[nSLUG] Xen and Heartbeat
jim at on-site.ns.ca
Wed Jun 3 08:53:39 ADT 2009
Miles Thompson was looking at some options for his hosting. I wonder
what his pipe size is. Today any server worth its' salt can easily fill
a 100 Mb line with web server or other data.
A decent quad core Xeon based server with 12-32 Gb of RAM should be able
to handle all of his needs in a single box. Then use the clustering
software to have an online clustered system for redundancy. Functionally
mirror the servers. One dies the other keeps going. You can have
multiple network cards to accommodate your unique IPs or use a single
card for multiple IPs.
Depending on the mail traffic volume, that is in versus out, and how it
is picked up or delivered, you should have no problem running all the
tasks on a single server. You may want to virualize a server for the
mail separate from the web server, but it should all be on the same box.
Then if you use a NAS for the data storage, the server hardware is quite
compact, and mirrored NAS storage is inexpensive and easy to clone.
I am not sure of the mail volume or the the size of his pipe to the
Internet, but powerful single boxes are somewhat easier to get ones head
around and possibly to administer and they can process a lot of data very
fast. I find they are less expensive to buy.
Splitting the load, and having the servers not identical in respect of
the default tasks they are setup for, makes each of them very unique.
Failing over one to the other could be more complex.
Am just surmising here. Why not use enough horsepower to do a job on one
server for the assigned task. In behind have Gigabit Lan and on the
outside have 100 Mb Lan.
We have no indication of the bandwidth useage or bandwidth available
here. Maybe you need multiple pipes to support the bandwidth and that
can change the design.
If one task seems to hog bandwidth and makes either the mail or web
serving seem slow maybe traffic 'shaping' is an option. If there are
mulitple separate Internet gateways on redundant providers, then the
design becomes waaay more complex. Then carefully programmed routers /
switches could provide much of the load balancing function upstream of
This is an interesting scenario. I would love to see more details of the
infrastructure and the objective, so we could all throw around some
James A. Haliburton
On-Site Computer Services of Halifax
Suite 100, 25 Walton Drive
Halifax, Nova Scotia
Canada B3N 1X6
Office/ Cell : (902)499-5250
Home/After Hours : (902)477-8342
e-mail : jim at on-site.ns.ca
More information about the nSLUG