Take the 2-minute tour ×
Server Fault is a question and answer site for system and network administrators. It's 100% free, no registration required.

"Back in the day" we always segregated our OS drives (in Windows) from our Data drives. In the Linux world, although I am much less familiar with it, I am aware that the wisdom dictates even more volumes defined and used in a best-practice configuration.

Now that server storage is just as likely to be on a SAN (where the disk resources are shared by many individual operating systems and applications), does it really matter any more that the OS and Data partitions be segregated at the volume level?

What are your thoughts?

share|improve this question

7 Answers 7

up vote 6 down vote accepted

There are three main drivers for keeping OS and Data separated storage-wise.

  1. Space. As ErikA points out, you really REALLY don't want your OS volume running out of space. All kinds of bad things can happen. Separating these two growth methods
  2. I/O Access Requirements. The type of I/O used on your OS volume is generally a lot different than the kind used by your Data volumes. Keeping your I/O types separate is a very good idea on many levels.
  3. Storage Portability. When it come time to upgrade your server OS, you can nuke the OS volume and keep all the data. Or in SAN or VM environments you can just move the Data volume to a new, freshly installed server and save time on upgrades.

Also, some operating systems (Windows is among them) don't take too kindly to resizing the OS volume, which means you generally need to give as much of it as it will need in its lifetime when you format the server. Contrast this to Data volumes which can and frequently are resized many times over the lifetime of a server. Even in fully virtualized environments where the OS and Datavvolumes themselves are being housed in the same actual storage, not being able to resize your OS volume can be a major handicap. Windows 2008+ is now recommending 30GB for the C:\ drive these days, a far cry from the 10GB we were using on Server 2003; this is something that will nail many Windows admins as they make the conversion from 2003 to 2008.

share|improve this answer
    
These are good points and they touch on deeper issues related to that shared storage management. Example: if your C: and D: are on the same set of spindles in the same RAID config, etc, you can't optimize for IO type. For storage portability, equally important to make sure the C and D drives are actually different LUNs whatever object is backing the virtual storage. Otherwise, if they're just partitions on the same LUN then you can't move one to a new server without doing a full copy. –  Jeremy Sep 17 '10 at 19:12

Yes, most certainly separate OS from data. I've seen it time and time again where, with a shared partition, the partition ends up filling up and making it impossible to patch the OS, impossible to extend the partition (due to various reasons), etc.

IMO, the overhead of managing two partitions is a small price to pay for the isolation provided.

With regards to the SAN-backed systems you referred to, that still won't protect you from data filling up your OS partition. With fully-virtualized storage, you don't need to worry as much about ensuring that OS and data live on separate spindles.

share|improve this answer
    
Funny, the reasons you give for separating the OS and data are in fact the very reasons I have for NOT doing so. –  John Gardeniers Sep 10 '10 at 2:42
    
Yes, I suppose there are cases (especially for non-virtualized servers with direct-attached disk) where it might be advantageous to have one large bucket o' disk at your disposal. –  EEAA Sep 10 '10 at 13:10
    
As an interesting side note, due to something odd happening during OS reinstall, my Windows OS boots off of drive F: and my extra data drive shows up as C: –  Brian Knoblauch Sep 10 '10 at 15:07

I would say it depends on what you are doing with the system. If you may need to reinstall the os you might save yourself some hassle by putting all of your data on a separate partition. Otherwise I don't see the necessity any more. My two cents.

share|improve this answer

In general principle, I think segregating the default OS space (such as C:) from the Data (D:) is a good idea, but I would also recommend creating a smaller partition for Log files (L:) to keep them a little more secure and prevent some types of Denial-of-Service attacks.

Linux is very nice in that the file system remains hierarchically under one root directory no matter how many physical disks or virtual partitions you use. I would definitely partition the disk, but not necessarily for data vs. OS separation (since very often the two get mixed up anyway).

I would look at:

  1. Which subdirectories are likely to fill their disk and cause space problems for other directories (i.e. partition off /home and /var/log for example).
  2. Whether different parts of your directory structure need different file systems for performance reasons (i.e. XFS for stability, Ext3 for all-around use, etc.)
  3. Which directories might need to be expanded in the future -- these are good candidates for partitioning because you can simply rename the directory, partition and mount a new set of disk space to the directory location, and copy the data from the old to the new location.
share|improve this answer

Historical Linux (well, Unix really) partitioning recomendations are partly due to its origins as a (networked) mainframe server OS, which in turn i suspect was influenced by the (then) relative unreliability of hardware. For instance logs and temporary data were typically seperated because those storage areas got a lot of wear-and-tear, but it wasn't much of an issue if they were lost.

If you're building a desktop system, i'd go for the data/non-data/swap split. Unless you are building a server that's expecting to take serious abose, stuff like seperate /usr/local and /var/tmp just becomes a space allocation headache.

share|improve this answer
    
I'd say logs and temp were separate because they had the potential to be filled with crap from a rogue user - as the OS is usually a shared, multi-user environment. I'd not sure reliability was as much a factor. –  gbjbaanb Sep 10 '10 at 23:52

I'd say that its still nice to have - you have 100Gb of data (too much pr0n dude :) ) and you need to reinstall the OS (or, in keeping with Windows history, re-install it regularly to remove built-up cruft) then its very simple matter to keep it intact, than if it was on the C partition as well.

However, I'd say there is a problem there as Windows especially likes to stuff all kinds of stuff in directories on the C drive - its not just 'users' directory, but all the app data and various bits and pieces that end up stuck in ProgramData too.

Also, there is another factor - apart from the really big stuff (yup, that pr0n again) there are plenty of online backup tools (or local backup utilities) that perform continuous backups. Given these, its not such a priority to separate the data, as you can easily restore it from the backup location.

Personally, I try to split data + OS. I also try to put apps on a different partition too, so that my OS backups are a lot smaller.

share|improve this answer

I'll be the devil's advocate for a different school of thought.

Suppose for performance reasons, your vendor recommends that the OS partition isn't "sparse" and wants you to allocate the full OS partition upfront. This results in 10Gb to 20Gb (or more) of unused space on the SAN drive.

This is fine for a single VM, but chances are you will have several "performance critical" servers, each with its own 10 to 20Gb of whitespace overhead. In our environment this whitespace accounted for 20% of our SAN disk. Keep in mind that there are limits to which we should fill a SAN disk (but that is another story).

Management's had a choice

1) Absorb the 20% wasted space on the SAN, which is in addition to other requirements of "white space", and isolate any "full disk" scenario that might occur

2) Put everything on the C:\ drive and risk the drive filling up due to application logs.

What did they do?

Considering that Windows 2008R2 can dynamically expand the host OS's C:\ drive, and can expand the drive when full, management took the cost "savings" and reinvested it in monitoring tools like SCOM.

Now we are getting more than just simple protection of a C:\ drive filling, but we have a more complete systems monitoring in place to address other concerns before it happens.

share|improve this answer
    
Thin provisioned SAN volumes have a habit of becoming thick provisioned due to data churn over time, unless you have software running in the OS which communicates back to the SAN when a file has been deleted. So in a SAN environment, you're generally better off just allocating as much space to the boot drive as needed, and accepting that it'll all be used up in time. SAN does make it more complicated though, I will give you that! Perhaps you could have allocated a single SAN volume (depending on many other factors, such as disk performance and workload requirements) for both the C: and D:? –  Jeremy Apr 21 '12 at 22:37

Your Answer

 
discard

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

Not the answer you're looking for? Browse other questions tagged or ask your own question.