Take the 2-minute tour ×
Super User is a question and answer site for computer enthusiasts and power users. It's 100% free, no registration required.

I am a freelance developer. I have a PC, a laptop and an old testing and file server machine. I might add one or two in future.

I want to have an on-site backup machine that can handle backups of ALL these machines - file backups, MySQL backups, backup of subversion repository, etc..

When building the machine, which components should I invest more in?

Examples:

The cabinet should have lots of room for expansion.
Hard disk size should be large.
But I guess hard disk speed need not be high (?) Also, Do I need a very faster processor?

But other components like, RAM, PSU, Processor, Network card, Cooling, etc.. how much relative importance do these have in a backup machine? Which of these components should be high-end or large, and which ones need not be?

Some Idea of the load: There will TBs of data. File backups and subversion repository backups will at least be done daily. MySQL backups done weekly. assume 3 machines at the moment and somewhere around 10 machines in the future.

share|improve this question
    
Is this going to be your only backup or are you also taking important stuff off site? –  Linker3000 Feb 20 '11 at 16:33
    
There will be an off-site backup. This on-site backup is to cover the cases where someone deletes files, Windows installation goes south, some machine's hard disk fails, subversion repository is corrupted, etc.. These things happen more often than natural disaster, theft etc.. So this is for quick restoration and running. There will be another off-site backup in case of something really bad happens. –  user57813 Feb 20 '11 at 19:01

2 Answers 2

up vote 0 down vote accepted

It really depends on your budget.

RAID is obvious requirement. I would invest on good hardware RAID controller for performance and stability, but others may recommend software RAID. Speed requirements really depend. We are backing up about 8TB of data using backuppc, and powerful Dell server with 8xSATA (7200RPM) disks in RAID5 is awfully slow. On the other hand, for example SAS disks are way too expensive.

Enclosure: buy large enough, but not overkill. Extending RAID array with new disks is anyway difficult, when your server gets too small, most probably you want to buy another one.

Network card: as long as it works. If you have 1Gbit/s network, fetch one that supports it.

For memory and CPU: there can't be too much. Grab processor and memory with good price point (not highest-end, but not absolutely cheapest ones either).

When you aren't taking backups more often, processor and memory aren't really important. If you buy too powerful machine, you have to throttle backups, because otherwise you are slowing down your production environment.

share|improve this answer
    
+1 Thanks Olli :) That gave me a good idea about the importance of different components. –  user57813 Feb 20 '11 at 15:59
    
@Senthil: if that answered your question, you should accept it as an answer. –  Olli Feb 20 '11 at 15:59
    
As Olli hinted - go for software RAID OR buy two identical hardware RAID controllers so that if/when the main one dies you can fire up with the spare and still get at your data. Good call on BackupPC - I use it on several sites - but remember that it doesn't natively cope with backing up open files so something will have to dump out the databases so they can be backed up reliably. –  Linker3000 Feb 20 '11 at 16:36

If the machine is only doing backups, i.e. you have no requirements otherwise as to what applications should run on it, it shouldn't run as an HTPC, etc., you might want to look into using Solaris, a Solaris spin-off (IllumOS, OpenIndiana), or a *BSD variant, as they all support ZFS as native kernel modules (Linux also supports ZFS, but only as a FUSE module - it's slower, though very stable, I've been using it for almost two years in a 14TB array).

ZFS supports compression, snapshotting (enabling you to roll back to the time that any snapshot was taken), deduplication (discarding duplicate blocks to save space - but very very slow), and filesystem-level redundancy. The latter being the coolest of all, in that you can set up RAID arrays that are equivalent in redundancy to "normal" RAID solutions, e.g. you can have single, double or even triple parity on each blocks, but parity-checking and recovery is done on the file level (all files are checksummed and continually verified on-the-fly on reads), so you'll never experience silent corruption of data, as the FS is protected, not the array.

As far as cases go, it really depends on how many disks you plan on stuffing in there. I have a fantastic Lian Li case with 12 disk bays in a dedicated section at the bottom of the case. It's all-aluminum, very cool, and very quiet. The build quality is amazing. I recently modified it to hold an extra 6 drives by dismounting the door and removing all 5.25" bays. My particular model seems to be out of production, but they have quite a few really neat cases. Check out their collection of larger ones here:

http://lian-li.com/v2/en/product/product03.php?cl_index=1&sc_index=25

About the hard disks, then no, I wouldn't focus on their speed so much. Using a lot of disks with parity, you'll soon run into bottlenecks elsewhere. In my particular setup, I have disks on an internal ICH10R controller (PCIe) and other on old PCI32 SATA expansion cards. The PCI bus is completely saturated when doing contiguous reads, but I still get 175 MB/s throughput. Plenty, plenty, plenty. Twice what I can expect from gigabit ethernet anyway.

The Western Digital green drives are somewhere in-between slow and fast, but the two 2TB ones I have are already showing a little too many relocated bad sectors for my liking, and they're under a year old, so I wouldn't personally recommend those. I personally just go for the cheapest until getting enough aversion against the brand to try something new. Currently, I'm not too fond of Samsung and Western Digital, but have a lot of very old Seagates (almost a decade) that are still 100%, so my last two 2TB drives were Seagate, too.

With regards to temperatures, keep in mind that the Google harddrive paper from a couple of years ago hinted at temperatures not being as important as previously thought. That is, when still within sane values, of course, but whether they're spinning at 20C or 50C it seems isn't as important after all.

share|improve this answer
    
Thanks :).. what about RAM, PSU, Processor? What capacities/performance should they have? –  user57813 Feb 20 '11 at 15:52
1  
If only fileserving and backing up, the solution I propose here would be happy with 1, maybe 2 gigs of RAM. I'm personally using the very cheap E2200 Pentium dual core processor for my Linux install with FUSE, and it's more than enough, so if you have any hardware in the attic waiting for a second coming, that might be their time now. About the PSU, then it's tricky, because drives do not use a lot of power once running, but they "surge" when you power on your computer (check a spare harddrive, it's right on the label ;) ). So you'll need a PSU with enough amps in order to power up... –  DanielSmedegaardBuus Feb 20 '11 at 16:15
    
Personally, I went through this when adding the additional 6 drives to my 12 drive setup, I simply couldn't boot. As PSUs are really expensive when you need near a kW, and I'm a cheap-ass, I ended up using two spare PSUs from my attic which I modded so that they both power on when the ATX power button is pressed (i.e. they work as one). If you're considering this, let me know, and I'll add links. I have one feeding juice to the mainboard and a few disks, the other feeding juice to the rest of the disks. –  DanielSmedegaardBuus Feb 20 '11 at 16:17
    
Just to clarify about the PSU: you probably won't need more than 100-200W to drive a 12-disk setup, but booting it is a different matter. Harddrives may use as little as .1 amps when running, but as much as over 1 amp when booting up. If your controllers support hotplugging, you can just plug in power plugs in sequence after booting, but for most people that's not very convenient ;) –  DanielSmedegaardBuus Feb 20 '11 at 16:26

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.