TCO and human hours for large deployments

Colin Paddock cpaddock at friendster.com
Wed Sep 19 17:55:41 UTC 2007


We have a somewhat largish mogilefs cluster, with 10 storage nodes and
something like 100 million files, without counting replication, added to
mogile to date (we deployed about two months ago now)

It took two people about a month to deploy, but this also included:

* Design of how we were going to set up the cluster
* Racking the machines needed
* Modifying our applications to use the mogile backend. (This required
about two days of an engineer's time, on top of the two ops people
dedicated to the project.)
* Testing various db/config options with an eye towards performance of
reads/writes. Testing lighty versus perlbal, etc.
* Actually sitting down and learning how mogile works (and we're still a
bit behind on that one :) )

After the deployment, it's been fairly smooth, without too much
maintenance. We've had to spend some time writing plug-ins for
nagios/ganglia, but beyond that, it's been a low-maintenance cluster.
We've also spent a good deal of time creating web-tool front ends for
mogadm, but this was more in our spare time. 

Colin Paddock
www.friendster.com/cpaddock 


-----Original Message-----
From: mogilefs-bounces at lists.danga.com
[mailto:mogilefs-bounces at lists.danga.com] On Behalf Of Lance Reed
Sent: Wednesday, September 19, 2007 4:07 AM
To: mogilefs at lists.danga.com
Subject: TCO and human hours for large deployments

I am working out the Total Cost for a large (er, scalable...) mogilefs 
deployment.  The numbers for hardware are relatively easy depending on 
design.  However, like all opensource initiatives, calculating the time 
/ resources for administration by humans is tricky until you do it for 
awhile.  Our company has plenty of experience using OpenSource tools for

the majority of our needs, with the proprietary solution thrown in where

applicable.  So I was wondering if anyone who is attempting to setup a 
mogilefs storage setup, or who is using one now could shed some light on

the subject.  I figured initially, on a new deployment, about 60% of a 
single Administrators time would be taken up to maintain a large 
mogilefs cluster.  I figure design and implementation would take much 
more than that (maybe 2 or 3 people to save time).  I am comparing this 
to a couple proprietary solutions we have been looking at that are known

to scale well into multi TB or even PB setup that are basically Scalable

NAS solutions that use a "brick" mentality.  The theory being that you 
pay upfront for stability or support if needed.  But I am wondering if 
say 60% for the first year of learning how to maintain and grow a 
production mogilefs is close or not.  Too high / low?

Any thoughts or real work data would be great.

Thanks all!!



More information about the mogilefs mailing list