Last week, I looked at the various types of virtualisation:
Hardware Virtualisation VMware), Para-Virtualisation (Xen) and
OS-Virtualisation (Linux Vserver) with the various merits and vices of each. After
looking at what virtualisation actually is, I think its important to look at
why one would want to virtualise and consolidate multiple hardware platformsinto one.
So why virtualise? Because everyone else seems to be doing
it? I hope not! Most server systems are under-utilised; general estimates
say that an average Windows server is running at around 15% utilisation and
UNIX environments at about 20-30%. Thats pretty low. In order to gain a more
personal analysis, I ran a small audit of six servers. All run Linux and host
various high-demand services such as mail serving, mail routing/scanning, file
serving, web services, and database hosting. The results were interesting. Two
servers showed intense usage--one server had constant usage of 90-98% (may be
time to upgrade this one) and the other fluctuated with the norm being 10% but
regular peaks of 65% were observed. A third server showed an average usage of
10% with irregular peaks to 20% utilisation; the remaining three systems were
showing very low utilisation of 0-5%. These last four servers could be good candidates
for virtualisation. Why? Well, with combined peak utilisation of 35% and an
average figure of just 20% they can happily run on one physical platformwithout encountering performance problems.
Other than increased
utilisation of resources, what benefits would be gained migrating from these
utilisation of resources, what benefits would be gained migrating from thesefour systems on to one physical server?
Saving physical space
is the most obvious answeras a rule, our server rooms dont grow along in sync
with demands from the business. Unless the entire company is considering an
office move, IT departments will have to work within their given space despite
increases in demand. Its obvious that replacing four 2U servers with one more
powerful 2U server is going to free up 6Us of space, although this is
optimistic, and realistically 4Us would be freed (Ill explain the grounds for
this later). This is still a huge saving in real estate and will enable twice
the number of services to be hosted in the same physical space. Many firms
turning to virtualisation will look at moving to blade servers at the same timeto maximise space savings.
Reduced hardware costs are another advantage of virtualisation. In the
example of my four under-utilised systems, all could be migrated to one server
of the same specifications and still have adequate resourcesthats at 25% of
the current hardware cost. Even if I were to over-spec the new system to allowfor future increases in usage, the savings are not to be ignored.
Reduced power consumption and need for cooling are benefits which come hand in hand. While the power
consumption and heat output of a system with high levels of utilisation will be
greater than that of a system under a lesser load, the consolidation of
multiple low-load systems should still produce less heat and demand less power
over all. Data centres are finding it increasingly more difficult to keep up
with demand for power at the rack and the cooling demand which comes with
increased power consumption (and that additional cooling also requires power,increasing overall running costs).
The ability to rapidly deploy a new system without ordering new hardware, building/installing the server, and updating
firmware can be a big time saver for sys admins (whose time is usually at apremium).
Although the above does not give an in-depth explanation of
the many advantages afforded by virtualisation, I hope it gives an informative
overview of benefits when compared to the more traditional server farm model. There
is an abundance of information to be found on both TechRepublic and Google,
this briefing should give those interested in the topic a starting point for
further research. I was hoping to also discuss the potential weaknesses of
virtualised services and the conceivable actions which can help to neutralise
these risks (and in some cases make virtualisation a more solid and lower-cost
approach)however, I seem to have run short of time so will look at this nextweek rather than hastily fumbling over the subject now.
If you have recent experience of consolidating
services with virtualisation, please share your experiences with us all. Were
problems encountered during planning/migration? Has the new virtualised model
realised the benefits expected beforehand?