What flavor for home server?

[Headcase_Fargone]

I'm looking to take the plunge into the world of Linux and build a little home server/NAS. My question though is: What distribution do I run?

I have some small bit of experience with Ubuntu desktop (never played with the server version), but I'm wondering if a lighter weight distro is in order. If Ubuntu, desktop or server? If not, what? Debian? Puppy?

I'm currently running a hacked DNS-321 with funplug and a few small apps installed, so I also have some experience with the command prompt. A pretty GUI isn't 100% necessary.

This build will consist of an Atom N270 and 2GB of RAM. OS will be running on a 4GB CompactFlash drive through the IDE bus, and storage will be two 5400rpm 1.5TB drives, possibly RAIDED, possibly JBOD'ed, depending on the software RAID options Linux offers (which I hear are exceptional).

I'll be streaming media both to televisions on the same LAN and my also to my phone, torrenting, etc. Nothing out of the ordinary. Primary focus is on power consumption.

What say you?

 

[Crusty]

I see no reason why one of the tailored distros like FreeNAS wouldn't work just fine for your purposes. If you want something that just works I would definitely go that route instead of trying to roll your own from a basic Linux distro.

Personally my home network uses an Ubuntu server box to store my files and run various tasks for me in the background, but I've had plenty of experience using Linux and setting it up for server use.

 

[JD50]

I'd go with either CentOS or an Ubuntu LTS release, depending on your preference between RHEL and Debian based distros.

 

[bobross419]

If you want to play with Linux while building the server, I'd give a thumb to CentOS.

If you just want it to go then you might look at Amahi. Its pretty slick for home server purposes and was my original choice before I decided I wanted to actually learn linux.

 

[Headcase_Fargone]

I'm certainly not averse to learning more about Linux. But at the same time I don't know that I want to jump into the deep end of the pool either. Looking a bit into CentOS I'm wondering if that would be the case.

Looking into Amahi now. Will both of these run alright on such an anemic platform as the Atom?

 

[Nothinman]

I'd avoid CentOS unless you have some need for RHEL compatibility or just want to learn that system more. I'd personally just use Debian, but Ubuntu might be a bit simpler to start off with.

 

[Circlenaut]

I'm currently running ClearOS for my atom router and Freenas for my fileserver. They both are relatively easy to use and the web gui is intuitive. For you I'd recommend either ClearOS (CentOS based) or amahi if your looking to do extra customization.

 

[Binky]

I know this is a *nix forum, so people are going to throw tomatoes at me for saying this, but don't overlook Windows Home Server. It's brain-dead simple, and it works very well with low power systems.

IMHO, experiment with something that isn't important, like a dual boot, or a secondary system. Don't make your first linux box something that serves files and media all around the house.

 

[child of wonder]

Debian. Stable and solid.

 

[Nothinman]

I know this is a *nix forum, so people are going to throw tomatoes at me for saying this, but don't overlook Windows Home Server. It's brain-dead simple, and it works very well with low power systems.

IMHO, experiment with something that isn't important, like a dual boot, or a secondary system. Don't make your first linux box something that serves files and media all around the house.

I know lots of people have had great success with WHS, but I'm not willing to pay to be limited by something like that.

 

[bobross419]

I know lots of people have had great success with WHS, but I'm not willing to pay to be limited by something like that.

My first real Linux box is a home server (CentOS) and it works great

 

[Hooray_Yogurt]

If you want to go the "full" linux route there will be little performance difference from one distro to another. Just install what you're familiar with.

 

[Headcase_Fargone]

I know this is a *nix forum, so people are going to throw tomatoes at me for saying this, but don't overlook Windows Home Server. It's brain-dead simple, and it works very well with low power systems.

IMHO, experiment with something that isn't important, like a dual boot, or a secondary system. Don't make your first linux box something that serves files and media all around the house.

I looked into it, but apparently WHS requires an 80GB installation drive. Right now I'm planning to install the OS to a 4GB compact flash drive (via IDE adapter) allowing the two 3.5" drives to spin down when not in use as part of my super-low power master plan.

I've been piddling around with Ubuntu for the last week or so and am really enjoying it. I may just end up going with Xubuntu or a full Ubuntu distro. I'll have to see how the little Atom I intend to build it with handles it.

I must say I'm anxious about entrusting all of my media to a Linux machine knowing next to nothing about the operating system and its inner workings. It seems I've already locked myself out of root access on my DNS-321 by altering a telnet script, heh.

 

[PCTC2]

I looked into it, but apparently WHS requires an 80GB installation drive. Right now I'm planning to install the OS to a 4GB compact flash drive (via IDE adapter) allowing the two 3.5" drives to spin down when not in use as part of my super-low power master plan.

I've been piddling around with Ubuntu for the last week or so and am really enjoying it. I may just end up going with Xubuntu or a full Ubuntu distro. I'll have to see how the little Atom I intend to build it with handles it.

I must say I'm anxious about entrusting all of my media to a Linux machine knowing next to nothing about the operating system and its inner workings. It seems I've already locked myself out of root access on my DNS-321 by altering a telnet script, heh.

If you're anxious, I would definitely recommend buying a few 2TB external harddrives and backing up all your data to it.

For a home server, Atoms are perfectly capable as long as you don't need more than a few hard drives without any form of RAID. I used an Atom 330 board for awhile for a SOHO server. It's my day-to-day desktop computer now for reading these forums and listening to music running Ubuntu 10.04 LTS.

Personally, I generally use Debian and Red-hat based distributions for my servers. Red hat is running my main production server and ClearOS (CentOS-based) is running my proxy/VPN/firewall. Ubuntu gets virtualized for web applications, etc and also gets installed on all of my desktops.

 

[electroju]

Using compact flash or any NAND Flash based SSD will have problems under Linux because of syslog daemon. The daemon syslog writes every second, so your compact flash will wear out. At this time, it is best to use a regular hard drive.

I suggest either Gentoo or Arch. Arch is good if you do not want to take a lot of space and do not want to compile programs to install programs. Arch gives you the ability to compile to fine tune the features for security or add features to add more functionality. Gentoo requires that you compile programs to install them. Both Gentoo and Arch provides utilities to simplify the process of checking for dependencies and compiles to your specifications. Though these distributions are suggestions, but I recommend to anybody setting up their own server and they want to use Linux is use a distribution they are familiar with instead of trying something out.

If I had to setup a file server for home use, I will go with OpenSolaris. OpenSolaris has ZFS support to help minimize data corruption and it has a specialized setup that does not get hurt by too many software upgrades. The processor I will go with is AMD because even their desktop processors includes ECC for memory which helps reduce data corruption.

Linux software RAID or mdadm is good if you have ECC memory, but it hurts your data if non-ECC memory is used. I am sure OpenSolaris has the same software RAID abilities as Linux. Software RAID is great because it consumes processor usage more than you think. If you setup RAID-5 through software RAID, performance of the computer will suffer. The Intel Atom processor will have problems with it. All the streaming you want it to do will be crawling. I suggest a hardware RAID if you are going to use an Intel Atom setup. True hardware RAID controllers that work in Linux are from 3ware or Areca. There are stand-alone hardware RAID cards that connects to SATA and they do not require software to work, but they require port multiplier support. The following is a stand-alone hardware RAID card.

Addonics 4X1 Hardware Port Multiplier with eSATA/USB
http://www.addonics.com/products/host_controller/ad4sr5hpmus.asp

 

[Nothinman]

Using compact flash or any NAND Flash based SSD will have problems under Linux because of syslog daemon. The daemon syslog writes every second, so your compact flash will wear out. At this time, it is best to use a regular hard drive

Or you can just redirect the syslog output to a ramdisk and flush it periodically, point it to a remote server or just disable it. Although disabling it completely is probably a bad idea.

I suggest either Gentoo or Arch

I haven't used Arch, but I wouldn't recommend Gentoo for anything except maybe as a lesson as to what not to do with Linux.

If I had to setup a file server for home use, I will go with OpenSolaris. OpenSolaris has ZFS support to help minimize data corruption and it has a specialized setup that does not get hurt by too many software upgrades.

Except then you have to deal with Solaris. I'd rather go with Debian and md+LVM+XFS than put up with Solaris just for ZFS.

The Intel Atom processor will have problems with it. All the streaming you want it to do will be crawling.

Software RAID5 shouldn't slow reads, if anything they'll be almost as fast as RAID0 because of the data striping. Writes may be slower because the CPU has to do the XOR calculations, but I can't imagine it would be unbearable on a single-user home machine.

 

[Headcase_Fargone]

Using compact flash or any NAND Flash based SSD will have problems under Linux because of syslog daemon. The daemon syslog writes every second, so your compact flash will wear out. At this time, it is best to use a regular hard drive.

Ouch, no one's ever mentioned that. Does that prevent the system idling?

I'm assuming this is some integral part of Linux that can't be disabled as well? Or at least turn the logging down to minimal?

 

[PCTC2]

Ouch, no one's ever mentioned that. Does that prevent the system idling?

I'm assuming this is some integral part of Linux that can't be disabled as well? Or at least turn the logging down to minimal?

Even at "idle", a lot of processes log their actions that are either chronological or whatever. However, given you will not be using it for anything other than an OS, this is a little bit of a stretch. SSDs may have a little more durability when it comes to writing (better NAND), but you will have to write the size of the drive thousands of times over to kill the NAND. The largest logs I've ever had were a 2GBs and that was from a few year old enterprise web server dealing with 10 heavily trafficked sites. I don't see you killing FLASH anytime soon. If you want, you could spend a little more, and buy a little better quality flash such as an industrial application IDE SSD that sits in the IDE port on the motherboard.

 

[Headcase_Fargone]

Enh, the CF card and adapter are $20 combined. I suppose I should have it create a backup of itself somewhere nightly.

If the CF card fails, will I lose the data on the RAID volume since it's software RAID?

Thanks for all of the information by the way, folks.

 

[Crusty]

Linux software raid persists a superblock onto each drive that is part of an array that details what it contains. Everything mdadm needs to rebuild/recreate the array is stored on the drives themselves.

 

[Headcase_Fargone]

So if I kept a nightly or even weekly backup of that little CF drive and just tossed in a fresh $15 card I'd be good to go again with no data loss? That's pretty cool.

 

[electroju]

Headcase_Fargone, I still recommend you to use regular hard drives. You could use notebook hard drives. If you select a model that is efficent, then you do not have to rely on a compact flash or NAND Flash based SSD for low power consumption. Linux at this time does not support TRIM command well enough for production systems. Sure you can use a compact flash or NAND Flash SSD for Linux, but the throughput will be around 1 megabyte per second or lower. If you want to go with solid-state disk, go with DRAM based. You do not have to worry about wear and tear. Also you do not have to worry about using the TRIM command. The following device is a DRAM based SSD.

ACARD ANS-9010
http://www.acard.com/english/fb01-p...&type1_title= Solid State Drive&type1_idno=13

OR

ACARD ANS-9010b
http://www.acard.com/english/fb01-p...&type1_title= Solid State Drive&type1_idno=13


If you do put two 1.5 TB hard drives in RAID-1 and one drive fails, the time it will take to reconstruct the data is about 17 hours. Probably in the best condition it will be about 9 hours.

Or you can just redirect the syslog output to a ramdisk and flush it periodically, point it to a remote server or just disable it. Although disabling it completely is probably a bad idea.

I do not recommend putting logging in ramdisk or tmpfs because all the logs will vanish every time the computer is powered down. If there is any problems like memory glitches, this setup will not catch it. Third, if the size of tmpfs is too big or larger than RAM, the system have to swap pages out to disk and this causes a performance penalty. Though this is not just syslog you have to worry about because there are /var/run, /var/spool, and /var/tmp that also writes to disk multiple times, so again only use regular hard drives.

I haven't used Arch, but I wouldn't recommend Gentoo for anything except maybe as a lesson as to what not to do with Linux.

I have used both Arch and Gentoo. I prefer Gentoo. They work great for servers because you are not force to install tons of programs to just get a server going. I use Gentoo on my desktops and notebooks on a daily basis. I use Linux as my primary operating system.

IMHO, Gentoo and Arch handle Linux the right way because Linux has tons of different standards to handle. These distributions combines all the standards into one standard that all maintainers have to comply to. Other distributions mainly pre-compiled versions have different variations of standards that are caused by the maintainers of each program. When using programs for these distribution, you are at their mercy of selfish thinking. Some maintainers from these pre-compiled distributions may include all features with out thinking about stability, reliability, and security. Others may just go for security.

Except then you have to deal with Solaris. I'd rather go with Debian and md+LVM+XFS than put up with Solaris just for ZFS.

Understand that yu prefer Debian over other distributions, but using Solaris as a bad thing is not bad. People trying out Linux wants a Windows like operating system. OpenSolaris and Ubuntu provides that. OpenSolaris is better for a file server compared to Linux because you do not have to handle multiple standards.

Software RAID5 shouldn't slow reads, if anything they'll be almost as fast as RAID0 because of the data striping. Writes may be slower because the CPU has to do the XOR calculations, but I can't imagine it would be unbearable on a single-user home machine.

RAID-5 slows down reads compared to RAID-0. Two hard drives in a RAID-0 will have a throughput of 50 megabytes per seconds given that one drive can handle 25 MB per second. RAID-5 provides parity checking and writing, so the bandwidth will not be as good. RAID-5 throughput will be around 40 to 45 megabytes per second using three disks although it has a little advantage that it can handle multiple writes or more than one write queues. RAID-5 with three disks spreads out the data in two disk data and one disk parity that take turns storing data or parity, but two drives are always striping in a three disk setup for RAID-5. I recommend for software RAID is to stick with RAID-0 or RAID-1 to minimize processor usage.

Linux software raid persists a superblock onto each drive that is part of an array that details what it contains. Everything mdadm needs to rebuild/recreate the array is stored on the drives themselves.

Sure if you want to rely on that to have a reliable setup during upon boot up. Not always it is reliable. It is best to include RAID configuration in initrd or initramfs. Though only kernel version 2.6.28 and up includes this autodetect feature, so double check it before relying on it.

I prefer stay away from software RAID, but I do not use RAID for my destkops. If I had to use RAID, I would use hardware RAID.

The largest logs I've ever had were a 2GBs and that was from a few year old enterprise web server dealing with 10 heavily trafficked sites.

The size of logs can easily reach +2 GB easily for any desktop or server within minutes. I place /var/log on its own partition to minimize the log from over filling my root. The reason why I do this is because logs can grow with out any warning. There are house keeping scripts to manage logs, so they do not get huge.

 

[PCTC2]

Headcase_Fargone, I still recommend you to use regular hard drives. You could use notebook hard drives. If you select a model that is efficent, then you do not have to rely on a compact flash or NAND Flash based SSD for low power consumption. Linux at this time does not support TRIM command well enough for production systems. Sure you can use a compact flash or NAND Flash SSD for Linux, but the throughput will be around 1 megabyte per second or lower. If you want to go with solid-state disk, go with DRAM based. You do not have to worry about wear and tear. Also you do not have to worry about using the TRIM command. The following device is a DRAM based SSD.

ACARD ANS-9010
http://www.acard.com/english/fb01-p...&type1_title= Solid State Drive&type1_idno=13

OR

ACARD ANS-9010b
http://www.acard.com/english/fb01-p...&type1_title= Solid State Drive&type1_idno=13


If you do put two 1.5 TB hard drives in RAID-1 and one drive fails, the time it will take to reconstruct the data is about 17 hours. Probably in the best condition it will be about 9 hours.



I do not recommend putting logging in ramdisk or tmpfs because all the logs will vanish every time the computer is powered down. If there is any problems like memory glitches, this setup will not catch it. Third, if the size of tmpfs is too big or larger than RAM, the system have to swap pages out to disk and this causes a performance penalty. Though this is not just syslog you have to worry about because there are /var/run, /var/spool, and /var/tmp that also writes to disk multiple times, so again only use regular hard drives.



I have used both Arch and Gentoo. I prefer Gentoo. They work great for servers because you are not force to install tons of programs to just get a server going. I use Gentoo on my desktops and notebooks on a daily basis. I use Linux as my primary operating system.

IMHO, Gentoo and Arch handle Linux the right way because Linux has tons of different standards to handle. These distributions combines all the standards into one standard that all maintainers have to comply to. Other distributions mainly pre-compiled versions have different variations of standards that are caused by the maintainers of each program. When using programs for these distribution, you are at their mercy of selfish thinking. Some maintainers from these pre-compiled distributions may include all features with out thinking about stability, reliability, and security. Others may just go for security.




Understand that yu prefer Debian over other distributions, but using Solaris as a bad thing is not bad. People trying out Linux wants a Windows like operating system. OpenSolaris and Ubuntu provides that. OpenSolaris is better for a file server compared to Linux because you do not have to handle multiple standards.


RAID-5 slows down reads compared to RAID-0. Two hard drives in a RAID-0 will have a throughput of 50 megabytes per seconds given that one drive can handle 25 MB per second. RAID-5 provides parity checking and writing, so the bandwidth will not be as good. RAID-5 throughput will be around 40 to 45 megabytes per second using three disks although it has a little advantage that it can handle multiple writes or more than one write queues. RAID-5 with three disks spreads out the data in two disk data and one disk parity that take turns storing data or parity, but two drives are always striping in a three disk setup for RAID-5. I recommend for software RAID is to stick with RAID-0 or RAID-1 to minimize processor usage.





Sure if you want to rely on that to have a reliable setup during upon boot up. Not always it is reliable. It is best to include RAID configuration in initrd or initramfs. Though only kernel version 2.6.28 and up includes this autodetect feature, so double check it before relying on it.

I prefer stay away from software RAID, but I do not use RAID for my destkops. If I had to use RAID, I would use hardware RAID.



The size of logs can easily reach +2 GB easily for any desktop or server within minutes. I place /var/log on its own partition to minimize the log from over filling my root. The reason why I do this is because logs can grow with out any warning. There are house keeping scripts to manage logs, so they do not get huge.

Where do I begin?

1) DRAM based disks are as bad as a RAM disk. They require battery backup for any power failure or moving of the server. Use a flash based drive or use a hard drive. Stay away from RAM-based solutions unless you want to rebuild your root every few weeks.

2) Many distros don't install many side-packages if you use their alternate install disk or net-install disk. Ubuntu server is very light-weight and you can select an even lighter install using the standard install media. Debian is very decent. Gentoo, for a beginner, is a good learn experience but for a server I would say stick to a pre-defined packaged server distro.

3) OpenSolaris is a pain. It is. For a home server, there are plenty of better options. ZFS is not the end-all-be-all filesystem. And it's not worth trying to get all the packages you want onto it. Debian and Ubuntu have some of the best repositories when it comes to having packages.

4) Stay away from soft-raid. PERIOD. If you want RAID, get a "cheap" LSI/Intel/HighPoint hardware RAID card and do a RAID 5. RAID 5 with 3 disks is a distributed parity with a 1-disk failure tolerance. It's space efficient (compared to RAID 1/10) and it has decent fault tolerance.

5) Where do you get 2GB of logs in a few minutes? Logs are PLAIN TEXT. In an enterprise web server, I have logs from September 2009 until today, and the logs are only 9GB. I don't know how you could get 2GB of plain text that quickly. My dev server, for it's entire lifetime of a few years, only has 500MB of logs. My network gateway may have a few GB but that's only because it stores network bandwidth usage for every second for as long as I run it. Logs are not an issue for a HOME SERVER.

 

[VinDSL]

Except then you have to deal with Solaris.[...]

I hear ya!

I'm running OpenSolaris 2009.06 on a multi-booted Toshiba lappy.

OpenSolaris is quite nice! It *looks* the same as any other Gnome-based Linux distro, blah, blah, blah.

The thing that pisses me off is...

The GRUB they use is incompatible with GRUB in Linux. Every time I boot into OpenSolaris, it wipes my Linux GRUB, and I have to rebuild it before booting into Mint 7 or Vista.

Er...

Is this what you mean by "deal[ing] with Solaris"? LoL!

Really, as a standalone, OpenSolaris would be fine! It just isn't 'multi-boot' friendly!

I'm certainly not averse to learning more about Linux.[...]

I'm running Slack on my home box.

My HDD just went 'read only'. LoL!

Guess I should be fixing that instead of hanging out in the forums...

 

[Nothinman]

I do not recommend putting logging in ramdisk or tmpfs because all the logs will vanish every time the computer is powered down. If there is any problems like memory glitches, this setup will not catch it. Third, if the size of tmpfs is too big or larger than RAM, the system have to swap pages out to disk and this causes a performance penalty. Though this is not just syslog you have to worry about because there are /var/run, /var/spool, and /var/tmp that also writes to disk multiple times, so again only use regular hard drives.

Which is why I said it was probably a bad idea. And on most machines those directories don't get that much work, /var/run really only holds a few PID files and sockets, /var/tmp is barely touched by anything and /var/spool only gets a workout if you process a lot of mail or print jobs.

Essentially, unless it's a mail server the only thing you need to really worry about is syslog.

I have used both Arch and Gentoo. I prefer Gentoo. They work great for servers because you are not force to install tons of programs to just get a server going. I use Gentoo on my desktops and notebooks on a daily basis. I use Linux as my primary operating system.

Gentoo is terrible for servers because there's no consistency. Sure, if you're only dealing with 1 or 2 it's no big deal but as soon as you've got to worry about a half dozen or more you don't want to have to think about what hacks you've done on each individual system. And binary packages are much better for patching. I know 100% that every package on my system is exactly the same as the one that was put together and tested by the Debian developers. There's no chance that my local compiler flags, installed packages, etc affected it.

And from a security standpoint it's practically idiotic to have a compiler on a system you want to consider even remotely secure. On my workstations they have to be there things like the nVidia driver, but for a server it makes no sense.

IMHO, Gentoo and Arch handle Linux the right way because Linux has tons of different standards to handle. These distributions combines all the standards into one standard that all maintainers have to comply to. Other distributions mainly pre-compiled versions have different variations of standards that are caused by the maintainers of each program. When using programs for these distribution, you are at their mercy of selfish thinking. Some maintainers from these pre-compiled distributions may include all features with out thinking about stability, reliability, and security. Others may just go for security.

You're always at someone's mercy, unless you write all of the software yourself from the boot loader to the compiler to the apps you're using. I would rather use the same binaries that are known good and working on millions of other installations than some locally compiled one that may or may not have unintended side affects from something I did or installed.

Understand that yu prefer Debian over other distributions, but using Solaris as a bad thing is not bad. People trying out Linux wants a Windows like operating system. OpenSolaris and Ubuntu provides that. OpenSolaris is better for a file server compared to Linux because you do not have to handle multiple standards.

Yes, I consider having to use Solaris a bad thing. I would probably be slightly higher on my list than Windows, but not by much. And that's compounded by the Oracle buyout. They're going to have to do something about ZFS and BTRFS and in either case I'd guess that Linux will be the beneficiary, not Solaris.

Two hard drives in a RAID-0 will have a throughput of 50 megabytes per seconds given that one drive can handle 25 MB per second.

Except that will never happen in real life. Filesystem overhead, processes competing for I/O, etc will make sure of that. The affect of RAID5 parity calculations will generally fall into statistical noise with the speed of processors relative to drives today.

I prefer stay away from software RAID, but I do not use RAID for my destkops. If I had to use RAID, I would use hardware RAID.

And I would go the exact opposite. The flexibility of Linux software RAID far exceeds most of the benefits of hardware RAID. And I'd rather not be tied to one vendor, with software RAID I just plug the drives into any controller and off I go.

The size of logs can easily reach +2 GB easily for any desktop or server within minutes. I place /var/log on its own partition to minimize the log from over filling my root. The reason why I do this is because logs can grow with out any warning. There are house keeping scripts to manage logs, so they do not get huge.

Yes they can, but it's not normal by any means as logrotate should handle that for you. Mine are under 500M here and I've been using this installation for nearly 3 years old now. My home mail server is slightly larger at ~800M, but nowhere near 2G.

And I've never understood the fear of filling up the root filesystem, it doesn't really cause any problems. Filling up /var or /home is going to have much, much worse effects.

OpenSolaris is quite nice! It *looks* the same as any other Gnome-based Linux distro, blah, blah, blah.

No, Gnome looks nice. OpenSolaris is ugly as ever.

Is this what you mean by "deal[ing] with Solaris"? LoL!

No, I mean the pain of actually having to install, setup and maintain it.