How much of our current resources are we consuming?
Is there reason to believe we'll outgrow them in... two years? Four? Six?
Do we have a need for a third server? A fourth?
What needs would those be, what benefits would they provide? What costs
would that impose? Would they need to be full featured servers like our
current servers, or would smaller machines be adequate?
What about alternatives like the cloud computing solutions someone recently
posted somewhere?
Also, a few software questions, such as what tools or analytics do we need
that will help provide useful information to sysadmins? And/or to our
development team?
Actually, it would be a really good idea to list each server with its RAM,
processor, and storage. Using top should give us current utilization, but I
believe we’re also running Nagios for alerting, so it’s probably monitoring
processor, RAM, and storage.
Andrés, can you speak to the current specs…I’m not nearly as familiar with
these systems as our previous. Are we running on vSphere or some other
hypervisor?
Ideally, we want to build a highly available system that uses fault
tolerance and redundancy. Let’s start with RAM, if these were physical
systems and not VMs, then we would have the options of hot spare RAM.
Regardless, I’ve never found it to be cost-effective (you basically have RAM
installed that you can’t use unless there’s a failure), so I’d recommend
against this were it possible. Storage would be next…I’m willing to bet that
we’re on RAID 5 storage, probably in a SAN environment (please correct me if
I’m wrong) which gives us quite a bit of fault tolerance and redundancy when
it comes to storage. Plus, we are also performing regular backups as part of
a disaster recovery plan. I’m going to ignore processors (for physical
servers, hot spare procs are like hot spare RAM…a waste of resources, in my
opinion) since we’re in a virtual environment and jump up to the servers
themselves.
We are running 64-Bit CentOS (Linux) on all servers. We have a front-end web
server which hosts the OI (application) and a back-end database server
running MySQL. We also have a development server (I believe) that utilizes
the same database server, but separate, development databases. In building
for the future, a part of me would like completely separate dev and
production environments (add a separate MySQL server for development only),
but another part of me would love to see us cluster both the front-end and
the back-end MySQL database server for high-availability and redundancy, but
that will, of course, come with a price. Is it worth it? Honestly, probably
not at this level of utilization, but that’s really a Board decision based
on the Board’s goals (5 9s uptime, or some measurable objective).
Bandwidth/networking should probably also be brought up here. Once again,
virtualization renders multiple NICs and creating adapter bonds unnecessary
as the underlying hardware will already have redundancy in this area. We do
need to double-check our bandwidth utilization stats, though.
Before I forget, it would probably be a really good idea to document the
current configurations (hardware/software/OS) for the system administrators
group. In addition, disaster recovery procedures should also be
included…maybe the files section of the system administrators list?
As to growth, I’m on a complete hardware refresh cycle of 3-5 years based on
performance needs at the day job. The really cool thing about virtualization
is that hardware swaps can be done WITHOUT impacting running VMs, so no
downtime. And, we won’t be doing them, our hosting provider will handle
hardware upgrades.
Storage requirements will only grow…that’s what they do. :-) However,
storage is cheap these days, and I believe that our current utilization is
well below our allocation.
Now, we need specifics… Andrés, can you get us some stats and our
configuration when you have a chance?
Lionel, I realize I didn’t answer your questions, but I’m hoping that the
above informs our thinking when we do formulate our answers.
Will
______________________________
Will Allred <wal...@gmail.com>
http://willallred.com/