What's in a "File Server"?

April 7, 2008 3:43 PM

ITGeekFilter: My office needs a new (Windows) file server. My boss is kinda cheap (and won't give me a budget); why do "File Servers" sold as such cost so much?

Unfortunately, we have to stick with Windows due to some lightweight license management programs. Right now we're running an old Dell desktop with WinXP Pro (on a Celeron 2.4GHz and 256MB RAM) , and two 250GB hard drives that are nearing capacity. There are only six users on our network, and I think that our system runs really slowly. One problem is the software backup that is mirroring files from one hard drive to another - it runs daily and takes about 12h. (This is the first thing that has to go.) I'd like to add some sort of RAID implementation to improve speed (of backups and serving), but have a couple of questions:

1) Why are new file servers sold as such so powerful and expensive? Even the Dell machines that are billed to only serve flat files have a Quad-Core Xeon processor...

2) Are standalone NAS devices worth their salt? Even if I can find one with RAID?

3) Would it be worthwhile to find a RAID controller that will work in our current box and put more drives in it?

4) How is Windows Storage Server better?

posted by danbeckmann to computers & internet (25 comments total)

Ok, keep in mind with NTFS that you're going to need to keep your partitions to 2GB or under I think. (I'm a linux admin and not a Windows admin, so don't quote me on this. I've just helped my windows admin buddies build linux file servers and get 'em workin' with Active Domain, so...) Or you have to buy an operating system like (I THINK) Windows Storage Server that supports large partitions in a different type of filesystem. Hence the price.

The dell machines that are sold as such are built to scale to hundreds of users. You didn't say how many you need to scale to, but you're probably right that you don't need an actual brand new SERVER for it.

A decent approach would be to get a SCSI card for your current Celeron machine, and buy a SCSI drive array to plug into it. This would not cost you as much. We like AC&NC JetStor devices where I work.

One of my buddies is trying out OpenFiler ...
posted by SpecialK at 3:48 PM on April 7

You could look for something like this:

AC&NC JetStor Sata 405U ... I have no idea waht the price runs on it, but it can't be more than a few grand.
posted by SpecialK at 3:49 PM on April 7

what sort of "system" are you running?
posted by seanyboy at 3:53 PM on April 7

"Ok, keep in mind with NTFS that you're going to need to keep your partitions to 2GB or under I think." - SpecialK

This hasnt been true for years. The XP Pro computer I'm typing this on is running 3 x 500 gig drives, all formatted NTFS (single 500gig partition for each drive).

"A decent approach would be to get a SCSI card for your current Celeron machine..." - SpecialK

Please for the love of all things holy, DO NOT DO THIS. Celeron processors were never intended to run Windows Server. You WILL be disappointed if you do this. I can guarantee it.

Tell your boss to quit being such a cheapskate. Ask him how much the 2 x 250gig drives worth of data is worth. How much would it suck to lose that?.... A new robust Server is a small price to pay for long term payoffs.
posted by jmnugent at 3:56 PM on April 7 [1 favorite]

Also, get a new backup system. Twelve hours? Daily? That's a little bit insane. Even without any RAID, any backups you execute should be incremental - backing up only what's changed since the last update, which probably isn't much.
posted by Tomorrowful at 4:03 PM on April 7

Dell small system fileservers are actually quite cheap.

A basic small system with Sata Raid0, 256GB and 2GB RAM can be had for about $600. There may be minor issues installing your current o/s on it, but it can usually be done. (Windows 2000 and above).

Updating to 500GB Hard drives adds another $400 onto the price, but this is probably standard across all hardware setups. Getting rid of the raid on 500GB (not recommended) drops it down by $300

As for backup, you can knock this out onto an external USB Hard drive.
Depending on the files you're trying to incrementally back up, this could take no time at all to a couple of hours.

SCSI is fine, but can be expensive.
posted by seanyboy at 4:05 PM on April 7

Why are new file servers sold as such so powerful and expensive?

The cynical answer: because businesses can expense them. That goes into 1) -- BigCorp marketing people are very good at overselling, and their spec sheet is designed to assist that process.

The less cynical answer: file servers do have some specific demands (I/O, redundancy, reliability). Buying from a vendor also means buying support if/when something goes wrong. The BYO options (with decent, low-cost hardware RAID for hotswap) are pretty damn good these days. You could even look for a Linux-based file server with a virtualised Windows installation for the files that require license management.
posted by holgate at 4:06 PM on April 7

1.) Various reasons: Most likely because DELL mass produces systems, its more cost effective to standardize on certain hardware. Another reason is because building your server at the beginning with a quad-core Xeon makes it more "future-proof" (in other words, get the best processor you can, because the last thing you want to do 1year down the road is have to do the delicate open heart surgery of upgrading processors.) A lot of people buy servers saying "We only need to serve files", but then down the road decide to expand,etc and its easier to add more tasks to 1 server than buy a 2nd Server. Most OEM's know this, so its smart to put a strong CPU in to begin with.

2.) I dont have a lot of experience with Standalone NAS, however my understanding is that the benefit being you can expand the NAS a lot easier than you can add drives to a Server. Another advantage is it helps keep your operating system (Windows Server) and your data seperate. If you do go this route - please buy quality. It IS your data we're talking about.

3.) A Celeron with only 256meg of Memory is a poor, poor, poor choice for a Server. Please dont do this. No matter how you try to upgrade it, its still a (weak) workstation in Server clothing.

4.) I've never dealt with Windows Storage Server. So I have no advice there.

Look, I know its hard as a small company when you grow to the point where you start to need a Server. Its a big investment, and if you dont understand it, (or its not setup right) it doesnt seem worth it. At all the small companies I've consulted for, one of the biggest and hardest things to get them to do is change mindsets to buying quality instead of being cheap. But after doing so, almost without fail, they eventually ask me "Why didnt we do this sooner?".....

Do your homework. Find local consultants who might generate you a quote. Get several quotes (5 atleast ) and compare/contrast them to see if you can find some middle ground. When you buy the Server, DO NOT just buy the "bare minimum". Spend enough so you have some wiggle room to expand. You have to think of it as an investment. (because it is)
posted by jmnugent at 4:10 PM on April 7

Please for the love of all things holy, DO NOT DO THIS. Celeron processors were never intended to run Windows Server. You WILL be disappointed if you do this. I can guarantee it.

Dude, he's running Windows Server on it now. All he needs to do is attach more drive space. If his boss is cheap, his boss is cheap. Not everyone's working for a blue-chip company with a million dollar/year hardware budget. WinXP Pro handles NTFS just fine, so as long as you stay under 2TB per partition.

And I meant 2 TB, not 2 GB.
posted by SpecialK at 4:25 PM on April 7

(And by Windows Server I meant Windows XP Pro. Gah. Need to read before I hit post. Anyway, there's NO REASON he needs to do anything, if his boss wants to keep it cheap, than to add more storage via SCSI, SATA, or Firewire. I'm a fan of the latter so that when the machine bites the dust he can just plug it into another computer and have it work.)
posted by SpecialK at 4:27 PM on April 7

One of the reasons Server hardware is sometimes more expensive than PC hardware is the vendor makes a commitment to support the hardware with spares for a specific period of time (often three or five years). This makes administration easier for enterprise class businesses.

The I/O requirements of a server are different than that of a PC. However unless you are working with large data sets six people shouldn't really strain any kind of modern PC serving files. (If you're running a DB or Exchange or something other app that really racks up cycles all bets are off).

SpecialK writes "Ok, keep in mind with NTFS that you're going to need to keep your partitions to 2GB or under I think. (I'm a linux admin and not a Windows admin, so don't quote me on this."

Maximum partition size under NTFS without any tweaking is 16 TB non bootable and 2TB bootable. With tweaking you can run that up to a significant portion of a Terabyte even under XP.
posted by Mitheral at 4:30 PM on April 7

And I see you corrected that and I made the same mistake. You can run the file system up to a significant portion of a Petabyte with tweaking.
posted by Mitheral at 4:32 PM on April 7

"..there's NO REASON he needs to do anything..."

"I'm a linux admin and not a Windows admin..." - SpecialK

No offense, but how many Windows Servers have you setup? If we were talking Linux, I'd be more inclined to agree with your assessment because Linux hardware requirements are more forgiving. However, I've spent the last 10 years doing Windows Sysadmin consulting, 3 of which were being the IT admin for a (cheap) school district (no, not a "million $$ blue-chip company") where I was responsible for 10 to 15 Windows 2003 Servers. So I'm confident I know what I'm talking about.

Its true, with only 6 users, you dont need some super-duper $10,000 Server where every single component is redundant. I also realize cost is a major factor. That being said though, going cheap on a decision like this is a bad idea. Going cheap WILL come back to bite you.
posted by jmnugent at 4:40 PM on April 7

This question is a little light on the details of what is actually WRONG with the current server, so I think you need to get that clear in your head first...

If the only problem with the server is that you are running out of space, then you probably don't need a new server... Go out and buy a RAID controller card and a couple of big hard disks for it and set it up... RAID will also solve your backup problem, because you can set it to do mirroring, which should mean less of an issue with HD failure (but note that RAID mirroring doesn't solve the problem of the boss accidentally saving over a file or something, which I've found is actually the more common use for backup in smaller organisations!).

Of course, eventually you will need a bigger server machine. There is a lot of bulldust around why "servers" (as sold by the big vendors) are better than workstations, but it essentially comes down to build quality and future proofing.. Most servers you buy from Dell etc are built to run 24 hours without any breaks.. They say that this means higher quality components etc, and I guess also gives you some support if it does break down (you ring Dell).. A lot of these machines are also built to be upgraded as the business grows. So they usually have space for more hard disks, memory etc, maybe even a spot for a redundant power supply! It's up to you whether you buy this type of machine, but the future proofing argument usually works ok with the boss...
posted by ranglin at 4:43 PM on April 7

Just add a couple more hard drives. You should be able to add at least 2 more, or even replace your current drives with larger ones.

Contrary to what other people have stated, requirements for a file server are extremely low. Much lower than a workstation.

Mirroring 250gb should not take 12 hours. How are you backing up the data? Compression can slow it down, if you have it on, just turn it off. Drive space is cheap.
posted by mphuie at 4:53 PM on April 7

1. They don't make their money from selling you old hardware at extremely low margins so you can scrape by.

2. I looked at these pretty hard, but for the cost and inflexibility I couldn't justify it.

3. Yes, this is what I would do (to start) in your situation. I would probably buy a hardware RAID controller, 2 500GB SATA drives, and upgrade to at least 512MB RAM, if not 1GB. Then run the 2 250s in a RAID 1 and the 2 500s in a RAID 1. RAID should not take the place of backups, but if you're skimpin it then you gotta make due.

Also I would run PerfMon over the course of a couple typical days, with the following counters, to see where your bottleneck is. It's probably Disk IO, but if it's CPU you might need to look at upgrading that as well.

# Disk\Current Disk Queue Length
# Disk\Disk Reads and Writes/Sec
# Memory\Available Mbytes
# Memory\Pages/sec
# Page File\% Utilization
# Processor\% Processor Time
# Processor\Interrupts/sec
# System\Processor Queue Length

For the money I would say this is your best bet without knowing your exact situation.

4. If you're not sure if you need it, you don't.

I've been running XP Pro on a 2.53 Celeron, cheap MoBo, 1GB RAM, old 20GB IDE Boot Partition, 1.5TB RAID 5 (4x500GB SATA 3) Storage Partition (running off a RocketRaid 2310 RAID controller), for almost 2 years now. It was a domain controller doing DHCP and serving movies and files to a house of 35 guys, but now just sits in my corner running BitTorrent and streaming movies to my XBOX. I couldn't be happier with my choice. The only issue I had was heat (5 HDs) so I added a front fan and now it runs ~30 C.
posted by Kupo? at 5:52 PM on April 7

RAID will also solve your backup problem, because you can set it to do mirroring,

For the love of god, no. RAID is NOT a backup solution, rather, its a method for minimizing downtime due to a specific kind of failure (usually, a disk failure). Tape is much, much more reliable in the long run, which is why it's used as a backup medium.

1) Server grade hardware is supposed to mean higher MTBF. Its designed with cooling in mind and has redundant fans. Its designed with easily serviceable parts. It has lights out management (maybe) and monitoring of hardware. Power supplies are often redundant. Hardware is a commodity, and you get what you pay for. You'll easily spend more on software licensing, so get over it.

2) Standalone NAS is great. Its a plug and play type of deal. You don't use it for anything except file sharing. They'll integrate into Active Directory nicely. The Windows powered ones support all the nifty Windows Server features like DFS, versioning of files, etc. RAID is a must, but again, RAID is not a substitute for backups. When you lose multiple drives and are looking at several thousand dollars for data recovery, you'll be glad you have tapes. You get what you pay for here, too. Many of these devices use software RAID, which can be more difficult to recover in case of a failure.

3) I would say no, for the reasons above Downtime is far more expensive than a one time outlay for hardware, so bite the bullet and get quality gear.
posted by kableh at 6:45 PM on April 7

Another vote for standalone NAS here. We use the Netgear ReadyNas NV+ at work and it's pretty easy to use. You can integrate with Active Directory or just set up in a workgroup (a slight assumption with only 6 users). Once it's up, it just works and life is good.
The ReadyNas nv+ is $1000 for 1tb with built in raid5 and once you have it configured, it just works- comes to somewhere around 600megs with the ability to upgrade the HDs to 2tb storage. There are other very viable NAS solutions, but that's just one example.

To add onto kableh's point, the point about Raid not being a backup solution is absolutely crucial, and another reason I like NAS. Most appliances I looked at have a built-in USB or Firewire port with the assumption you'll be using it to attach an external HD to backup the NAS to external HD, which is at a minimum what you ought to be doing.
posted by jmd82 at 7:40 PM on April 7

Right now we're running an old Dell desktop with WinXP Pro (on a Celeron 2.4GHz and 256MB RAM) , and two 250GB hard drives that are nearing capacity.

256mb of ram isn't enough for XP, that's the reason it runs slow. Bumb it up to 1 or 2 GB, and see how your performance changes.

Next, remember that RAID isn't a backup solution. Depending on the implementation, RAID can improve performance, and it can improve uptime (ya, it improves reliability too, but only sort of). To improve performance without sacrificing reliability, you have to go to a fairly complicated RAID implementation (like RAID 0+1). A RAID controller card or a NAS can handle that just fine, but I don't think Windows software raid will do it.

Finally, backing up from one always on hard drive to another always on hard drive is a pretty terrible backup solution. That single backup drive can fail just as easily as the main hard drive. Also, what if you want a file that some klutz accidentally deleted two months ago? At the very least, you need to have two different backup drives that get swapped back and fourth, and which get stored in a drawer when not in use (at least one drive is disconnected and powered down at any one time). Really though, you want to take snapshots of your filesystem offline once in a while.

The real question you have to answer is how much downtime can you suffer without significant losses. If shutting down the server over a weekend is no big deal, you should probably keep going with the bargain basement hardware you've got. Upgrade the memory, stick a 750GB drive in there, make sure you've got USB 2.0, get a couple of external USB drives, and etc.. It is redundant power supplies as well as hot swap and hot spare hard drives that start costing the big money in file server hardware..
(the other thing you have to do is fix your backup solution.. I'm not up to date on modern backup procedures, but it seems to me you should be periodically retiring drives with full copies of your data on them, and storing those drives off site)
posted by Chuckles at 8:37 PM on April 7

For the love of god, no. RAID is NOT a backup solution, rather, its a method for minimizing downtime due to a specific kind of failure (usually, a disk failure). Tape is much, much more reliable in the long run, which is why it's used as a backup medium.

Here is an old question about backup. My question/observation there is still on the table.. The thing that makes tape more reliable isn't its tape-y-ness, it is the fact that the tape sits in a drawer when not in use. Not powered, not wearing out, possibly even off site where theft and fire can't hurt it. Hard drives can be used in this way too. I expect HDs are perfectly acceptable alternatives to tape if you do it right.

On the other hand, I totally agree about RAID and backups.. Just this past Christmas I had two drives fail at once in my 4x300GB RAID-5 array full of media files. I didn't lose anything critical, because that is on a different file system which gets backed up fairly regularly, but I did lose lots of stuff, and it really sucks!
posted by Chuckles at 8:58 PM on April 7

Hard drives are currently selling for well under thirty cents a gigabyte. That makes them cheaper per byte than all but the crappiest tape cartridges. Plus, a backup on an external hard drive can take the form of a set of ordinary files, which means you don't need to fool with specialized backup software to get at your backups. Plus, USB2 is everywhere, and you can be pretty confident that its successor will be backward-compatible, so even when your old tape drive blows up and you can't buy a compatible one any more, you can still read your archived disk backups. Plus, hard drives have fewer completely show-stopping failure modes than tape (a snarl of tape wound up inside a drive is a truly ugly thing).

I can see no reason at all to prefer an offsite, offline tape cartridge to an offsite, offline USB2 hard disk.

The main thing, though, is never, never, never have only one backup. Personally, I like regularly-switched, offsite external hard disks for rapid disaster recovery, and online backup (e.g. Jungle Disk) for everything else.
posted by flabdablet at 1:45 AM on April 8

The NAS suggestions are the lowest-pain option (I have a ReadyNAS box at home that I lurve) but they're not the lowest-buck option.

Here's what I would do. In fact it's what I have done.

Beg borrow or steal a new machine, keep the old one running just as it is. Install as many hard drives as will fit in the machine or your budget, whichever comes first.

Now install Linux. Come back here, I'm not done. Set up your hard drives as software RAID under LVM with a RAID-1 boot partition across all drives. Newer Ubuntu installers have made this much less painful, but if you decide to skip LVM to make the learning curve a bit shallower you can. Install Samba and copy all the files off of your old machine to the new machine. Mount the new machine's exported share on the old machine and point your license manager at that location. This is the sticking point, if the windows program balks at the managed files being on a network share skip ahead to the very end of this post* or just ditch it all and follow someone else's advice.

Now install windows in a vmware guest, bridged networking and install your license management software in it. Mount the new share in the same way as it is on the old machine and turn the old machine off. Point your clients at the vmware guest's name/IP and go for a beer. The underlying linux machine will keep serving files even if you have to reboot the windows instance, you'll be able to get away with muuuuch lower hardware requirements for the fileserving part of the show, and you can migrate much more easily to new hardware by simply moving the disks and using LVM to expand the appropriate volumes.

*) If the windows software insists on mounting a physical drive, you can export an LVM lv as the physical boot drive for a vmware instance. You don't really gain anything by doing this vs just buying a new machine because you're still using Windows as a fileserver (which is silly) but it will work.

Also be careful just dropping in a RAID card without having good on-site backups. If the raid card itself fails you usually can't just pop down to best buy to get a replacement and many surprisingly expensive hw raid cards are excruciatingly specific as to how similar a replacement card must be to be able to mount the existing drives (i.e. same exact hardware down the the firmware(!) version). Under high throughput most HW cards are actually slower than SW raid on an otherwise unloaded machine anyway.
posted by Skorgu at 9:13 AM on April 8

Skorgu: Good ldeas, but why not do it much simpler?

FreeNAS.

Stupidly simple to set up.
Free (as in beer).
All standard file server services available through web interface config.
Built-in rsync and software RAID.

That would be my suggestion for a cheap way to get better performance. I use mine as a storage repository for my movies installed on an old, crappy Celeron with 384MB of RAM, and we can have three computers playing from it simultaneously with no frame drops.

FWIW.
posted by liquado at 9:58 AM on April 8

liquado: the OP sayeth: Unfortunately, we have to stick with Windows due to some lightweight license management programs. You could have a FreeNAS server box and a second license management server, functionally equivalent to my vmware hack but with slightly larger surface area to fail.
posted by Skorgu at 12:49 PM on April 8

A lot of bad advice mixed in with the good on this thread...

To answer your question(s).

1) Why are big servers so much more expensive? Because they are using high end hardware that costs more to produce; and more of it. ECC memory, SCSI drives, RAID cards, tape backups, Xeon processors, etc. Better cooling because they need to be able to live in a rack with other servers. Because they often have tons of management and monitoring software and hardware that comes with them. Because they often have longer and better warranties than desktops. It's worth the price if you need that kind of stuff.

2) Sure, if it's enterprise grade. Or at least pro-sumer grade. A little thing meant for occasional home use will probably overheat under heavy use.

3) Maybe, if your bottleneck is hard drive performance. It probably isn't.

4) No clue.


* There's nothing wrong with a Celeron processor. IF the applications you are running on the server are not too dependent on L2 cache. And they probably aren't. All a Celeron is is an older Pentium processor with less cache.

* Unless you buy an appliance that is Linux based and turn-key, don't bother with it. It'll be a management headache, unless you are already an experienced Linux user.

* First think I'd do is max out the memory in your current server, and see how that goes. If that works, just buy a couple of new 1TB drives and copy the existing data over to them.

* Actually, the first thing I'd do is figure out where the performance bottleneck is occurring. It's an almost sure bet that memory is the problem, but you will have to do a little analysis to make sure. Look at swapfile usage, processor usage %, network utilization. If your network card driver has it, make sure the thing isn't receiving or sending bad packets. Maybe the server is sitting there doing nothing because you have a network problem?

PM me if you want...
posted by gjc at 6:41 PM on April 8