current number of users that exploit the respective
application does not mandate for actually exploiting such additional resources.
Actually, with this query, we needed to indicate the worst possible scenario
where a big number of concurrent users issue the most complex query to the
LMS system and to highlight the cost-effectiveness as well as the scalability of
the system in terms of query processing time. The results obtained indicate that
our goals have been achieved for both queries and thus for any other query
having a complexity in between the complexity exhibited by these two queries.
To summarize, we can clearly see that in both experiments, the scale-out
activities occur as soon as possible. In addition, the high limit on instance number
to be reserved is indeed the most appropriate as the less complex query obtains
both a great reduction in CPU load, which does not require reserving additional
resources, and a query time speed-up while for the second query there is a
significant speed-up but due to its high complexity, the CPU load is not highly
decreased, which mandates for spoiling cost-effectiveness if additional resources
are reserved. The low limit has been posed based on the current application
usage which does not go over more than ten concurrent users. If this changes
in the near future, then we will certainly attempt to raise this limit to better
handle the new and increased query trac.
Search WWH ::
Custom Search