Ideal Server Hardware Choice
Tony Dietrich
td at transoft-adsl.demon.co.uk
Tue Feb 28 23:17:44 UTC 2006
On Tuesday 28 February 2006 21:06, Timothy Alberts wrote:
> I'm researching purchasing 2 new servers for our network. Ideally I
> want to be 100% compatible with FC Linux (obviously). Our network is 50
> clients with 164K internet connection (slow). We run:
>
> -Internet gateway via iptables/firewall
> -DHCP/DNS for local network
> -Sendmail averaging 3-4k messages/day with POP3 via dovecot
> -Apache web for intranet with perl/php/mysql.
> -MySQL database is large, but access is maybee 3-5 clients at any given
> time.
> -Ideally converting to Tomcat application server for intranet (replacing
> perl/php).
> -FTP (maybee SAMBA again someday)
> -OpenLDAP for unified login, company address book
> -Remote access may be added when we get the bandwidth to support it...
>
> I have been researching:
>
> IBM eServers x206 x226 anyone know if these are good systems for FC
> linux?
>
> Sun's X2100 systems.
>
> Dell 1800, looks fairly standard.
>
> So, do any of you Fedora Users/Network Admins have any recommendations
> for me? Does Fedora have a list of their test farm servers so I can see
> what they use? I remember seeing a note IBM donated some servers to
> Fedora recently.
Timothy
I've installed systems/networks for several customers with similar
requirements.
I firmly believe that for a small firm, the priority isn't necessarily getting
a high quality (and also potentially expensive brand-name server/system), but
making sure that whatever system you source can be quickly and easily fixed
if it breaks down. (OK, we all hope that never happens, but sods law says it
will for the one client who expects perfection.)
I tend to suggest that a single-location firm of your size takes a look around
at the local system-builders .. people who have been around a few years, can
put together a decent system, but tend to be less expensive than the
brand-names, but more importantly, are willing to stand by their products and
provide the immediate backup you hope you'll never need. Often the need to
maintain a good reputation within the local community means that local system
builders are the ones that are willing to get out of bed at 11pm just to run
a new network card over to you. I'm luck, my local favorite supplier lets me
have his home number, and has on several occasions worked on a problem far
beyond the call of duty .. simple because he knows that if the word gets
around that he doesn't want to bother, he won't survive. A good relationship
with such a supplier can be worth its weight in gold in a very short time.
I also wonder if you are over-speccing. As long as you haven't got any
unusual hardware installed, FC is pretty happy with most systems ....
It may pay to look at lower spec machines, but more of them. The one thing I
can never impress enough on customers is the need for backups, backups and
more backups, both of data and of systems. Physical redundancy for primary
systems seems to be out of favour in the modern world, but I swear by it, and
I seem to have happy customers
I have one firm that runs its network (with similar requirements except for a
much higher quantity of email) from a cluster of servers that are barely
capable of running a decent desktop.
My own firm's email server is running headless (no X installed) on a salvaged
1998 spec desktop (1.4mHz 512mB RAM) machine, and quite happily dealing with
2k emails a day using MailScanner, clamav, spamassassin, dovecot and postfix.
(FC4, recently upgraded from RH9). Yes, there's a similar spec salvaged
system, with an identical install, sitting next to it unused ready to be
switched on and take over at a moments notice .. I can even do a remote
wake-on-LAN if I have to. I boot it up once a week and get it to sync its
user files in case we've added/deleted a user that week. Both systems
together cost me £250.00GBP.
Our primary and secondary DNS/DHCP servers are even more ancient ... 350mHz
systems .. but they ONLY do DNS/DHCP, and apart from updating the packages
are still running their original RH9 install, circa 2001.
A couple of systems dedicated as routers, again older systems, but very
lightly loaded ... most of the time their hard-disks are powered down and
they are running on memory only.
Our webserver is only a bit better off as far as its spec goes, and that deals
with around 2.5k external hits per diem/ perhaps 5k internal (we run an
electronic web-based diary system) , running apache/MySQL/PHP on an FC4
install.
A couple of LDAP servers complete the setup.
Our heaviest used server supports NFS and Samba shares for up to 80 systems
(local and remote), so runs on a 2gHz system with RAID and tape backup, but
even so its not in truth a 'server spec' machine. We have another 'backup'
NFS/Samba server which again is a rescued system, and sits there quite
happily doing nothing most of the time apart from syncing its files off the
main server, ready to take over if needed .. I think that system cost us
£50.00GBP. Providing I see regular logwatch emails showing me the disks are
OK, I know that at least I won't have to panic about all the staff being
unable to work while I get the main server repaired/reinstalled/restored from
backup (or whatever it needs).
I'd guess the entire setup cost us about 2k GBP over a period of two/three
years, with about 1.25k of that on the main server. Our UK tax system means
that computer prices are about pound-for-dollar at the moment.
The way we sourced the systems both for ourselves and for several of our
customers means that we can provide a larger number of systems, each of lower
spec, but each capable of doing one particular job ... but then we set them
up so that if need be they can take over the job of one or more of the other
servers on the network if *that* server goes down .. its old fashioned
redundancy in numbers, but it works for firms around your size.
Of course configuring such a 'server farm' is a little more work .. but once
its up and running, its pretty stable, and tends to mean that a firm which
might not have a dedicated IT support doesn't have to panic too much if one
particular system goes down .. just about anyone can be taught how to hit the
power-on switch for the right backup system, and then a more leisurely call
can go out to whoever is supporting the network (you in your case I guess?).
I've also had it commented to me by one customer that insurance companies LIKE
to see customers with more physical redundancy .. his data retrieval
insurance policy got cheaper by around 30% after we converted him.
The final icing on the cake is that some of the less powerful systems draw
less power and create less heat .. in fact one or two of the systems are
fanless. Few moving parts sometimes means less things to go wrong. It also
means less noise, which some customers really like.
YMMV
(BTW, did I mention I'm a miser .. and I've found being a miser with
customers' money tends to mean they come back again and again, so in the end
I earn more out of them (just over a longer period) than if I went for broke
and got them the best/most well-known systems available :p)
Each to their own :-)
Tony
--
Tony
More information about the fedora-list
mailing list