Building a Budget Storage Server

An anonymous reader noted an article running over at Firingsquad talking about building a budget storage server. Talks about cooling, power, RAID, expandability, etc. Good overview type article, with practical application.

Comments

a tip

[flok]

A tip which I don't see mentioned very often: when using multiple drives in a raid-array, use drives from different batches. Or even better: from different vendors. Why? You don't want your complete raid-array failing because to much drives fail because of a common problem in their hardware/firmware.
Ok, chances on that might be slim, but in my opinion you're better safe then sorry.

Re:a tip

[supersmike]

Interesting tip. I didn't know that. At the same time, 3Ware recommends using identical drives if you want maximum performace for reads on RAID1.

Re:a tip

[pmz]

You don't want your complete raid-array failing because to much drives fail because of a common problem in their hardware/firmware.

Also, you don't want drives failing due to unpredictable failures of unmatched drives failing to interoperate.

If there were truly a statistical benefit to mixing drives like you say, I would have thought the analysts and Sun, EMC, and IBM would have adopted this strategy by now. Or have they?

Why is it that Sun's drive model numbers are also specific to a firmware revision? Why are arrays sold with matched drives and why are patches offered to upgrade firmwares to know revisions?

How is it even possible to integration test sets of unmatched drives and have any notion of the long-term MTBF of drives with firmwares who have never met before?

Re:a tip

[supersmike]

What constitutes an identical model? Is it number of cylinders and heads? I don't know much about hardware, but I'd love to try this if I could be sure I would still get that read performance they talk about when using identical drives. I seem to recall they even said from the same manufacturer. Probably because they assume I'll screw it up otherwise. =:-o

Re:a tip

[Otter]

What constitutes an identical model?...I seem to recall they even said from the same manufacturer.

I assume he's saying to buy one Seagate UltraATA 160 Gb from one vendor and another one from a second vendor. Certainly "identical model" implies the same manufacturer.

Re:a tip

[flok]

Also, you don't want drives failing due to unpredictable failures of unmatched drives failing to interoperate.

True, but what if they're all on a different channel of your controller. Problem solved, isn't it?

If there were truly a statistical benefit to mixing drives like you say, I would have thought the analysts and Sun, EMC, and IBM would have adopted this strategy by now. Or have they?

Maybe nobody analyzed it. But nevertheless, you have some good points.

Re:a tip

[bconway]

This is extremely poor advice to give, and I hope no one takes your word on this and jumps in the deep end. There are a host of unforeseen problems that can arise from using unmatched drives. This is not what the array manufacturers designed for, and they even warn against it in their documentation.

Re:a tip

[mindstrm]

head tracking.

ITs' not just identical numbers of cylinders and heads, but identical mappings of those to virtual cylinders/heads that you see in IDE.

No, it won't screw anything up, but performance will suffer.

The idea is that all heads can move in parallel with some types of raid. you don't want one drive writing it's block to have to move farther than the other one..

As for 3ware, I have a 12 port 3ware controller with 12 120 gig drives in it... and though its' tough for me to judge performance, as I use it chiefly over 100mbps ethernet, I will say it never seems to have hte IDE blocking problems when moving around large files that you see with normal IDE (which makes sense, considering the OS is talking to a real controller, which is taking care of all the IDE stuff behind the scenes.)

I don't think the performance is the same as a high performance scsi setup, but it certainly is good. and cheap.

Re:a tip

[kfg]

Performance and data security are often at odds. So much so and so often that I think we can nearly take it as a given that increasing one decreases the other. It is the nature of the beast.

One must ask one's self, "How fast do I want my corrupted data delivered?"

KFG

Re:a tip

[fatdave]

We had ca 25% of our deathstars fail inside two years. Out of about 30 drives we have had to change 7 or 8 due to failure. That is a failure rate we don't see with other brands.

Re:a tip

[pmz]

It doesn't mean anything.

For an organization with a very large number of deployed drives, it helps them estimate the number of spares they need for a given amount of time.

Re:a tip

[Macgruder]

Man, is it the day for everyone to switch off their brains?

He said different batches and different vendors. Not different models.

Use the same model all around, but buy them from different vendors (such as CDW, NewEgg, etc.) That way the chances of having a batch failure is minimized.

Re:a tip

[pmz]

We had ca 25% of our deathstars fail inside two years.

Enterprise-class drives come with a five-year warranty. A truly bad batch of drives should be purged by then, leaving the manufacturer to pick up the tab for replacement. Also, you cite having to replace a drive every two to three months, which is a pain but not as bad as patching Windows.

Re:a tip

[karnal]

I can actually back you up on this with real world experience.

Just for grins (since my older motherboard supported it), I had a 7200rpm maxtor 30gb. Thought, hmmm, can do raid 0 - and get better performance.

Bought a 7200rpm seagate -- performance dropped through the floor. Why? Well, depending on where the data was, the seagate would have to reposition the head while the maxtor was still reading the same track....

Finally bought a similar maxtor to replace the seagate, and my performance did increase. Not by any amazing amount above the norm, but it wasn't dog slow anymore on reads and writes.

Re:a tip

[k12linux]

While we're handing out tips, here is one I learned the hard way. Create your RAID paritions at the lower side of the "specified" drive capacity. In other words if your new 180Gb drives actually have 180.5Gb available, DON'T use the extra .5Gb!

I had to replace a failed 180.4Gb drive on a 1Tb server and the replacement was exactly 180Gb. I had to back up 400+Gb of data, re-create the RAID array with 180Gb partitions and then restore. If you think backing up 60Gb is slow... ha!

Unfortunately, the 3ware utilities don't seem to allow you to specify the partition size.. they just use the whole drive. Mixing one 180Gb drive in with the 180.4Gb drives made it use 180Gb for all of them. Unfortunately that isn't very practical when you are creating a raid array on a batch of brand new drives. (You'd have to find one slightly smaller drive.)

Re:article text

[Dav3K]

You were worried about this article being slashdotted??

Whoa...

[eurleif]

Their definition of "budget" is $3,140? Someone give me their budget right now!

Re:Whoa...

[AchmedHabib]

I've got a $25000 mail server
Yes, but would you want to replace that server and all the others with a budget server?

I mean, the people that bitch about the hardware is to expensive, are the ones that would not understand that you have to bring the entire server down to replace a harddrive. and who's "spare/family time" is going to be allocated for the job? :)
However for a near-line backup system it could work, or any other system where you can afford it to have downtime during business hours.

Besides from that, they seem to fall between 2 chairs here, whats with the geforce card for a storage server.

Re:Whoa...

[Anonymous Coward]

"Besides from that, they seem to fall between 2 chairs here, whats with the geforce card for a storage server."

I think they missed the point, they went budget, then went picky with things like expensive keyboard and not using onboard video because they really needed a DVI input. Lame.

I could save $ on that stuff alone.

Keyboard and mouse combo are beyond needs, gimme a $5 keyboard and $5 mouse. A display requiring DVI is probably LCD, we'll say he wants a $250 display (to be cheap to say the least) and %70 for the excess video card. I'll go with onboard video and a cheap $50 monitor (hell I can save on that by just switching one from another PC) for.

I saved $300 there at least.

It's a storage server, you're not supposed to connect to it ever once it's started up. And what's with the statement that it shouldn't run Linux? I couldn't think of a better OS to run a server on under a budget. Why spend bulk cash to pay for an operating system on a machine that's just sitting there? His concern is USB/Firewire compatibility to access old data as it's moved out of the machine... uh, I have no intention on ever bringing my storage servers down after they're up, especially to do a dirty op in removing a storage disk. And what's with DVD writers for backup? You're storing 1TB of data on the machine and you're going to back it up 5 GB at a time? Please connect to a tape library if you're going to back up that much crucial data PER YEAR (as stated the drives should fill in a year).

This is not a good article for any of it's goals.

Re:Whoa...

[RevDobbs]

Firing Squad is a gaming site, and their goal is to also make it a workstation, and a 1337 one at that. As such, they dumped money into silly things like LED displays, rounded cables, an uber kewl case, graphics card, DVD burner, etc., all of which upped the price with out improving the performance that one would look for in a "storage server".

First, the last thing I would want is someone doing daily work on such an important computer. I don't need Word or IE to crash & burn and eat the OS with it, rendering everybody else's work unavailable to them.

Second, no comments are made about the OS, or even about what the performance of their "Budget Storage Server" will be. It seems odd that they so blithely dismissed RAID, even though the performance and reliability improvements are known.

Finally, there's no mention of tape backup, pretty much the quickest and cheapest way to archive tons of data; why collect a terrabyte worth of information and leave it to the reliability of the HD?

Based on the title, I was hoping the article would be about a file server with high performance-to-cost ratio, something like integrating a BSD/Samba machine into an mixed NT, MAC, & UNIX enviornment, with some concreate numbers regarding sustained data transfer or multiple users in an office workspace. Instead, it was an article about stuffing 5 HDs into a gaming rig. BFD; I find the case mod articles to be more interesting than that.

Self?

[jargoone]

CDs may self-destruct at sustained speeds of greater than 56x

The author (or the person who wrote the sidebar comment) needs to learn the meaning of self-destruct...

Re:Self?

[swordboy]

As a side note, engineers *never* use the term "self-destruct" in a technical report. The same goes for "explode" and other synonyms. The correct term is:

"Spontaneously Disassemble"

If your laughing, make note that I am being completely serious. I've seen this term on paper too many times.

Last time I checked

[bigjnsa500]

Last time I checked, motherboards only come with 2 IDE channels. According to the article, they are using 4 250gig IDE in 'standard' configuration (ie, no RAID or SATA) and using a system drive. Uh, 4+1 don't add up to 4.

Since they couldn't afford RAID, what about software RAID? Way faster than normal IDE operations.

Re:Last time I checked

[mindstrm]

Lots of motherboards come with 4 IDE channels now, and onboard IDE raid. Very common, not expensive.

Re:Last time I checked

[Space cowboy]

I've never trusted software RAID on a multi-CPU (ie: one of our typical servers) system. I've had the raid screw up far too often for co-incidence. I'm pretty sure there's a race condition in there somewhere. I've had a server run for 5 months with no problems, then suddenly I get an SMS that a node is down. Bike over to the co-lo, and the filesystem's completely screwed... Never again. Hardware raid all the time :-)

Simon.

Re:Last time I checked

[Zapman]

In principle, I agree with you. HW Raid (if you can get/afford it) is the way to go. Dead drive? Yank it, and pop in another. Automatic resync. No down time.

SW Raid however, isn't (always) as bad as you make it out to be, at least under solaris. Their 'disksuite' product has been very reliable here. Every server has it, and we've yet to see any real problems with it. Great for mirroring the OS drives.

Before version 4, disksuite had some teething problems, but the 4.2.1 release rocks.

Re:Last time I checked

[antifun]

Been using Linux SW raid in the 2.4 kernel series for a year+ now and it has worked like a champ, with both IDE and SCSI devices. All disk servers were SMP (overkill but management wanted it that way). Dunno what you screwed up.

If your criteria for an adequate disk server include either (a) high performance or (b) long-term maintainability, then you should choose SW raid.

Most HW raid systems, especially cheapo PCI cards, but even expensive Fibre Channel-SCSI3 rackmount monsters, offer either extremely primitive performance metrics or none at all. With SW raid, you normally get the full performance-monitoring and tuning capabilities of the host OS. Big win. You will also get better performance from a SW raid, given the same drive layout, and as long as you do not use the box for anything else at the same time. It should be obvious but some people don't believe this.

The other big win is more important when you spend more money than $3000 (a pittance in this market): there's no hardware manufacturer to get bought, go out of business, or change product lines. No multi-thousand-dollar support contract or custom software to configure the RAID or any of that other crap. Trust me, when your dedicated RAID box's motherboard flakes on you and you discover the manufacturer has gone out of business, you'll be cursing yourself for choosing HW raid every time you search Ebay for a replacement part.

Not to mention that commodity, general-purpose HW is always cheaper to replace, and its performance/price ratio grows much faster than special purpose HW. The HW raid system with the 200MHz i760 and 64MB RAM might have looked great in 2000 but now you're stuck with the proprietary on-disk format of an out-of-business vendor with no way out except to build a new system of the same capacity and copy everything over. (In the case of large data warehouses, "full backups" don't usually happen.)

HW raid was compelling in the past. Now, with commodity hardware so cheap, and open, stable SW raid systems floating around, you'd be a fool not to prefer them in many situations. If you want a fire-and-forget dedicated box, go for it. But be ready for the "forget" part in a year or two.

Re:Last time I checked

[stmfreak]

Never again. Hardware raid all the time :-)

Riiiiight, because that hardware RAID doesn't have any of that untrustworthy software in it. No bugs there. Move along.

Re:Last time I checked

[Lumpy]

and the other funny part is that they talk that reliability is important and yet dismiss SCSI right away.

please find me IDE drives with a 5 year warrenty. Or server class IDE drives.

I can't. I tried. and I decided that for our "cheap" server we use U160 drives off a 29160 scsi card and use software raid 5 on a linux box running samba.

I came in less than they did, yes I have less storage but I know that my drives will still be spinning and running happily in 2007 you can't say that for today's IDE drives. I also added a DLT7 drive (anyone that spec's a server WITHOUT a backup solution is a hack.) to back things up daily.

IDE is great for consumer class stuff. I would NEVER EVER trust critical business data to any server running IDE drives and without a good backup system like a DLT drive.

Re:Last time I checked

[Ledskof]

Linux Software Raid works nice. You can install the OS on it too. Or at least I have, MANY times.

I've never had any problems with it. I have a production server that has been running it for a year. I have a ResierFS MDRaid5 volume for storage. Some of the other stuff is raid0 and raid1, depending on my attitude for different mount points at the time. No problems with any of it on the storage level, ever. Even when I had to swap a bad drive out. The rebuild went faster than the hardware rebuilds I've had to deal with too.

So far my experience with software raid is that it is so configurable and flexible that hardware raid is just too asphyxiating. Oh yeah that plus I've never experienced a stability issue with it.

Plus, there's that beautiful EVMS package that makes volume management really fast and easy.

Griping about software raid should be specific to operating system, application, and method, etc.

Ask the HP-UX guys how solid software raid is for them.

Did I miss something ?

[tmark]

In an article about building your own storage server, why are they spending so much time talking about irrelevant things like *video card's 3-d performance* (128 MB in a storage server ??), mouse and keyboard choice, and yet fail to even so much as mention (as far as I could tell) OS choice or software ?

Re:Did I miss something ?

[23_Elders]

I agree, that seemed much more like, "Watch us build an expensive PC with a lot of hard disks" than "Watch us build something useful for reliable network storage."

I am currently trying to put together a RAID 5 file server and they do not cover any topic of use to me in that article. For example, practical backup solution? They chose a DVD burner, why that over similar tape solutions? I would guess price, but it would be nice if they at least mentioned some of their considerations. Especially since it would take 112 DVD-R's to back up a terrabyte?

Also, aside from their DVD backups, they seem to have no data recovery plan in case a hard drive fails. I guess they aren't storing anything important on these drives?

Re:Did I miss something ?

[ashkar]

There seem to be many similar problems in this article.

• A file server would be better off without a floppy or dvd drives. A cdrom drive would suffice and provide a small security benefit as well.
• A second ethernet port is definately needed in a production machine such as this, but why connect it to the internet. It seems that that would be unnecessary and possibly even dangerous firewall or no.
• His suggestions for memory and cpu are probably way too high. The gig of memory I guess I can understand, but even eight ATA drives shouldn't require a 3.0 P4. Someone correct me if I'm wrong here. I've never had more than four drives going at a time.

This article should have been passed around a bit more before publication and maybe these errors would have been fixed.

Re:Did I miss something ?

[jmt9581]

They did mention OS choice briefly:

This is also the reason Linux was not a good choice for our system -- it doesn't make sense to put XFS/ext3/ReiserFS drives into a USB2.0/Firewire external box.

After skimming the article, I have some questions:

• Why does it not make sense to put a journaling filesystem in an external box?
• Why not just use ext2 if they don't want a journaling filesystem?
• Do they mention their choice of OS anywhere else in the article?

To me, this read more like an advertisement for some of the latest and greatest in computer technology than a real article on building a 1TB fileserver from commodity parts. For example:

"For this application we'll use an NVIDIA GeForce FX 5200 128MB card to stay within budget. No one should mistake this card for a 3D powerhouse though; it is among the slowest 3D cards on the market -- fast enough for us, but slow compared to the other NVIDIA and ATI offerings. Most "servers" do with an integrated graphics chip like an ATI Rage XL or less.
Cost: $70 with free t-shirt or hat"

Why the heck would you need a fileserver to support "at least 1024x768 resolution" and have "good drivers?" Are you actually going to be using it as a workstation?

Re:Did I miss something ?

[tlk nnr]

Yup.
And they build a storage server without ECC memory. We had that - by chance we noticed that a memory chip generated single bit corruptions. We were busy for days comparing every file with older backups. And that was around 50 GB data, not 1 TB.

Re:Did I miss something ?

[Jeff DeMaagd]

You are right. They are probably unbalanced gaming twits. Ignore them.

Few to no real servers actually need 3D, and 8MB is often judged to be plenty enough if you look at the boards designed for server use.

The only exception is if these people are making their every day system into a server, which may not be advisable for anything.

Re:Did I miss something ?

[kiwimate]

In an article about building your own storage server, why are they spending so much time talking about irrelevant things like *video card's 3-d performance* (128 MB in a storage server ??), mouse and keyboard choice, and yet fail to even so much as mention (as far as I could tell) OS choice or software ?

Look at the banner on the pages -- "Home of the Hardcore Gamer". It's because they're gamers and know everything about tuning a system for games, but don't know the first thing about building a server. What they've ended up with is a mish-mash that won't serve any particular purpose well, except possibly as a rather decent PC for a secretary (except no secretary would want something that big at her desk).

As one reads through the article, what leaps out is that they're most comfortable when debating relative merits of 3D video cards and building uber-fancy custom machines designed for gaming excellence. Good for them, but this is far removed from building a server.

It's got a terabyte of utterly unsafe storage. No RAID, no nothing.

It's got a video card which is overkill for a server but which they disdain as a low-end 3D graphics card.

They've got one hard drive for the system and everything else as data, so they're not building a "high performance" system or else they'd have a separate drive for paging.

They haven't discussed the types of files they'll be storing at all -- will they be tiny text files, medium sized spreadsheets and documents, or massively large presentations and CAD files? This affects how you configure your system.

Their approach to planning for hardware failure is "we bought the better quality stuff so we don't have to worry so much about MTBF". No need for RAID or redundant power supplies. (Although oddly enough they've chucked in two NICs.)

Did I mention no RAID? Yet they've bought a 3D graphics card (overkill), a nice mouse (in case they want to do graphics editing or perform fast wrist actions on their storage server), a wireless keyboard, and a fun little LED display to tell them how fast the CPU fan is spinning.

Look at how they're future-proofing the system, by the way. They anticipate going through 2 TB of data every year. So every six months they're going to pull out the existing 1 TB of storage, plop into an external array, and put in a new set of disks. I wonder how long this system is supposed to last...

All in all a very odd system indeed. In fact, a pseudo-server built by gamers with no understanding of how to build a server.

Re:Did I miss something ?

[Stephen Samuel]

agreed. terrible "review". out of touch with its "budget" title. no mention of OS/software at all.

Yeah. They did mention something about not using Linux because they didn't want ext3fs or reiserfs on their disks (or something like that). ah, shoot.. .here's the quote.

This is also the reason Linux was not a good choice for our system -- it doesn't make sense to put XFS/ext3/ReiserFS drives into a USB2.0/Firewire external box.

So it sounds vaguely to me like they've got 4x250MB drives with no RAID, an ill thought out backup system and probably running Wintendos because it's a more stable game^H^H^H^Hserver operating system... Then they wrote this article to justify it.

Re:Did I miss something ?

[guacamolefoo]

I agree, that seemed much more like, "Watch us build an expensive PC with a lot of hard disks" than "Watch us build something useful for reliable network storage."

My solution to building a cheap storage system was the following:

1. Buy old Netfinity 5000 on eBay.
2. Order 5 x 9GB SCSI drives from my trusty IBM parts guy (csaunders at itexchange.com)for $70 each.
3. Order basic RAID card for said box.
4. Install RedHat 7.1 from a CD in a book under my couch.
5. Install SAMBA
6. Run cron job to back up user data and relevant config files to an external USB hard drive attached to a windows box on the lan.
7. Take external hard drive to safe deposit box weekly. Get second USB drive out of safe deposit box and attach it to windows box at office to await next update. FWIW, I've been thinking about putting the USB drive that is in the office in a safe when the back up is not taking place. This is not for fear of fire or catastrophe -- I just don't want it to walk out the door.
8. The Netfinity server has the RAID 5 array configured for a hot spare drive so that there is failover operation if a drive quits.
9. Installed PowerChute software with a UPS to shutdown the box gracefully if power quits.

External USB -- $100 each (2) = $200 (got enclosures and cheap-o spare IDE hard drives from scavenged boxen)
SCSI Drives -- $70 each (5) = $350
Netfinity box = $300
UPS = $200 (I think)
Redhat 7.1 on CD in book under sofa = priceless

Total: $1,050.

Project results:
RAID-5 with regular offsite storage. Logical disk size is only 27 GB, but you can fatten this by using bigger SCSI drives. I didn't need mondo storage, so I saw no need to go with 36 GB drives, though you certainly could if you had more money.

I am currently trying to put together a RAID 5 file server and they do not cover any topic of use to me in that article. For example, practical backup solution?

External USB drives worked for me. Depends on how heavy-duty you need and how your office works. Perhaps simply connecting up two servers in different offices and doing mutual backups nightly for changed files might suffice. DVDs and CDs are an option, and tape is still useful.

Also, aside from their DVD backups, they seem to have no data recovery plan in case a hard drive fails. I guess they aren't storing anything important on these drives?

My data recovery plan (if everything pukes) is to buy a new chassis and drives and reinstall RH 7.1., connect it to the lan, and download old config files and user data. I think it would take a couple of days (mostly waiting on delivery of the drives and box). That time could be slashed if I were truly paranoid if I simply kept spare parts off-site. I'm just not that worried, however.

FWIW, our office is a small lawyer's office with about 10 people on our LAN. The data we need to store is not huge.

GF.

Insert standard joke..

[rf0]

Finally a place I can store all my p0rn/warez/dvds. But seriously why did they put in a 3D Graphics cards on a server. Surely any cheap AGP card even without 3D will do. Some basic ATI's are just $20

Rus

A fileserver is great in my home

[digitalgimpus]

I took a Beige G3/266MHz that I got for $50... put a 120 GB WD drive, ACARD IDE Controller, and Mac OS X.

An extra fan, to keep it nice and cool, and a 10/100 NIC.

Runs rather well. Smooth, reliable, and fast. For a very low cost. Mac OS X 10.2 comes with AppleShare, for Macs, and Samba for windows file sharing. Apache for a webserver, and PHP, Perl...mySQL.

You got whatever you really need.

I added webmin, for remote control. Makes it a bit easier.

Lame

[sardonic2]

Article is lame when it comes to the important stuff. Its great he gave us the hardware to do it, but thats not the important part now is it? Software.... something that can do backup's to harddrive and then take backups and archive on tape. we went with tapeware because of price, but we cannot archive a current backup to tape, so that means we have 4 week online and no archive really (bad). Are there any open source solutions? I saw a couple but they look hard to setup and manage. Tapeware gives a powerful interface and makes it easy to backup from multiple machines... plus linux boxes don't need special server license (unless they have a tape drive) where any Windows 2000 Server box needs a server license.

Re:Lame

[JamesD_UK]

I currently use Bacula as my open source backup solution. Clients are available for Windows, Linux and Unix although I believe the server works best with Linux or Unix. It supports most hardware, including some tape robots (something that would be useful for 1TB of data!) and appears to be extremely flexible. It's done everything I've asked of it and more without complaint. Best of all the support from the author via. the project mailing list is second to none. The interface is through a console application although there's also a UI available (still a work in progress). There's also Amanda, you might want to look at that too.

Re:Lame

[syzygy_001]

if your a window's shop take a look at Retrospect

our Windows guy here has used it at the last few places he's worked, and it provides good backup and disaster recover, you can bun a recovery CD which will reinstall OS and then connect to the backup server and restore the box automatically to it's last (or whichever you tell it to) backup.

Got one. Did it cheaper :-)

[Space cowboy]

I built a similar system for the web rack (disks are bulky, compared to 1U motherboards). Gave me 1.5 TB of SATA hardware RAID-5 in 2U. All the other machines boot off it - much better use of space :-)

Simon

Completely Stupid

[Anonymous Coward]

No raid? Going to rely on the drive's MTBF? WTF. A raid controler is like 80$ MAX and one additional drive is like 250 or so. Spend the damn money. While you're at it. Invest in a tape drive. You're data is more valuable than the drives.

Re:Completely Stupid

[McSpew]

In the article, they point out that RAID will only help if a drive fails on its own, and not if there's a MB failure or some other badness. And one of the points was they were going for a cheap offering. Adding more stuff means adding more $$$$.

Okay, first off, what planet are you living on where solid state electronics are less reliable than hard drives with moving parts? Also, RAID is cheap, relative to the total cost of the system. These clowns spent almost $3200 building a storage system, but were too cheap to spend an extra $350 to support RAID 5? Yeah, that's freaking smart.

They're relying on the MTBF of the drives, but they forgot that your chances of failure go up with every additional drive you add to your storage setup. With four drives, you've quadrupled the chances that one of those drives is going to fail early. I don't remember enough about MTBF calculations and don't know enough about statistics to know what that does to your real-world MTBF numbers, but I'd bet that your real-world MTBF can be no better than one-quarter what it is for a single drive.

The people who wrote that article are serious morons. They completely misspent their money, if their goal was to build a storage server. For a storage server, what's most important is reliability and IO throughput. If you're on a budget, there's no sense in spending $500 on a CPU and motherboard and another $350 on 1GB of RAM. An Athlon and MB combo with 256MB or so of RAM could be had for probably the cost of their CPU alone. They could have then spent the rest of their budget wisely, by buying a high-performance RAID controller.

The problem is their idiotic dual-purpose requirement. Storage servers aren't workstations and vice versa. High CPU performance isn't a requirement for speedy file-serving: High performance hard drives and a high-performance caching RAID controller do that. Putting a gigabit Ethernet card in the server and hooking it to a switch with a decent bandwidth will do more to speed up your file fetches than putting a 3 GHz P4 on the MB.

Re:Completely Stupid

[Sqwubbsy]

Easy there, cowboy. I'm not an apologist for the writer, I'm just pointing out what he said.
Heck, if you're spending $3K, what's a 10% increase to prevent failure?
I was also ticked about the lack of OS explanation other than "Linux won't cut it."
For what they ultimately achieved, you could get fairly comparable results with a well tweaked setup from mwave.com. (Which is basically what they did.)

Today's News:

[JamesD_UK]

... it's possible to buy a large PC case and fill it with a large number of drives that add up to a volume of storage that was once considered to be large several years ago. What's new here?

The article could have covered a little more than just the hardware needed to run such a setup, perhaps covering some sort of remote management interface for the storage? It would also have been nice to hear if they solved the problem of backing up this data on a budget too. (Ingoring the possiblilty of burning the data to DVD).

Mini-itx

[herrvinny]

Wouldn't a mini-itx system make more sense here? You're building a simple storage server, doesn't need to be massively huge. A 533mhz processor (the low end with mini-itx boards, I think) is plenty fast enough to run a simple storage server.

Video card? Why on earth would you need a $70 video card for a storage server! He should have gotten a motherboard with integrated graphics, so even if he needed to attach a monitor, integrated graphics would be more than enough to handle anything. What is he building, a storage server or a full fledged PC?

Re:Mini-itx

[wilper]

Well, in the specs he said he wanted it to be a workstation AND a storage server. However, he didn't want to run Linux on it because of lack of drivers (Which I presume rules out the BSDs as well.), so he'll probably end up running some form of Windows on it.

So he basically built a windows workstation with lots of disks, guess the other users on the network will learn to hate the poor man who uses it when he reboots after every change in the configuration in the server, depriving them of access to the files.

I would have gone with a dedicated server, hid it in some closet somewhere and left it alone, and bought another workstation for the money saved on the server. (Give or take a few $100.)

Re:Mini-itx

[Zak3056]

He should have gotten a motherboard with integrated graphics, so even if he needed to attach a monitor, integrated graphics would be more than enough to handle anything.

Because if he wasn't blowing $70 on a video card, and $160 on his keyboard and mouse, he wouldn't be able to complain about how RAID would blow the budget.

His calculations for the power supply have SEVENTY WATTS budgeted for the video card, which, of course, forces him to spend $190 on the 450 watt power supply.

His motherboard has dual gigabit LAN, because "an extra NIC is essential for a server." Note, he doesn't say WHY he needs that extra gigabit NIC (fault tolerance? Performance? It looks cool?) only that he considers it "essential."

He has a hundred dollar add-on that "displays the latest stock-quotes and surf reports."

I feel dumber for having read this article.

Re:Mini-itx

[crucini]

The mini-ITX spec only calls for 55W of power, which isn't enough for a bunch of disks. Of course you could use a mini-ITX mainboard with an ATX power supply in some custom or semi-custom case.

As for the rest, I agree the author's goals are unclear.

stupid.

[Polo]

I don't think they used RAID. Drives aren't as reliable as they've been spec'd out to be.

I guess if they have everything important backed up on DVD and/or their data wasn't worth much, it'd just be a hassle... But when the system fails you end up with a big panic: running out to buy a new drive, then trying to get everything back up and running again.

I've built similar configurations and lost a drive (twice now!) and it's a big mess. At least with a separate system drive they eliminated one problem... if they lose the main drive they can reinstall and if they lose a data drive, they can at least reboot.

I would recommend raid -- at least raid 5 which would give them 3/4 terabyte and less headaches.

DRDB and or Linux Virtual Services

[_LORAX_]

If you want reliability you cannot just rely on ONE server anymore. Just get the cheapest boxes that meen the requirement and get *2* of them. Use DRDB and heartbeat to make the failover seamless. With these two cheap boxes you get 24x7 reliability at a 7-11 price. Raid, cooling, ... will all help in the one box senario delay system failure, but that box *WILL* fail. Two boxes can help not only with outages, but upgrades as well since the primary can be taken offline for upgrades without any upseting of the system.

The latest issue has reduntancy and scalability articles that go from 2 boxen to as many as you want.

http://www.linuxmagazine.com/

My solution

[olympus_coder]

I salvaged a derilict dual P3x450, dug up enough 256meg sticks to give it a gig a ram and a salvaged video card.

For drives, I watch and wait until I need more space, then I add a drive, ussually whatever Fry's has on cheap. I use LVM to add it to my partitions. Of course, I can only add a total of 4 drives this way before I'm forced to by a off board controler (I'm at that point now).

The other downside is that there is no redundancy, but oh well. Redundancy is expensive.

Performance stinks as I violate the rules about one device per controler. Of course, I don't care because I'm accessing it over a 10mbit network (via the phone lines in my appartment). It is sufficient to stream video to 2 or more machines so no worries.

Total cost ~$500 worth of hard drives. Everything else was "free".

Andrew

My budget storage server

[EmagGeek]

ECS Fully-Integrated motherboard
Athlon 1800XP, 256MB Ram
4x 40GB IDE Hard Disks
Promise SX4000 Raid-5 Controller
All in a micro_ATX chassis

Can't get much cheaper than $700 for a 120GB storage server with at least some measure of redundancy.

-1 didn't RTFA

[TubeSteak]

4+1 = 5 (and add in the DVD+/-R/RW for a total of 6)

ATA-133 controllers
We added a PCI Promise ATA-133 controller so we can run our four Maxlines as all master drives. This will improve simultaneous access performance and allows for an easy upgrade to eight storage drives.

Cost: $30

Seagate 120GB Barracuda 7200.7 SATA - 78%
A good hard drive but nothing to brag about, or at least nobody will care if you brag about it. It is a little noisy during operation.

Cost: $110

Maxtor Maxline II Plus 250GB PATA x 4 - 86%
A hard drive with potential, given its SCSI-like MTBF ratings and design for 24/7 applications. Having a hard drive fail is no fun, so if Maxtor has really built a more reliable drive any additional cost is well worth it. It gets extra points for presumed reliability. It is also faster than our SATA 120GB Barracuda.

Cost: $1,000

Re:WTF?

[pivo]

Yeah, if you want to build a mickey-mouse file server.

Which is what they've done. Notice they're also using it as a workstation to play games.

Morons!

[phlyingpenguin]

Since when do you need a 3ghz processor and a gig of ram let alone a GeForceFX (yes he noted it's slow, not slow enough mind you) for a fileserver?

And why is he putting a keyboard/mouse in the picture? Oh he's putting windows on it... he forgot to buy a license for that! I'm not sure I understand the comment on it not being smart to put XFS/JFS/ReiserFS/Ext3 on a firewire drive... can somebody explain why that's not smart?

$3,100 dollars is REALLY steep for a machine that shouldn't cost anything more than the drives it serves data from.

just built something like this actually

[StandardDeviant]

Just yesterday I brought up a server here at work to server as a 1.0 TB-range backup server using 8x200gb WD 8mb cache drives strung off a 3ware escalade controller (raid5, two hot spares). The build process was suprisingly painless (used an athlon-based solution but that's relatively unimportant. you'll want 64bit/64mhz pci slots for things like the 3ware storage card, scsci card to drive a tape drive, etc. the cheapest board I found that could do this was ironically a dual CPU MPX chipset board from gigabyte, sub-$200), with a total cost for a total beast of a machine coming in at about 3400 USD with shipping and such. I'd recommend heartily the 3ware controller cards if you want to try something like this, they're worth every penny of their ~200-300 cost simply for the increased performance and reliability they bring to the table as well as the reduced hassle (the array just shows up as a single huuuuuge scsi drive to linux... always nice when /dev/sda is reported to contain something like two billion 512 byte sectors ;)). I went with a black aluminum Lian-Li case because it has enough 3.5" drive bays to hold all those drives, comes with lots of fans by default (as well as cooling a bit better than your average plastic / steel case due to the thermal properties of the material), and a monster 550w "vantec stealth" powersupply for reliability and the ability to sustain all the devices in the system. Debian stable installed with zero hassle and now I'm just left with the pain of fighting with backup software. ;) True, I'd trust something from Sun or similar more than this homebrew thing, but this is also a mere fraction of the cost of something from the commercial Unix vendors, so for the same total cost I could have multiple redundant servers... or more ale-and-whores money in the departmental budget. ;)

he's right...

[snooo53]

While for a large business, $3000 must be dirt cheap.... for the rest of us it is WAY too expensive. I could either build a kick ass entertainment center for $3000 or their "budget" server.... I'll give you one guess to figure out which one I'd choose.

I've learned to be very skeptical of any of these articles on "budget" this or that, because they rarely are. To me, a budget server means less than $500. How about an article on how to build and configure a home network server for that price?

Re:he's right...

[Cramer]

An ok "budget" fileserver with 1TB of space can be built for around 1k$. And maybe even cheaper if you don't care what it looks like :-)

I built a 1.4TB array for about 2500$ two+ years ago. That was with 80GB drives. So, I know damn well it can be done much cheaper than this guys box. (We tested it as a full news server for a month. [click] yeah, you lose a lot putting a filesystem on it.)

Re:he's right...

[cornjones]

I used a belkin omni somthing 8 port had no problems at that res.

ej

Interesting?

[jargoone]

Somebody didn't RTFA...

At the same time, we wanted this server to act as a workstation with as much capability as the other systems attached to the storage server.

This article has some "WTF"'s in it

[ducomputergeek]

First off, you don't need a 128MB vid card for a server. Most have 2MB intergrated RAGE ATI cards. Old as dirt, stable, and works.

Next, what are you uses? I mean most small business work groups I have seen might store larger Powerpoint, excel and other files. It takes them a while to fill up dual 160GB hd's in a raid 1.

Still, for our company we purchased 1.6TB Xraid's from apple with Fiber cards. Why? well we are doing a lot of work with FCP and need the quick access times that come with fiber vs. ethernet.

My config...

[tinrobot]

Built a storage server two years ago, it's run like a tank since I put it online.

Dual 800MHz PIII in a Supermicro Motherboard.
Cheap-O video card
Gigabit card
40 GB system drive.
6x80MB Maxtor drives (5400 rpm)
Escalade RAID-5 card.

I chose 5400 rpm drives for several reasons:

A) A little bit cheaper
B) Used half the power of the 7200
C) Runs a lot cooler
D) Higher MTBF

Every drive that has ever failed on me has been because of heat. I put several fans in the case to make sure the drives don't overheat. So far so good (knocks wood)

How Funny

[WndrBr3d]

My company just recently invested in a mass storage solution, since it's obvious that mass, redundant storage on SCSI (>300GB) isnt a cost effective option for a small office environment. We took the easy way out and purchased the following:

Dell PowerEdge 1600SC Server:
Xeon 2.0Ghz
512MB RAM
18GB U320 15k RPM (OS Drive)
32x CD-RW/DVD Drive

We chose this server because it has both PCI33, PCI66, AND PCI-X slots on the bus, supports up to SIX internal hard drives and has two 5.25" drive bays.

For the mass file store we chose Maxtor 300GB 5,400RPM 2MB Cache Drives. You have to remember this is not going to be an active file server but more just a file repository and source control/backup server for a small office (10 Clients).

Our Mass Storage Solution Is:
3Ware 7506-8 RAID Controller
4x Maxtor 300GB Drives

We're going to put the Maxtor Drives on a RAID5 and since the 3Ware is a Switching HARDWARE 64-Bit/66Mhz PCI RAID card for IDE Drives, performance should be stellar.

I think all in all the entire solution ended up costing us around $4,000 for parts and systems, BUT, we also got OS (Win2k 5 CAL) and a 3 Year Dell Warranty on Parts.

I think $4,000 for a 900GB Hardware RAID5 on a Xeon server aint too shabby :-)

Budget

[herrvinny]

Total $3,140

Okay, I just looked at the article again. $3,000? Damn. I wouldn't mind having that budget...

Seriously folks, if you think you need $3,000 to build a server, then you're out of your minds. I don't want to be modded as Flamebait, but anyone here at /. (including me) could build a server for less than half that, and I would bet that for storage activities, it would be equivalent or faster than this moron's PC.

Video Card? Keyboard? Mouse? No. Shouldn't even be there. Yeah, sure, during initial setup, connect a secondhand monitor, mouse, etc (who doesn't have a spare monitor lying around? I have one 10 yrs old lying around somewhere and it still should work). But after initial setup, after you install and configure Linux/Apache, Windows/IIS, FreeBSD/whatever combos, forget it. After that, you should be able to telnet or remote admin the server.

I'm going to issue a challenge. Alexis Dang (the author of this piece), if you're listening, here's a challenge. Give me $1500 and I'll build you a server that can beat your server in storage related activities. Not video games, not music, not Paintshop testing.... just pure storage. Hell, give anyone on this board $1500, and they can beat your "server" upside down.

Re:Budget

[cornjones]

hmmm.... i'll bite. just grabbing some quick prices from pricewatch i would say:
65$ - M811LU Socket A DDR MB with Duron 1.1Ghz (1100Mhz) 200FSB CPU & Fan (don't know anyting about the board but it has video/sound/lan built in)
63$ - 512MB PC2100 (figure 512mb should be plenty for our file server but more wouldn't be amiss)
18$ - Mid Tower ATX Dual Front USB (USB 2.0) opt., 8bay, Case only (mainly for the 8 bays)
28$ - 600Watt EXTREMO power supply Dual Fan Aluminum Super Silent (600 should be plenty)
12$ - 4 X 80mm Sleeve Bearing FAN FOR CASE W/4PIN CONNECTOR (not sure if i can even fit 4 but the more air circ the better. )
18$ - LanReady - 32bit PCI 1000/100/10Base-TX Gigabit Fast 1000Mbs Ethernet NE2000 Adapter (gigE is a must.)
4$ - KB 107-Key Standard PS2 Keyboard
2$ - Generic ps/2 mouse white
7$ - 1.44 Floppy Drive 3.5inch NEW beige w/faceplate
16$ - Cyberdrive - 56X Internal IDE CD-ROM Drive
35$ - Image Western Digital - 20.0GB EIDE 2MB 5400rpm Ultra-ATA/100.

Ok, So there is our base system. That comes in at $268. (this wouldn't be a bad system for the 'rents)

Now we do our storage.
Everybody seems to be on there knees for 3Ware so we are looking at either 4 or 8 drive solution
For the 4 drive:
245$ - 3Ware escalade 7506-4LP, 4 channel udma raid adapter
For drives we are looking at
732$ - 4 X MAXTOR 250GB EIDE UDMA-133 HD 5400 RPM on the higher but reasonable end
and
324$ - 4 X IBM/Hitachi 120.0GB EIDE 5400Rpm 2MB,8.9ms for a cheaper option

So that gives you about 750gb w/ raid 5 (500gb w/ one spare) on the high end, 240gb (160gb w/ one spare) on teh low end.

if we go w/ the 8 drive setup
373$ - 3Ware escalade 7506-8, 8 channel udma raid adapter
For the drives we are looking at:
1464$ - 8 X MAXTOR 250GB EIDE UDMA-133 HD 5400 RPM on the higher end
and
648$ - 8 X IBM/Hitachi 120.0GB EIDE 5400Rpm 2MB,8.9m for a cheaper option

this gives us 1750gb (1500 w/ 1 spare) or 560gb (480gb w/ 1 spare).

So our final numbers:
4 drive machine:
1245$ gives us 750gb
837$ gives us 240gb
8 drive machine:
2105$ gives us 1750gb
1289$ gives us 560gb

The prices all come from pricewatch.com and should include shipping. some massaging may be necessary depending on your location and vendor choices. The raid sizes come from RaidCalc (http://www.ibeast.com/content/tools/RaidCalc/Raid Calc.asp)

I didn't include a monitor on these but you can get a compaq 15in for 64 bux if you need it.

Use your favorite free OS w/ samba/mac share or your stolen copy of windows and you have a nice weekend project.

mmmm... toys

back to work

Re:Budget

[Idarubicin]

Hell, give anyone on this board $1500, and they can beat your "server" upside down.

Indeed. If you want to spend the full $3000, build two of your $1500 boxen, and then you have a complete backup.

Yep. I have one too

[The Tyro]

but I admittedly went a little over the top...

- Raid rack-mount server chassis (space for 8 drives)
- 3ware RAID controller (great linux support)
- multiple 120gb drives in RAID-5
- dual-athlon MB, bunch of RAM
- CrystalFontz LCD running LCD4Linux
- Samba, Postfix, etc.

It has enough extra horsepower that I can run a counterstrike server along with providing network services, primarily huge storage, for all my other machines. It's full of high-bitrate oggs (reripped everything; it took weeks, even using Grip's auto-rip feature). Oh, let's not forget the high-quality DivX.

Apart from giving me room to grow, it's made me a huge fan of dualies. I've never worked on a machine that's as snappy and responsive.

Rename the article

[bigjnsa500]

They should rename this article:

"How to build a budget file server without knowing what we're talking about"

3 grand is on a budget? What happened to raising from the grave an old AMD K5-166, throw some big IDE drives and you really got a budge file server.

Just don't hook it to the internet.

[emil]

OS X 10.1 users are still waiting for a patched SSH.

While Apple includes server software in OS X, Apple is not excited about you actually making use of this software (they would rather that you buy OS X Server), so it will constantly be a thorn in your side.

I've thought about OS X server applications, but...

• I don't care for their support policies
• Apple likes to use support to cut OS functionality (itunes)
• /etc/passwd is a rump, and the c library pulls from an Apple licensing daemon, which makes me uncomfortable
• The FS also makes me uncomfortable (non-UNIX-native OS9 FS, journaling is a 20% performance hit)
• The "repair filesystem permissions" maintenance gives me the willies.
• Apple doesn't seem to be jumping up and down about any sort of standards compliance (POSIX, UNIX98, etc.)
• There is a lawsuit between Apple and X/Open (OSF) about Apple's claims that OS X is "UNIX." X/Open says that it most certainly is not. Apple should license and certify.
• Prebinding/optimizing Objective C applications has become a voodoo rite.

It seems like a good deal at first, but look before you leap.

and assuming one works for free ...

[aoj]

that looks like a fun home project but I would start here when looking for a small office server Penguin Computing Relion Servers

Re:Wow

[kwerle]

Mac OSX.......... $129 (you DID pay for it, right?)

Actually, assuming they only have < 5 OSX systems in their house, they probably bought the "family pack" for $199. So the /system cost for the machine could have been as little as $40.
For a supported system (good, easy software updates) and an OS that's a pleasure to use, I'd say the money was well spent.

No RAID == begging for trouble

[PSC]

If you believe the numbers, running a drive in RAID mirror will double the effective MTBF, we have done that by choosing the Maxline series vs a standard consumer IDE hard drive.

(Shakes head and bangs it violently against concrete wall)

MTBF and RAID is about entirely different things. The R in RAID stands for REDUNDANCY. You can have a MTBF approaching infinity and you would still have no redundancy.

Mirroring does NOT just double MTBF. It folds two probability functions. With RAID1 not only have both disks to die for data loss, but both disks have to die at the same time! (Or in fact, during the recovery window.) With a MTBF of 1.2 mio hours and a recovery window of maybe 5 hours, this really makes the difference.

Using non-RAID IDE disks, especially on a server, no matter how small the budget, is just playing russian roulette with your data. With at least 5 chambers loaded. It's wantonly negligent. It's unprofessional. Don't do it.

(As a side node, the MTBF is an utterly useless bit of information. It is determined by e.g. running 10,000 disks for 10 hours, with one disk failing. That is one dead disk in 100,000 hours of operation, so MTBF is 100,000. It's a bit like saying that if one woman can make a child in 9 months, 9 women can make a child in 1 month. Reality just doesn't work like that.)

A Budget!!??? A point?

[westyvw]

I have many different storage servers, at different locations. They have no clue on how to build on the cheap. They mention no Linux (or BSD) and thats just plain stupid. They put a video card in a server? WTF? You dont neeed one at all, all admin access should be with an xwindow on a main computer.

Slashdot, News for Nobody. This was the lamest article I have read in awhile.

Re:Here's a real budget storage server

[crombie]

Nice rig. Like the case. Why do you need an extra 500 watt power supply? Did you remove the one that comes with the case?

How are you doing backups?

Isn't the esclade a 64 bit pci card? I don't think the a7n8x supports 64 bit, correct?

Re:What OS?

[jeffhot]

Yes, this seems like a huge oversight to me too. Especially when comparing it to offerings from Dell and Apple, who surely include a Server OS with their hardware. It's not like you can just throw Microsoft Windows Server 2003 into the mix without affecting the budget. I know that XServe from Apple they are referring to comes with an Unlimited Client version of OS X Server. MS Windows Server 2003 is $999 for only 5 Clients, and $3999 for 25. Now compare this build it yourself to the Xserve! There reasoning for saying no to Linux was the formatting of the hard drives into any common Linux format would reduce compatibility when the drives were eventually moved into External FireWire enclosures.

Having looked at it thoroughly..

[mindstrm]

I can't figure out why these guys thinkg a DVDR is a backup solution
a) Likely to fail
b) Look how much time, and how many discs it will take to back up 1TB.

The realistic backup solution for stuff like this is: stuff like this.

Back up to a set of hard drives. Seriously. The cost/MB is still the cheapest out there, and it's more flexible, and heck, way faster than tape.

Re:Wow

[TheCrazyFinn]

OS X is quite fine on that hardware, especially if you turn off the GUI (Which ain't needed on a server anyways). Yes, you can turn off the GUI on OS X. It's a one-line change.

Where is the ECC memory?

[chunkwhite86]

If it's a server, isn't data integrity a higher priority than sheer performance? Why aren't they using ECC memory modules? Price is not an issue - I have a dual Athlon MP system which supports ECC and I'm running 1.5 GB of PC2100. The 512MB ECC modules were only like $112 each.

Plus they complained about not having front-panel firewire and USB! WTF? This is supposed to be a server isn't it? Not an iMac!

And my final rant - An NVIDIA FX video card? Are they smoking crack? A Matrox Millenium PCI card is all you need in a server. GeForce FX is the last thing I would ever imagine to find in a budget storage server.

Dangerous

[SomeOtherGuy]

Talk about working without a net. I mean, why call it a file server -- sure it will serve files...But it will not do anything about redundency or recovery. Thus it is just a Desktop with lots of standalone drive space. The whole file server moniker should be reserved for machines that not only collect and serve your data -- but also protect and back-up your data. No raid, no mirrors, no tape backups -- no nothin. And some good the 3d graphics card or MTBF will do you when one of the drives goes south taking your data with it.....(Well at least you may be able to replace it under warrenty with a new EMPTY hard drive and play a mean game of Unreal Tournament or something....:)

This is budget?

[Anonymous Coward]

You want budget? I bought a pair of Maxtor 40 GB HDs, a stick of 256 MB of RAM and stuck them in an old P150 with Windows 98.

That thing is used by a workgroup of about 6 for storing just about everything they need with their business. (They're using at most 10 gigs.) Every night, a batch file starts XCOPY to backup all files that have changed during the day from the main disk to the backup. Every little while, I burn CDs of everything with the snappy little script I wrote.

Total cost at the time (two years ago?), probably 500 bucks.

That's what I call a "budget storage server".

Article never really explains

[0x41]

Alexis builds a budget storage server and explains why you can't take a random desktop and add a bunch of disks.
Funny, the article never really explains why you need all the super-super parts that he uses to build the server, except for the reasons like the UPS looks cool on your desk. Whoever wrote this really just took a budget and blew it on a bunch of cool parts to build a kick ass server, but would have been out of a job if he did the same at my IT department for wasting IT budget and time.
-$0.02

And we're backing this up how?

[vasqzr]

Anyone can build a PC and fill it with 300GB IDE drives. Big deal.

A few problems:

#1, if you've got BIG files (read: audio/video) or a quite a few users, it won't be fast enough.

#2, It's not a real SAN device. Where's the SCSI and Fibre channel? Can you plug it in to multiple servers?

#3, How do you plan on backing this up? a stack of DVD-R's? A bunch of 80GB tapes?

Re:Tape still competitive

[red floyd]

The equalizer is the cost of the drive. CD drives are dirt cheap; if you back up to hard drive, the drive is the media; but if you amortize the cost of the tape drive over, say 100 terabytes (not so unreasonable given the durability of tape drives), you bring up your cost to maybe 55 cents on the gig for tape. Granted, if you back up only 10 terabytes, it's no cheaper than hard drives.

No, the cost of the tape drive is not necessarily the equalizer. The equalizer is the cost of the operator sitting there and swapping CDs/DVDs, as opposed to getting a tape solution that can hold your typical incremental backup on a single tape, that can be dropped in at COB, and removed in the morning.

Another stupid bit

[rhizome]

Well, this is where I stopped reading:

Another possibility was RAID 5, which allows 5 drives to act as 4 drives. An additional parity track is written on each drive, so if one fails, then the other drives can recover the lost data. This is available through software or hardware. This is a great solution if you do not plan to upgrade your maximum server capacity. When the time comes to replace a drive with a higher capacity drive, you will be forced to replace the entire array.

Right. The thing reads more like an excuse to play with some SATA drives they got for review or something. At any rate, the article presents some seriously flawed and amateurish design decisions under the guise of "budget" architechture.

1/3 disks, 2/3 bloated computer

[billstewart]

They spent about $1000 of that cost on disks, and were too cheap to spend an extra $250 for RAID, but they spent $100+60 on a really cool keyboard and mouse and $100 for a really cute front-panel display.

They spent $300 for a Pentium-3 and $200 for a high-end motherboard and $350 for the fastest most expensive memory they could find, when a "budget server" could do just fine with a ~$100-150 2GHz CPU+motherboard and $200 for 1GB of average-speed memory. (Their motherboard does sound good, though.) After all, the bottleneck here is the disk drives and network, not the CPU, though even on a budget server it's probably worth having the 1GB of RAM for caching and for staging CD or DVD burns.

The $190 power supply seems expensive, but that may be realistic for a system that can expand to 8 drives. If you've got a UPS, you may not need as high-end a power supply, and a "budget" system might get away without it, but since they were too cheap to buy a 5th drive for RAID they're probably much more in need of highly reliable power. And their 3GHzP4 CPU and overpowered-for-a-server video card use too much power and put out too much heat - you can easily save 50-75 watts by making better choices, and probably 100. You could save even more by using a motherboard with built-in 2D video, but most of those don't have the high-performance networking support yet.

Also, they didn't have a price for an operating system :-). That means that they're planning to use Linux, which is another reason not to waste power or cooling or money on a gamerz video card...

SCSI/ATA compromise

[Sloppy]

One thing I would like to try out some day, when building a big RAID, would be a compromise between ATA cheapness and SCSI lots-of-drives-on-one-PCI-slot goodness.

There are adapters that turn a cheap ATA drive into a SCSI drive, roughly in the neighborhood of $100 apiece. If these things work well, then one may be able to put a whole bunch of cheap+100 (which is still pretty cheap) ATA drives on a SCSI (160 or 320 or whatever) channel, instead of adding a bunch of ATA cards and using up all the PCI slots.

Yeah, yeah, ATA sucks, whatever. I have actually had good luck with cheap ATA drives (better than my SCSI experience in the 1990s, for sure) and besides, you RAID 'em based on the assuption that yes, they will die. And when one dies, you pay chump change to replace it. (And when they don't die, then you smirk and say, "That's because I'm ready for it. What would be the point of them dying?")

Another good thing about this approach is that SCSI makes it easy (compared to ATA) to use external enclosures. I'm party convinced that a lot of my good and bad drive experiences have been related to cooling issues. (My expensive Micropolis SCSI drives roasted inside my Amiga 3000 case, but my Coolermaster and Lian Li cases are doing a great job of keeping my cheap Maxtors comfortable.)

Anyone have any experience with building RAIDs of SCSI-adapted ATA drives? Can they transfer at Ultra160 (or 320?!) speeds?

Re:offsite

[ivan256]

Tapes are guaranteed to survive a bumpy truck ride off site, and the time you accidentally drop it. You also don't have to worry about wether your tape will spin up after sitting unused on the shelf for a few years. You get no such guarantees with a hard drive.

If your data is worth anything to you, or you have any interest in archiving, hard drives are a poor choice for backups.

Re:Thread idea: what do you have at home?

[Lysol]

Ah yes, all excellent. I love the home network.

Currently I have a few of the Shuttle boxes that run 1ghz p3 and 1.4ghz celeron & a white box generic Tyan 1.2ghz Athlon dually.

The p3-1hgz does my personal domain hosting - web (tomcat/apache), mail, openh323 gateway, ut server.

The celery does mail and public ssl egroupware portal for my home biz.

The dually is running rh 9 w/software raid across 2 ata-100 drives. It handles all my MySql stuff as well as the main repository for all my machines backups - it's on an the internal net. Once a week nightly backups are dropped to tape and once a month, the backup dir is wiped. So I always have at least 1 week of backups on tape, which I'm quite happy with. This is also my CVS server too.

All my development is mainly done on me 1ghz Powerbook. I also have a excellent 80gb Seagate Barracuda (fluid bearings) in an external firewire enclosure that has all my mp3z & mov/divx's. Sometime I just drop it on the dually and it mounts it up and shares it on my network, but most of the time I keep it local. My Powerbook is also dual head w/a 17" 760V Samsung LCD. Great monitor for the price.

My last machine is a kick ass little Shuttle box with a Athlon XP 2000+ and Maxtor ata-133 20gb drive. It's dual partitioned w/XP and Fedora. I use the Fedora side to develop for another Linux server that is just easier vs. doin it on my Mac. The XP boot is for Starcraft & UT. :)

All the machines have 1gb of ram except the two Intel Shuttle boxes. Everything internally is connected via a DLink Fast Ethernet switch. My external net is just a cheapie FE switch which is all bridged together via a even cheaper little broadband 'router'. It has dhcp, firewall and other things built it and has worked pretty well for me. My connecion is 1.5 down and 786 up DSL. No cable here. There's also an Airport Base station on the wall that provides wi-fi for family and friends when they drop by.

I've been running networks for many years outta my various cribs. The biggest and baddest was when the whole boom was goin on. I had a startup w/some friends and I had 9 machines in a closet of my studio apt. This included 2 big (1 size down from rack mount) UPS', a Dell quad Xeon 550 server (scsi 10k rpm drives with raid 5), 512mb ram and hot swappable everything - 4u also. This was out big app server. I was also running 2 RH 5 dns servers, 1 RH 5 mail server, 1 RH 5 ssh server, 3 dell p3-500 2u apache servers and one failover app server (Weblogic), 2 NT SQL 7.0 servers with full replication. I also had a separate tape backup box that was just a p2-350. Ah, the good 'ol days. After a bit we sold the company and I went traveling to Europe never to have another rack in my closet again. :)

Anyway, I love home networks and my two big obsessions now are: quiet and huge storage. I thought the article was kinda lame when I go to the video card and mouse part - something my servers have never really had or needed (cept maybe the nt junk). One thing I found higly annoying is when RH 9 required a mouse to install. I didn't like that much, but no big I guess.

So I have yet to find the magic bullet machine that will hold a 1TB array for cheap and for quiet. But maybe some day. Could it be that now that there's 5400 rpm 2.5" drives that that could be the thing??? Low power and quiet..

My 1TB Media Server

[meehawl]

I built this out of cannibalized parts last January 2003. I suppose by now you could probably double the media storage for the same cost -- there's a lot of rebates for PATA drives around.

Supermicro P6DBE (1997 vintage)
2xP3 600MHz

Adaptec 1940UW SCSI
Software RAID 1
x2 36GB Seagate SCSI drives
(web server)

1GB ECC PC100 RAM

x1 WD1600JB PATA drive
(apps)

Promise SX6000
Hardware RAID 5
x6 WD1600JB PATA drives
(media server)

ATI Rage Pro
(it's a server!)
Antec 1040SX Case

Antex True480 - 480 Watt PSU

Basically, all I bought new were the drives, the case, and the PSU. Total cost below $1300. Serves several thousand visitors a day, peaked at 30K hits for a while following a Slashdotting. CPU usage peaks around 20%. Using J River's Media Center, I've tested it serving 6 simultaneous 720x480 DIVX streams to clients over LAN and WAN with no problems.

These chumps spent 3 times what I did, and they don't even have disk redundancy. Who let the dogs out?

Re:Mini-itx - I did that

[RatBastard]

I have a 533Mhz Via Mini-ITX motherboard driving my file server. Here's what I built:
MB: Via 533Mhz Mini-ITX
Video: Built into MB, crap, but who cares?
NIC: 100 Base-TX built into motherboard
RAM: 1x 512MB DIMM
Storage:
- 1x 20GB Maxtor hard drive for the OS
- 2x Maxtor 120GB drives plugged into a Promise Ultra 66 PCI IDE controller, mirrored
Case: Some old piece of crap mid-tower ATX case
PSU: PC Power and Cooling 300W

It's not uber-leet, but it gets the job done. The system also has a minimum of fans: on for the PSU and one for the drives. Neither the CPU nor the video have fans.

My needs were for a reasonably large capacity (yeah, 120 MB is hardly "large" anymore), reasonable responciveness, low-as-possible power consumption and noise.

I wouldn't use this thing in a production environment or as a mission critical system, but for a home file server feeding files out to four/five client systems it works fine. (And yes, I am planning to put a backup system in there.)

This attrocity tha these idiots specced out is a sad and pathetic joke.

They misunderstand RAID

[jamie(really)]

If you believe the numbers, running a drive in RAID mirror will double the effective MTBF, we have done that by choosing the Maxline series vs a standard consumer IDE hard drive.

RAID isnt about increasing MTBF, its about not having to bring the system down and not losing data when (not if) a drive fails. When one of their four drives fails, they will have to stop the system and restore from an old backup. It takes a while to restore a 250Gb drive. And of course all the data since the last backup is gone.

Then there's the fact that they have four seperate drives, while with RAID you get one big one.

RAID controllers, especially the Escalade one, do a much better job of managing disk writes than your onboard IDE controller. For server usage you will see a much higher response for multiple users using RAID. No reference is made to the different behavior of drives in a server compared to workstation drives.

They remark that RAID wont protect your data because the PSU or motherboard can fail. Ok, I have never had a motherboard fail. That doesnt mean that they don't, it just means that their MTBF is way beyond discs. I have had a PSU fail, but not in a way that damaged the computer. You could consider dual PSU solution. Or a post-psu UPS (i.e. where the battery is between the PSU and the motherboard), as opposed to, or in addition to, a traditional pre-PSU UPS.

But then, the whole article is something of a joke. A $3000 budget server with the most expensive RAM, CPU, Keyboard and a bloody LCD panel??? I dont know what planet they are on but for $3000 I built a dual P4 XEON box with a Promise SX6000 pro raid controller. (I buy escalade now, I might add). And a $160 Keyboard for keyboard and mouse? What's it need a keyboard for? For $160 you can buy a 4 port KVM switch and use a keyboard you've already got. Or spend $5 on a basic one.

They use one gigabit port to connect to the internet. Why not connect both ports to a capable switch and get 2Gb/s?

BTW, I tried low-end Seagate, Maxtor and WD, and finally found that Samsung Spinpoint drives survived the longest in a RAID box.