[UPHPU] Amazon EC2 / High availability hosting

Joseph Scott joseph at josephscott.org
Fri Feb 5 09:06:03 MST 2010

On Fri, Feb 5, 2010 at 8:06 AM, Jordan Schatz <white.armor at gmail.com> wrote:
> Thanks for the info everyone, I am thinking strongly that we will use
> EC2. I am not sure that I understand EC2, it seems a little to good to be
> true. If I understand it correctly I don't need to set up a RAID, or
> consciously duplicate my data, because the virtualized EBS (elastic block
> storage) is already redundant, and I don't need to have a failover
> server, since the virtualized EC2 instance will persist even if the
> hardware fails.

Unless you are booting the EC2 instance from EBS your instance will
not persist in the event of a hardware failure.  If the hardware dies,
you'll end up starting an instance on a new piece of hardware, that
won't have access to your previous instance.

Keep in mind that you'll still need to make backups, even with using
EBS.  It can still fail, though hopefully at a lower rate (that's what
they report) than just a single drive.

> Is that your experience?
> Does all the virtualization great high latency? or other performance
> problem?

Keep in mind that this is still at it's core shared hosting.  You have
multiple instances using the same hardware.  This is not the same as
getting a dedicated server from theplanet.com or cari.net
layeredtech.com.  Now admittedly this is much more advanced shared
hosting (using Xen to run instances), but there aren't a lot of
details on disk and network I/O, so there's no way to know when you
start an instance exactly what kind of performance you'll get.  CPU
and memory are easier to virtualize and are less likely to be an
issue, but I haven't seen any numbers so we don't know if they over
subscribe the CPU or not.

> It just sounds to easy...

After all that I'm not saying don't use EC2 (or AWS in general), just
that it's good to look at what you are actually getting and not
getting.  If it does a good job of solving your problem then use it,
if not go some where else.

Another thing to keep in mind with EC2 (and AWS in general) is that
for applications that require a lot of bandwidth it's generally
cheaper to go some where else.  Again that's a general rule and may
not apply depending on exactly what your bandwidth demands are.

Joseph Scott
joseph at josephscott.org

More information about the UPHPU mailing list