Quad Xeon, want to make linux server

[ccb056]

I'm finishing up putting together a server box. Once I get the hard drives next week, all I will have to do is install an OS. In fact, I'm posting this from the machine right now, running knoppix.

Here's what I've got:
Quad P3 Xeon 500mhz 2mb cache
1GB PC100 memory
3 PSU's
SCSI cd
floppy
2x 18gb 10k rpm SCSI hdd's
Belkin 10/100 nic (I know, crap, but I didnt want to rip the 3com out of my tyan tiger mpx)

Anyway, im looking for a *nix os. Must be SMP capable, and preferably command line. Also, I'm not familiar with setting up programs in a *nix environment, but I've spent the last year ssh'ing into various purdue unix servers.

I don't want any worthless crap on the OS that I'm not going to use. The machine will not have a monitor, keyboard, or mouse, everything will be done via ssh.

The machine will be behind a linksys rv042, along with my pc, a dualie 2800+.

These are the services I know I will need installed:
DNS Server (thinking BIND)
Apache 2x
PHP 4x or 5x
MySQL 4x or 5x
FTP (maybe proftpd ????)
MAIL: POP and SMTP

I don't think I will need any sort of software firewall, however, I wouldnt mind putting one on.

Right now, I'm thinking about using either CentOS, Debian, or Slackware.

Any suggestions?

[FunkyStickman]

If you're not experienced with Linux, I'd recommend Debian, Fedora, or Suse (all package-based distros) for ease of setup, stability, and updating.

All versions of Linux support SMP, and (almost) all come with the standard network services, i.e. DNS, FTP, HTTP, PHP, SQL, SMTP, and so on. I'm partial to Suse because it has a fantastic control panel, and is super-easy to set up. Some like Fedora because it has probably the most packages of any distro.

If you can get around in Knoppix, you shouldn't have any trouble with Debian, which it's based on.

Let us know what you go with!

[Cuda68]

While I agree with FunkyStickman's choices, I am also interested in knowing what kind of server this is. You may want to look at the admin tools fairly close for each distro.

[Xamindar]

The Linux distro you choose depends on the person. Both Redhat/Fedora and Suse are backed by big companies and will therefore probably be the most stable. They are both based on the rpm system for installing packages.

Debian is pretty good and it uses its own apt get package management system. Knoppix is based off of Debian. If you are really lazy you can just install your knoppix to the hard drive and use it from then on.

Gentoo is the one I prefer. It uses portage for its package management. Portage is based off of BSD's ports system. When you want to install a program you just type something like \"emerge apache\" on the command line and it compiles and installs the program and any dependancies. Installing Gentoo is a long process and complicated, but it is well documented and teaches you a lot about how the OS works. Also, most configuration is done by editing config files. That may seem scary but once you get used to it you will love it.

Ubuntu is another one which is growing in popularity awefully fast. I don't know anything about it though.

It's really up to the individual to decide what they like. All distros are good and have their own strengths and advantages. I have tried all of the above with the exception of Ubuntu and I found that Gentoo is the best for me.

For what you are using that machine for (web server/mail server) one of the BSDs will probably be perfect. I do not know much about them though. Configuration of various things on BSD is quite different from Linux so if you already know Linux then you might want to stick with it.

Oh, and sence you are using ssh, screen is your friend

[ccb056]

Im probably going to use CentOS, its just a rebadged free version of redhat.

The server will be a website server...

[fliptw]

4 CPUs is overkill for a web server.

[ccb056]

hardly, especially with php, mysql, etc running in the background for 5 different sites

also remember, these are only 500mhz cpus.

[fliptw]

your running more than one site, thats good.

If you bought those CPU's your money would've have been better spent setting up a RAID5 array of IDE disks instead.

If things are going to get choked up, its at the NIC, and the number of CPU's won't help that.

[ccb056]

The cpus cost 20 bucks, total
Hard disks (2) are raid 0 scsi 10k rpm 18gb

nic is 10/100 and the internet connection is 10mbps up.

[fliptw]

The cpus cost 20 bucks, total
Hard disks (2) are raid 0 scsi 10k rpm 18gb

USER ERROR. RICER FAULT.

never use RAID 0 on a machine where you are responisble for someone's else's data, regardless of what kind of drive its composed of, you are doubling the chance of data loss, and setting yourself up for some hassle.

Use RAID 1, or get a 3rd disc and go to RAID 5, but don't use RAID 0 on a server, unless you got seriously deep pockets.

[Xamindar]

RAID5 array of IDE disks

LoL! That's funny.

The cpus cost 20 bucks, total
Hard disks (2) are raid 0 scsi 10k rpm 18gb

USER ERROR. RICER FAULT.

never use RAID 0 on a machine where you are responisble for someone's else's data, regardless of what kind of drive its composed of, you are doubling the chance of data loss, and setting yourself up for some hassle.

Use RAID 1, or get a 3rd disc and go to RAID 5, but don't use RAID 0 on a server, unless you got seriously deep pockets.

I agree with fliptw here. You DO NOT want to use raid 0 on a server. If you do, then you better have very good backup systems.

[fliptw]

the only useful backup system to use with RAID 0 is one that creates a backup for every write - don't make the assumption that the disks won't fail during a backup.

[ccb056]

The chances that one disk fails is less than 0.1%

So.. the chance that the system fails in RAID 0 is 0.2%

Anyway, I've got a 120gb drive, a 20gb drive, and 4x80 gb drives (raid 0) in the dualie, so I can configure the dualie to backup the quad's files every so often.

[Jeff250]

The chances that one disk fails is less than 0.1%

So.. the chance that the system fails in RAID 0 is 0.2%

In what interval? A day? A year? A lifetime?

[ccb056]

at any given time

[Krom]

/me has a RAID 0 with only my data on it and I am the most paranoid DVD burning backup nazi ever.

I agree with flip 100%, you never know with hard drives, 5 minutes from now might be the time that the 0.2% chance happens! Never depend on hard drives.

[ccb056]

I've never had a problem with any of my hard drives; and ive gone through about 20 of them

Hard drives don't just drop dead; there are many signs that show the drive is about to die.

[Krom]

Hard drives DO just drop dead, I've seen it happen a half dozen times. You are lucky if you get warning signs.

[fliptw]

Lets put things into perspective:

Disk recovery starts at 1500. to even attempt to do a RAID 0 recovery you to send all the drives in that array out, just for one drive failing.

And assuming you had a recent backup, you'd need to redo the entire array.

With RAID 5, you get the possibilty of hotswapping the drives, because that stores infomation for restoring the data on the array if one fails. This is called fault tolerance.

And don't Trust SMART - its impossible for it to tell you if you have a head crash or the drive's firmware gets corruptted, which recently happened to me on a two-year old drive in a dell machine, like Krom said, don't trust hard drives.

[Jeff250]

at any given time

OK, if there is a 0.2% chance that your data will be lost at any one instant, think about how many instants there are in one day. I think that it's safe to say that there are at least three quadrillion instants in a day. So there are three quadrillion opportunities per day for your data to be lost with a 0.2% probability. What do you think that says about the future of your data? I'd suggest rethinking the meaning of your statistic and seeking out its context.

[ccb056]

the chances of the drive failing during a period of time would involve the .02 and base-e

its a simple continous compound interest problem, you did study high school math, didnt you?

as for drives failing in a raid 0 array...

in a raid 0 array with 2 drives, each drive's chances of dying is cut in 1/2 due to the simple fact that each drive is used 1/2 as much

[Xamindar]

in a raid 0 array with 2 drives, each drive's chances of dying is cut in 1/2 due to the simple fact that each drive is used 1/2 as much

Ok, I don't even want to get into how flawed that logic is.

[Krom]

in a raid 0 array with 2 drives, each drive's chances of dying is cut in 1/2 due to the simple fact that each drive is used 1/2 as much

Ok, I don't even want to get into how flawed that logic is.

I second that. If that is his logic, only a total failure and data loss will cure him.

[Jeff250]

the chances of the drive failing during a period of time would involve the .02 and base-e

its a simple continous compound interest problem, you did study high school math, didnt you?

Funny, but a continously compounded interest problem only refers to how often the interest is compounded (continously). Continuously compounded interest is advantageous because, the more often you receive payback from interest, the sooner you can begin earning the interest off of that payback.

However, it does not escape the problem of still needing some interest rate for a given interval of time to "compound" (most interest rates are annual rates by the way, but you can be easily ripped off if you don't know the difference). Of course, this is all generously assuming that you can tie any of what you just pulled out of your hat into the problem at hand. So, once again, what interval of time here are we talking about for your .02% failure rate? A minute? Half an hour?

[fliptw]

in a raid 0 array with 2 drives, each drive's chances of dying is cut in 1/2 due to the simple fact that each drive is used 1/2 as much

You ain't doing to well at school, are ya?

If that was true, they'd be no point in RAID0, because you'd only get half performance from each drive using your logic...

[Krom]

And now a message from captain obvious:
Here is some more 'high school math'; 2 drives = 2 times the moving parts = 2 times the chance of failure.

[Xamindar]

yeah lol, it doesn't matter if they are being \"used\" half as much or not, they are still spinning just as fast.

Dude, seriously, get one more drive and do raid 5. Not only is it more safe but it is fun as well. I got an old proliant here with 10 drives in it and it is always fun to watch all the drives read/write at the same time. A bit noisy though.

[Instig8]

Don't use FTP unless in a secure environment (internal network) or for anonymous FTP. If you need FTP use vsftpd. Otherwise, use scp which is supported by most FTP clients.

Sending passwords in plain-text is old-school. You don't still use telnet, do you?

[Xamindar]

yeah don't use telnet. use ssh. you can use ssh for ftp transfers too, type \"sftp\" in linux. I personally only use normal ftp (proftp) for anonymous ftp and set it to some obscure port.