W H I T E P A P E R
www.persistent.com
© 2017 Persistent Systems Ltd. All rights reserved.
7
This calls out for a strategy based on checking with cloud service providers. Public cloud providers
may report that your specific industry regulation guidelines (e.g., HIPAA for healthcare) have been
implemented. Security and privacy guarantees (e.g., virtual networks, encryption, authentication and
authorization) may also prove to be sufficient. On the other hand, if for whatever reason one of these
two hurdles is not cleared, the remaining possibilities are either a private cloud deployment or a hybrid
cloud, where the more sensitive data is stored on premise or in a private cloud, and the remaining data
is stored in the public cloud, and a virtual network connects the two environments securely.
b.
Performance
and
scalability
– Based on the performance requirements of analytics queries, again
customers might decide to go public, remain private, or even opt for hybrid.
There are certainly more guarantees for predictable query performance for single tenants (this designation
includes both private cloud and public cloud IaaS providers) than on public multi-tenant clouds. While
auto-scaling technology in public clouds may scale resources up or down elastically, noisy neighbors
with heavy spikes in demand can still be an issu
e 2 .Even though there are ways to mitigate this (e.g., by
throttling mechanisms implemented by public cloud providers to prevent heavy tenants from consuming
too much resources), single tenant implementations are more of a fit for mission-critical workloads.
Some private cloud providers are also hardware vendors (e.g., HPE, Dell, IBM which are in the top 5
spots of this market), so customers can buy cloud hardware and services from them. Some also let
customers bring their own hardware. Thus, with private cloud providers, customers have more control to
choose the appliance or scale-up server that is known to suit their needs. On the other hand, technology
running on virtualized hardware offered by both private cloud and public IaaS vendors is generally
horizontally scalable to accommodate growing data volumes and numbers of users consuming analytics
services.
When requirements for query latency are in the sub-second range, transfer times for data from the
cloud to the client and vice versa will be higher than acceptable. In such cases, customers may want
to investigate direct network connectivity services offering more reliability and lower latencies than
typical Internet connections (all major public cloud vendors offer such products for lease or purchase –of
course, this adds to the cost). For queries on datasets demanding very predictable, very fast response
times, leaving these datasets on premise may be the better solution. As with security and compliance,
performance may also be a driver for hybrid clouds.
c.
Pricing models
and
cost
– Private cloud offers large cost benefits when compared to on-premise by
eliminating heavy upfront costs: no hardware must be purchased (necessarily), no data center costs,
and substantial operational costs are saved, as less IT personnel is needed. Nevertheless, there will be
some upfront costs as well as an ongoing cost. Public cloud pricing models can be either subscriptions
for a predefined period or “Pay as you go”, and it will be generally cheaper than ongoing costs of private
clouds. Of course, the cost related to technology used at a given level of service in public clouds is very
relevant; we will examine this in the next section below. Beware, pricing structures are not standardized
and may be complex. Hence, depending on expected timeframe of your analytical solution, whether it
is PoC or production deployment, compute the TCO in both scenarios, and set up trial accounts with
different cloud providers before deciding.
d.
Internal technical skills
– This is also one of the deciding factors: if the IT team does not have the skills
to manage infrastructure, going with public cloud will make more sense. Or they might have the skills,
but some may feel challenged or threatened by this new way of working. However, as with previous
technological shifts, cloud solutions open new opportunities for IT staff, so it may turn out to be a positive
change. Based on this dimension, enterprises can also choose to go with hybrid cloud.
2
In PSL we have experience of an extreme customer case where, because of cost considerations, the architecture that was decided was to allow 40 tenants in an Amazon
Redshift single node cluster. Beyond 5 tenants issuing heavy queries, the system could not scale and overall query performance suffered.