Slashdot Mirror
Can My Desktop Make It in the Big Leagues?
bionic-john wonders: "I work in an environment where the dollar is more than almighty (who doesn't?). One of my cost savings plans is to use desktop computers as servers. They cost much less, the parts are readily available and/or interchangeable - as opposed to waiting for overnight proprietary or obscure parts from a vendor, and so on. I understand that servers have redundancy on disk and power - but this can be emulated for a fraction of the cost, as well. Is there a performance difference between a desktop and a server with the same specs? Chipsets are chipsets, motherboards are motherboard, and memory is memory -- is there something special about a server other than looking at the rack of blades and feeling special?"
If you don't have much space to spare, I would go with rackmount servers anyway. Some also provide remote administration capability separate from the OS, meaning you can reboot it and such.
US businesses that currently accept chip and PIN/signature
Just be careful about checking and cleaning the power supply sometimes, and you should be fine.
In my experience, "servers" are merely designed with rack-mountable boxes as opposed to floor-sitting boxes.
Bob-
The Ludwig von Mises Institute. The reasoning individuals economics
- Disks fail. When you stick a server in a rack and leave it running for 5 or 6 years (unlike your average /.'ers desktop which probably gets a shake-up far more often), you won't regret being able to hot-swap a failed drive on your RAID array with a spare.
- Power supplies fail... To be honset, this isn't nearly as big a deal in the hot-swap arena as the hard drives. However, having 2 power supplies in a server machine means that things are significantly less bad when or if one of them happens to fail.
- Vendor commitment. From those old Compaq Proliants to the new Dell Poweredge machines, they were built to be stuffed in a rack and left untouched (unless something fails... see above). They'll come with hardware that those vendors usually stake their reputation on or even had a hand in building. Even the management software isn't always bad....
While there is something to be said for the "Server-Grade" hardware, and rack mountability at that, there is no good reason why intelligently chosen and configured "Desktop" hardware can't perform as well. The key is to recognize limitations of various components, such as being aware of SCSI vs. IDE specs, and the fact that standard PCI slots prevent total saturation of a 1GB NIC. If you choose your parts wisely, and with your goals in mind, you can save quite a bit of money without sacrificing performance or reliability, and maintaining vendor neutrality.
You are not the customer.
Chipsets are chipsets, motherboards are motherboard, and memory is memory
Well, chipsets can be argued... but memory is not memory. Server motherboards are equipped to use ECC/Buffered RAM which is far better for production servers.
It's only relevant if uptime is key, but with desktops, you generally wont have:
Redundant power supplies
Redundant disks
Hardware raid (other than 0/1)
If that's not important to you, then by all means go for it
-- (appended to the end of comments you post, 120 chars)
In the long run, it comes down to standardization and serviceability. If you've got maybe 2-4 servers, go for it. Otherwise, you're in for TONS of headaches. Desktop lines are changed CONSTANTLY, and you'll find yourself always trying to get a part for something that is discontinued.
Still, you can do it. But I stand by the statement you're a dumbass.
redundancy... speed... noise... that coolness factor you get when you say 'yeah, i'm running quad 64 bit opterons with 4 gigs of ram each. yeah thats right, this bitch has 16 gigs of ram, what you got?' and umm... well thats about it. if your needs dont require you to have dual true gigabit nics, dual or quad processors, a scsi raid array and a space heater/pink noise generator, then get yourself some decient computers with the basics. servers are usually built with better parts (i dont know for sure but i'm pretty sure that the same silicon wafers that make the 2.4 celerons also make the 3.2 p4's, the difference is in the testing - i think, plz correct if i'm wrong) but for the most part they're standard parts.
---- The first point-and-click interface was a Smith & Wesson
More details here.
Never take the cheap option when buying something new, but if it's for personal use old computers which have a proven track record of being reliable are just as good, if very slow in comparison, my old P2 400 is still chugging along 24-7.
New good quality fans, and filters are a good investment though. If you skimp on them you'll only end up having to clean out the dust from the case and replace the fans possibly after the damage has been done.
Bits of experience form my days of administrating a heterogenous network of desktops-as-servers in an ISV shop (disclaimer: I am professional software developer, I did administration because I was most knowledgeable OS geek). Several reasons why you don't want dektops to be servers: * Power supplies. Beleive me, PSUs DO fail. And more hosts you have the higher probability of failure you get. Even if you keep a stock of PSUs in the closet. It still takes you about 20 mins to get desktop/server up and running again, and night failure is far worse. * Rack mounting is not a vendor trick to charge you more money. If you have more than trivial infrastructure, wiring on desktops and "floor-tops" is going to be your favourite nightmare. * SCSI and SCSI raids are just a waste of money on a desktop but it is must have for intense, parallel access of many users to their homes, mailboxes, whatever on server. * not last and not least: having someone working on a server is probably most stupid idea in the whole IT. Whatever OS you use, beleive me, users will find a way to devour 98% CPU time and 99% of memory. That leaves for server applications.. well .. do the math :)
There are many other things, I just came up with whatever came into my mind right now.
my sstream of consciousness
Perhaps you've heard of a little technology called RAID? In most environments where space isn't an issue and you have a real IT staff, you could probably get by with a bunch of consumer-grade parts set up in a reduntant fashion.
In fact, the low-cost "servers" you would get from Dell aren't that much more than consumer-grade parts specifically configured to be ran as servers. The cheapest ones come with IDE and Celerons / Pentium 4s.
When it comes to hardware, you should only buy what you need and enough redundancy to keep running through the installation of the next level of redundancy. Computers depreciate faster than any other expense you could have; they aren't drill presses or factory automation.
Simple economics: if two "servers" cost $1500 each, and you can get "PCs" for $750 each, you can either get four times as many or save half the cost--which can help you move to better equipment as the budget goes along.
I have an "obsolete" low-end server that I use for running FreeBSD. It has SMP, ECC RAM, SCSI disks, a boring but very reliable chipset, extensive documentation, diagnostics software, and a high-quality case and power supply. It is also tested and certified to run all of the popular server operating systems. The manufacturer support is excellent. The video card would suck for a modern desktop, but who cares. It never crashes, it just works. If it does break, I can get parts and service.
Mea navis aericumbens anguillis abundat
What makes this guy a server? I'm no expert, but here's what I see:
- Lots of RAM. Came pre-configured with 1GB, and could handle many times that.
- There's only two 32-bit PCI slots, but four 64-bit slots. Handy if you want to add RAID or Fibre support, a nuisance if you want the more ordinary kind of add-in.
- No built-in sound card.
- No AGP interface. Instead, there's a basic 4MB video interface on the motherboard.
- Massive fans.
Anyway, bionic-john is correct in thinking that a workstation will do as a server, provided only that you don't demand more of it than it's designed to do. (Which is always a question anyway.) I work for a hosting/colocation provider, and I see all kinds of stuff pressed into service as servers: cheap white boxes, Sun and Apple workstations, even an X-Box or two. Ultimately, all computers are interchangable. Specialized computers are just a matter of convenience and cost-effectiveness.Really depends on how "BIG" you mean.
Clustering, then yes, PC's might be a ways to go, but you trade manpower time for uptime per box.
If its single app box, and you are not hot swappable, you wont be able to make it to a maintenance window for repairs.
Blades are good, but not the end all, normally you have dozens of blades around a big beefy database, and the database box and disc storage are the expensive beasts. Also licensing is a factor, support, lots of things.
Not all setups are the same, you should of listed what type of transactions you serve up.
I find people will save money on the short term, and spend more money 6 months down the road due to poor design.
well - it turns out that one of my white box servers crapped out on me moments after this article! I do not feel bad, nor do I feel like it should have been a 'server' quality machine. The machine was in fact a 1996 PII, it may have even been a cyrix. $200 later and a couple hours, I rolled out a new PIII-1000..the downside was working on SAT.
The load that these machines take are not much more that what that PII could handle (in fact I think that load handled everything great other than its nightly data mirror)..and the 'MISSION CRITICALITY' -- well that is debateable - I agree with the other poster that 2 for the price of one is a great deal (rough cost estimates, but it is close)
I can see a couple servers in the organization that are mission critical (web/mail) - but for some of the print/file servers - like I said, these little white boxes kick ass running a hybrid linux distro.
Yes - there have been times where PSU's died - but that was probably my own fault - maybe I should replace them yearly? In fact - a couple of the boxes have over a year of uptime - they just crank.
I appreciate your input on all this
PS = a lot of talk about RAID - dont forget there are some great IDE and SATA RAID cards out there - I use a few myself.
I guess it all depends on your environment and the availability you want to provide. There are reasons that servers have redundant "everything". The primary reason, is for availability. If you work in an environment that can easily deal with a few hours of downtime at ANY given time, then I guess you may be able to pull this off.
If you work at google.com, or some other high-availability company, please resign ASAP.
Seriously - it depends on your budget. If your budget is _that_ shoestring, you should probably be looking for another job. Because the only thing executives like to cut before operating expenses (ie, software and hardware), is salaries.
If you're young and new to the industry, take this advice: the big-wigs won't be as impressed by saving $200 on a server as they would by a $400 MORE EXPENSIVE server staying alive for ONLY 1 HOUR longer. When you're young $200 might seem like a lot of money, however; keep in mind that in a company environment, that amount of money is pennies. A corporate environment probably loses TENS of THOUSANDS of dollars for every hour their servers are down. Obviously, there are a ton of variables (your industry, etc); however, for the most part I'm completely correct. Ask your office manager, department head, or some other significant person how much money they lose every hour their network is down. Be careful, however, not to ask it in a way that makes you sound "opportunistic" - because that's the fastest way to 100% downtime (unemployment).
Post more details about your environment, and maybe we can help you come up with better solutions.
Don't think that a small group of dedicated individuals can't change the world. It's the only thing that ever has.
I've been running cheap desktop units as servers for email, http and win shares for 5 years. No problems. Just buy decent power supplies, check that all the fans are turning once in a while, and change out the hard disks every year or two.
"Eve of Destruction", it's not just for old hippies anymore...
Do not use a desktop as a server unless it is being used by less than 10-20 users. I say so because desktops have 1 CPU while most servers have at least two. what makes a desktop and a server different is the multiprocessing ability. any pc with 1 cpu is desktop or workstation, no matter what the marketrolls say. What you need to do is to buy a 2xOpteron board and two cheap Opterons 242-244 together with 2x512mb reg ddr400. stick them into a desktop case with the best atx or eps+12v PSU you can find and also some SerialATA drives and you will have just built your own server. do not buy IDE or SCSI: SATA and SAS is the future.
The correct answer to the question is what is the value of downtime to you. Often a few hours of being offline dwarfs the savings possible from this approach.
There is no question you will have more downtime with desktop hardware - it in just not engineered with 365/24 in mind. You can add in a few extra fans and make sure you don't have any proprietary parts like Dell and HP throw into their desktops, but in the long run you WILL have more downtime.
It's not officially a Good Idea, but is fine for some environments.
Just take into account that server and desktop hardware are designed with different goals in mind. Server hardware is meant for 100% uptime, even in the case of most hardware failures, and have good scalability under high loads, while desktop hardware aims to give you the best bang for your buck, understanding that your data is typically much less valuable.
I'm guessing you'll be using IDE drives.
Some of the more expensive (usually scsi) hard disks and controllers have a battery backed cache that can ensure that your writes are preserved in the event of a power loss. The lack of this requires you to sacrifice a great deal of write performance if you wish to ensure integrity. The sacrifice is a bit less if the hard disk preserves write order, which ensures integrity to the extent that the filesystem is capable, though you'll still lose data. Combining a desktop ups with a desktop server, set up to power down safely before the ups runs out and come back up afterwards, is sometimes enough to let you sleep some nights.
The mtbf (mean time between failure) ratings for hard drives intended for desktop and server use are calculated differently. For servers, a consistent high load is assumed. For desktops, a low load and lots of sleep time are assumed. So a 1 million hour server HD might be equivalent to a 2 million hour desktop HD, and most desktop HD's are rated at like 300000 hours.
Also, mtbf is not an estimate of how long a hard disk will last, just the chances of a fairly new drive going out unexpectedly. Like if they tested new hard disks for 500 hours to weed out the duds, then took 1000 of the survivors and tested them for another 1000 hours, and 4 went dead, they could claim an mtbf of 1000*1000/4=250000 hours AFAIK. But you can be sure most of them won't last that long, that's almost 30 years at full load. Like saying if 4 kids in 1000 die between ages 5 and 15, you can claim humans have an mean time between failure of 10*1000/4=2500 years. The real estimated lifetime of a hard disk may be roughly proportional how long the manufacturer is willing to warranty it for. Hard disks intended for server use tend to be warranteed for much longer.
If you use a desktop, max out the ram to minimize disk use and schedule very regular incremental backups, as full backups will also greatly increase disk use. A desktop server will last the longest if it almost only touches the hard disk to perform necessary writes. And be aware that cheap desktops have a high lemon rate.
If you buy a Dell PowerEdge 400sc, their cheapest line of servers, you're actually getting low end desktop hardware in an easy-access case for the about same price as their similar desktops, plus integrated gigabit. So using a desktop as a server isn't too horrible, if it's not vital.
A good raid 5 file server with scsi drives, plenty of ecc ram, and a reduntant power supply can live almost forever without maintenance. They've been accidentally sealed behind walls without anyone noticing until many years later.
Funny that you mentioned google. There was some blog entry (http://blog.topix.net/archives/000016.html) (I beleive that was discussed on Slashdot earlier), where author argues that Goolge relies on high redundancy, instead of high availability of a single computer.
my sstream of consciousness
Reliability is the only difference between a desktop and a server system. If you can tolerate an outage every few weeks, go ahead and use desktops. If you need 100% uptime, get a real server, it will pay for itself many times over.
What if a hard drive dies? In a server, you pull it out, pop in a new one, and the RAID array fixes itself. The users don't notice a thing. In a desktop machine, you have to turn it off, unplug everything, open the case, unscrew the screws, unplug the cables, remove the drive, put in the new drive, put everything back together, restore the array manually, and hope you didn't lose some data. And all while you do this, the server is down and nobody can do anything.
Just keep one thing in mind. If you pay too much, nothing will happen. If you get a crappy system, you will get fired.
A coworker did something similar to what you are talking about. While it did save cash up-front, he spent a huge amount of time researching and ordering parts, filling out purchase orders, assembling systems, troubleshooting systems, returning defective parts for exchange, burning in and testing assembled systems. Looking at the total costs, it only made sense if you treated his labor as free. In a rational organization, we would have bought tested and assembled systems. But our management had a fixed budget for labor and a small budget for capital expenditures, which led them to ignore labor costs.
Mea navis aericumbens anguillis abundat
For years, Google was a giant pile of dirt-cheap
no-name PCs. Each one had two IDE drives and a
single Celeron CPU. Failure? Oh yeah, but it didn't
matter at all. The software would just drop the
broken box out of the cluster. Nobody would even
bother to fix the PCs as they died! It was cheaper
to just replace the whole cluster whenever too
many of the boxes were dead.
Now Google is large enough to get a good deal on
custom-built rack-mount hardware. It's still IDE
and cheapo consumer CPUs of course. Assuming that
your server needs are a bit less that Google's,
this option won't be available cheap for you.
I've seen some of the hardware google uses and it's not fancy name brand redundant everything servers. In fact their setups might shock some IT traditionalists. They seem to use standard mother boards mounted on open shelfs(no case), with a psu and an IDE hard drive.
From what I've read about google their philosophy is it's better to have a number of redundant servers, then one critical server.
Most of the posts have been reliability yada yada...
Here are the real differences:
Chipsets are different - and focus on throughput.
RAM accuracy (yes... there is a difference)
Built in pre-failure diagnostics
Redundancy
Hot swapable components
When you look at pressing desktops into server use, analyze the cost of downtime. Let's say you have a sales team hooked to your server - 8 users. Server is down 1 hour. Sales are $8,000/day. You lose 1/8 of your sales for the day. You just lost $1K in revenue plus your time spent fixing. This happens 10 times... you can see where the desktop gets expensive.
-- $G
If you like cutting corners then i'm sure your software will die before your hardware will.
Some ingenuity can help cover the gap. Some idea's i've had for powerfull remote troubleshooting and repair get video cards with TV out, put them into an RF modulator and take the coax into a computer do video serving. This will allow you to see what exactly the computer is doing when its not responding.
I dont know if conventional UPS serial lets you power off and power on the computers but relay boards can be wired to control the power and connected to a serial port for doing hard reboots on computers.
At trade shows I saw hot swappable power supplies for pretty cheap (around $75?). Dont know how hard that is to price out.
The biggest part of this decision is how your software can deal with computers going down. If you can setup your software for redundancy, spend your resources on that vs getting branded servers. This isn't always an option though, dont assume it is.
we just bought five new 2u Dell Power Edge 2850's for 2k each!. That included two 2.8 Intel Xeons, three 36 gb seagate scsi 10K RPM drives (can have 6 total) with a 256MB RAID controlller , dual power supplies, dual gig ethernet, and no OS installed. Thats the price you just paid for a decent workstation. It's a bad idea.
There are differences but most do not really look.
Most cheap desktop motherboards have built-in video using "shared memory" - this is actaully taken from main memory and is a constant interuption to CPU to do what it needs to be done.
Bandwidth of the PCI bus and ACPI forcing all cards to use the same interupt adding to the overhead of the OS to sort out the conflict and order. This can also lead to lockups or frozen IO - I know using 100M NIC with 100M disk controller.
Multiple processors - and I am not talking about the CPUs! Server level parts most have intellegent controllers (ie their own co-processors) This way the main CPU can get work done and not worry about the reading a disk drive.
Now: Does very server have to built to server standards? NO
A old desktop box makes a great firewall, printer server or even departmental webserver. The key here, if it goes down how fast can it be replaced? With a firewall do not build one. Build two, the second just needs to boot and be plugged in. Same for a printer server or small localized webserver.
But if you are crunching data - a database server for example - buy a real server. I like IBM X440 maxs out at 16 CPU (build sets of 4) data busses 256 bits wide not 32 or 64 of most mother boards. PCI-X slots 64bits wide and hotswapable cards, plus maxs out with these at like 100 of them. Though on VMWare's ESX and make a pile of "little white boxes" all virtually.
You have also noted about RAID cards for IDE. besure they are intellegent (Co-processors) or the CPU is doing all the work.
In the end to me real difference between Desktop / Server Class / Servers is CPU loading. How much of the "housekeeping" the CPU must perform.
On desktop machine, the CPU does it all, It watches even byte the goes into and out of a disk drive or netcard. It gives up time to allow the video to share its memory. This all takes away from the base function of running an app. At one point a few years ago - the average machine was using up to 40% of its processing just to keep the screen updated.
Server Class machines have helping processors to off load the CPU. Adding these into desktop box starts the transformation into a server - except missing true server need hotswapable everything.
I have built machines with this in mind of years - My current home machine is dual PPro 200, with highend scsi and highend video (for the time, PCI Bus) working a large database and useing database design tool - it out preforms the 3Ghz P4 I have office, with IDE and shared video. Parts do make a difference.
True Server machines are built differently, PERIOD. Look at the X440 from IBM, look at the top end machines Dell, HP/Compaq you will see the difference.
Yes, they are sell servers that are really desktops in deguess. Dell 400SC small server is the same case and motherboards as Dell 800 desktop series. The difference ECC memory, and a front cover that covers the 2 USB slot and sound ports in the front. Also you can get this for less than matching desktop configuration. I got one for my wife's desktop.
Lastly clustering...
Clustering to me is the same as raid to disk drives. Lots of cheap servers sharing the load acting as a single larger machine. So all of this may be for naught.
One thing to consider when comparing servers to workstations is support. Yes, you can build a raid-5 array (probably not hot-swappable, if it's IDE, but I digress), but what are you going to do if something on the motherboard burns out? Ordering overnight from Newegg might seem fast now, but buying a server with a support plan might have someone at your door 24/7 with replacement parts.
How much will your company loose if this server is completely offline for two days? How much less does that server cost now?
I'll create an amusing sig when I have something meaningful to post.
Rackspace is usually at a premium. Desktop servers don't stack well and each year they are made in different sizes. Sometimes half an inch more width can be a problem if you need to swap one.
Reliability. PC computers and components just aren't made for a 24/7 vibration-ridden environment. Their MTBF is probably not considered a significant design factor, as people just reboot their machine if something goes wrong.
Open the case of an IBM or Dell rackmount server and prepare to be impressed. The design is clean and modular, every cubic inch is utilized to the maximum. The box is "littered" with redundant fans designed to blow the air in one direction through the case (unlike the standard PC case fans that cause air turbulence and little cooling), clean and modular layout, etc.
Most good servers come with a main-CPU-independent remote management port where you can check the machine state, power cycle it, access the console etc. That feature by itself can save minutes of not hours of downtime.
Plus, as a business you will want to lease the servers, so the price difference won't be a lot in monthly payments.
You probably *could* use standard PCs if you cluster them or if you have them in your immediate vicinity, or if you plan to change them pretty often.
I have had a MS 2000 SQL7 box up and running on a GENERIC built machine for over a year with 2 needs for reboot -- pretty good uptime for $1300 bucks a year and a half ago as for the replacing of hard drives - HOTWAP drive bays - may not be able to do the 'hotswap' but you dont have to open the case dependant on how you have you system configured, rebuiling the drive may not be that hard -- I keep the OS on one, data on one and batch mirror that locally --
if primary data drive fails - I have a mirror from the night before
if OS drive fails - yes there is down time
if backup drive fails - remote backup till you can get another drive out there
If your main factor is cost and you don't care about redundancy, performance, support, scalability, or space - a desktop is a great choice.
If you care about any of the above, you want a server. You just have to figure out what an apples to apples comparison is. Take all your servers / desktops configured as you would need - and get some quotes for them. If the price is drastically different, that would be one thing. If you are talking a small difference - go with the servers.
Also, most servers also have Xeons, which most desktops won't have.
I don't know how the rest of the world does it, and I don't really care.
The mail server where I work used to consist of a 733MHz Celeron, branded E-Machines. It was a disused desktop machine from Joe Random (Joe, of course, has a shiny new Dell on his desk to replace it). Complete with a $3 PCI RTL8139 NIC, it was the epitome of cheap.
If any part failed, including the 175-Watt PSU, the machine would die completely.
It'd been that way since I started with the company.
I mentioned it to a higher-up, who happens to be a rather important salesman of moderate technical inclination, and whose sales depend primarily on reliable email.
He insisted that I do something about it, and so I began doing so.
I fought with the RAID adapter in a Proliant that we had spare before I realized why people generally loathe binary drivers under Linux. I looked for another way to connect the hard drives, but the box only had one(!) real IDE channel, and it was consumed by a pair of CD-ROM drives.
I sat and fathomed that for awhile: Big server box, stout steel constuction, Serverworks chipset, ECC RAM, huge cooling, 64-bit PCI, one P4 Xeon and room for a second. Unsupportable hardware RAID. One bloody IDE channel. No SCSI. The sound of nonsensical madness was deafening.
So I just built one. I had a few priorities, like redundant PSU cooling, Pentium 4 (I'm an AMD fanboy, but thermal throttling is your friend, even if the chip is vastly overpriced), redundant storage, good IO performance, and the ability to replace any (or every) part with something that can be sourced locally within an hour or so. Oh, and it has to be cheap.
I also made a list of non-priorities: Don't need a lot of number-crunching ability, don't need redundant PSUs, don't care about multiple CPUs.
"Who makes server mainboards," I asked myself. I answered myself with "Tyan."
I've never read anything but good stuff about Tyan. So I got one of their P4 boards. Not a "server" board, but one of their lesser (single-CPU) models which were hopefully developed by the same engineers. Two channels of SATA RAID, four DIMM slots, very few other built-in goodies, except for two additional PATA ports.
It supports dual-channel ECC RAM, so I picked up a couple of quarter-gig sticks of that. Could've gotten more, but remember, this is a -budget- server. (It seldom swaps, and when it does, the disks are fast enough to make it a non-issue.)
Also picked up a couple of Western Digital 80GB SATA drives, because Moving Parts Are Important, MMkay?, and at the time they were the only ones still offering a 5-year warranty. This machine is supposed to live longer than that before it is outgrown.
And for good measure, I included a Pioneer DVD-R for offline backups. I hate tapes.
I tossed it all in the cheapest black case I could find (newegg, $24, shipped). I threw away the included PSU and replaced it with a big Antec Truepower.
Killed the hardware RAID in favor of Linux's software RAID1. I have no intentions of ever marrying a computer's software to something as general and failure-prone as a modern motherboard - out-of-the-box RAID is a great way to fuck yourself at disaster-recovery time.
It runs Gentoo, and and filters and tosses mail something like twenty times the rate of the old E-Machines consumerbox (which had buried itself in backlogged mail a few times).
We've got redundancy of cooling and storage, we've got a graceful fail-safe on the CPU fan, and we've got a disaster plan that includes being able to find parts from the mom-and-pop shop down the street, or mounting the SATA drives in that wretched Proliant with a PCI controller, or (at worst) setting up the Proliant's DVD-ROM and one of its 80gig drives as master/slave and restoring from DVD-R.
I'm pleased with it. It was cheap. It went together slicker than greased shit. I don't think it's going to fail anytime soon, but if it does, at least I don't have to worry abou
Kid-proof tablet..
What is usually the difference is form factor, quality of hardware, cooling, and type of hardware.
A serverroom is usually cramped so the smaller the case the better. Or at least a case that doesn't need open areas all around it. Those 9inch racks ain't just there to look cool. It is just more efficient then stacking PC towers.
Not all motherboards/hds/fans/etc are equal. Almost all can run 24/7 if your lucky but being under full load 24/7 is a different thing. I have had zero problems with HD's except the few IDE's that against my advice were used in a server wich put a pretty heavy load on them, because again agianst my advice they had far to little memory.
Cooling is important to. Servers run 24/7 so there is no cooling off period. They also stand next to others and so can only cool through the fans, not by radiating heat from the sides. In short, PC's placed in a cramped enviroment tend to run a little hotter then under your desk. (of course all depends on your desk and serverroom)
But most important is the type of hardware. Yes stuff like hardware raid, ECC memory, SCSI, no fancy crap Motherboards, real PSU, dual/quad CPU really do make a difference on your typical server application. Will you need that extra performance? I don't know.
I once had a low budget website that while taking a lot of hits would take them on a pretty small content base. So we could spend some extra on memory and save on the HD's, the idea being that all the content would be in memory with the HD's sitting pretty much idle.
It worked fine apache cached and the HD's idled. Then the users decided since the machine worked so well they could put more content on it. Then the HD's blew up. The IDE disks could easily handle just the few OS writes and reads (no logging) with apache using a memory cache but when that became full because there was to much content to serve the HD's suddenly had to handle hunderds of thousands of hits. I was amazed they even held out for two months before we started getting weird errors.
Next to it an other customer has a properly setup hardware raid SCSI and it handles a far greater load while barely being stressed.
It is not that the IDE disks were bad. They performed extremely well in fact. Until some idiot decided to use them for something they weren't intended.
So it all depends on what kinda use your machines will see. Just remember one thing, smarter people then you have been trying to save money ever since the bubble has burst. Yes sales of PC servers has gone up and sales of Server servers has gone down. There certainly were more then a few companies during the bubble who spend way to much on their server park but equally there are now companies experiencing downtime because that cheap PC box just ain't cutting it.
Buyer beware.
MMO Quests are like orgasms:
You may solo them, I prefer them in a group.
I've been doing this for more than a decade without the horrific failures described here (can't remember the last failure, it's been so long). But it's strictly low end and I'm only serving 10 users at the moment (has been as many as 35). I'm serving office documents and AutoCAD files almost exclusively. While not as sexy as saying I have "Server X", I'm quite happy with a dual CPU linux box (Netmax, so anyone can figue out what to do when I'm out) and I backup to Windows boxes with the bundled software.
My server is an older Intel server board with dual PIII-600s (as are some of my older workstations) and a 10/100 NIC (due for an upgrade, though, so we can go gigabit for the CAD machines - to match our new color printers). The server and workstations are all built with similar Antec cases, so I can maintain one stock of spares and change drive clusters faster than you can blink (and still with no need to do that).
Full backups of documents and drawings are done nightly, about 20GB total. Static files (photos, etc.) are backed weekly and monthly. Any time I want, I can have redundant backups, just backup to another workstation with $50 of spare drive space.
Servers are not magic. Pick the criteria that suit you and build/buy the machine and OS you need. Redundancy is a given. After that, no one but you will give a damn if it's sexy or not. They willl care if it means they get a better workstation out of the deal, though.
This is how most IT departments start, and it's a normal process of evolution.
... or it may not. I have seen a couple of
In the beginning, there isn't much money available, so most places cobble together 'servers' from spare desktop components, and throw them up in a closet somewhere. That generally works okay, and the company realizes that they like having servers, so over time, the installation grows.
As it gets bigger, the lower reliability of desktop components will start to become apparent; servers will go down, hard drives will fail. It's just statistics; given enough samples, the lower quality of the cheaper components will start to make itself felt.
Gradually, as IT departments grow, they tend to migrate towards better and better hardware. The really big outfits tend to use Dell and Compaq. Compaq in particular sells very, very expensive machines, which are very well engineered and hardly ever break. But you pay through the NOSE for this kind of service.
So how do you know how much to spend on your servers? When you gain the ability to numerically measure how much it costs you when they fail. When your department and company mature to the point that you can accurately measure costs of downtime, then with management's decision on acceptable risk levels, you'll have a pretty good idea of what you should be spending on servers. Many big companies find that the cost of downtime is appalling, when they actually are able to measure it, and that the cost of even very expensive servers is minimal in comparison, so they buy the best stuff they can find.
But until you can measure it, IMO you're fine with desktop components, as long as you buy GOOD ONES. Don't skimp on your drives, and make sure you have good cooling for them. Buy server cases; you can get good ones for a couple hundred bucks that will hold a billion drives, and then make sure to buy good cooling; you may want the boxes that mount 3.5" disks in 5.25" slots, with fans and hotswappability. I usually buy PC Power and Cooling power supplies for servers; even the Silencers are fairly loud, but they are very robust and well-built. Many of them are dual supplies in one box, which improves reliability even more. That's a lot of fans in each machine, so you may want to pick up a spare or two with each machine you buy. (Tape them inside the case). And the noise level, particularly once you get a number of them, will be high... but think of it as the sound of reliability and you won't mind it too much. Also note that when you get past a few machines, or if you spend a lot of time in server rooms, you should wear ear protection. I have worked in big colo facilities that were absolutely deafening, to the point that things sounded muffled when I left. That kind of noise DOES DO DAMAGE, and you want earplugs.
Make sure you understand exactly what onboard network chipset you are buying: you most likely want an Nforce3 or an Intel, um, 865 or better, I think it is... where the network card is directly on the northbridge, so you can get the true gigabit speeds. When they are on the Southbridge, and look like they are PCI devices, you can't run gigabit full out. And never buy a motherboard that uses Realtek 8139 networking, they are garbage. They make the CPU work way too hard, and are NOT good for server machines.
What you will end up with is a whole room full of Frankenclones, but if you've been smart and spent your money on good stuff, it'll be almost as reliable as the Dell/HP/Compaq/IBM clusters for a tiny fraction of the price. And you'll be able to get replacement parts anywhere. But you probably WON'T have spare parts on hand to fix things, unless you've been unusually clever in your design, because each new generation of machines will be different than that last, and you won't be able to use the same replacement parts interchangeably.
Someday, when you find out what downtime costs you, the extra cost of the big label servers may suddenly look wonderful
By the time you've bought a desktop with all the high performance, high reliability options you'd need for a server, you've bought a server.
Until last year, we had a very good run with using pretty standard machines as linux web and file servers that were accessed constantly over a LAN. The only things that needed replacement were harddisks (so ensure you perform nightly backups to another machine on the LAN), and the occasional birthday present of extra RAM or bigger harddisks.
This year we noticed Dell had very good rates for renting their rack servers, so we grabbed a couple, and will upgrade them on a 18-24 month basis. The affordability of these makes them much cheaper than buying a desktop machine, and the Dell warrantee/support has in our experience been sufficient when we've had problems (e.g. another damned harddisk crap-out).
So now, we have the leased rack 2.8ghz servers for our webservers, and our trusty P~500s still keeping up fine with our file serving, mail, and routing needs, etc (Thanks Linux... ).
I have a PII which is as much as server and a quad xeon (works fine as a webserver, no downtime in the past year due to parts (only had it for a year)).
then again, if you spend a couple of days more tinkering with these machines than you would the servers, you've lost your money.
There is no distinct line between server hardware and desktop hardware. A lower end server is easily build from decent desktop components. The bottom line is: buy good hardware.
Don't skimp on the harddrives, but go for reliable ones. SATA Raptors are as reliable as many SCSI drives, and go in any modern desktop. RAID5 them. RAID5 in software isn't much of a CPU hog in modern machines. RAID5 in hardware is faster, but more expensive. Fit to budget.
Hotplugging SATA is not really supported (tested) in Linux, but expect it to mature. When a drive fails at this moment, downtime is unavoidable. In the near future, expect this to improve.
As for the mobo, memory, network, case. Get quality stuff, but don't go overboard. Onboard vga is fine for your purposes: it will act as a server.
Depending on your needs, backup media need to be considered. Put DVD burners in the server. Backup often. When you need more storage, portable harddrives are great. You need more than one.
Most important: (stress)test your equipment before putting it to use. Most broken hardware is broken from the beginning. Failing hardware is much less likely. The biggest difference between so called server hardware, and desktop hardware is the amount of checking it had before it leaves the factory. So do that yourself.
the pun is mightier than the sword
You are not considering the actual cost of owning and running these machines, only the initial cost of hardware. If you learn how to do a proper analysis of the costs associated with each machine over a 3-5 year period, the typical server lifespan, you will find that purchasing an entry level server will be far less expensive. Better memory (ECC), server chipsets (Intel 7xxx vs Intel 865 for example), and chassis designed to provide adequate airflow for a server is a bargain compared to downtime while you fix your Dimension "server" every couple of months.
You can do a 1U P4 3.0 with mirrored Enterprise quality SATA disks and 1GB of ECC RAM for well under $2000. Take a look at the Intel SR1325TP1-E server platform. It's the server chassis with proper cooling with an Intel TP1 board installed. The board has dual onboard nics and the chassis has about five fans. Very nice, and runs $500. Add the CPU for about $200, memory, and disks (SATA, CD, floppy) and you are done.
What was this?
"Backup to a USB hard drive."
Do they make those in 64GB versions now? No? I'll just use another RAID array then, thanks.
You are unaware that you can put an external USB case on any 3.5 or 2.5 IDE or SATA hard drive and you are giving advice? 400Gig drives are standard now. Where ya been?
There is no Raid '0'. Read the ACM paper. Mirrored drives are Raid Level 1. And it's expensive to do this. Raid 5 is likely a better choice, but requires 3+ drives.
---- Teach Peace. It's Cheaper Than War.
Think for a minute. You can save lots of $$$. And you'll end up with trouble. Servers are built differently, and architecturally. They use SCSI instead of IDE and ATA/SATA. They have PCI-X bus instead of PCI. They use GBE natively, not 10/100. They usually have several NICs instead of one. They have more pre-fetch cache, and more lucid FSB. You can buy toys to do real work, and they'll last....for a while. If you've got a high duty cycle, then buy real hardware. You can't expect to get Peterbilt performance from Camry hardware. You'll be sorry if you try. Your time is worth money, and mucking with cheezy hardware isn't a good idea. This doesn't mean that you can't find 'enterprise' quality hardware from inexpensive sources-- you can. Just don't expect toy desktops turned into workstations to be worth much. Linux doesn't help hardware problems, and neither does FOSS. There's a real architectural and life-cycle cost differential between slam-it-together desktops and server designs. Don't be fooled by the seeming cost differential.
---- Teach Peace. It's Cheaper Than War.
The simple answer is the following
1)How much downtime can you afford due to lack of hotswap etc
2)Can the desktop box do the job? If your trying to do some massive process, the answer might be no
Lets face it - there are a LOT of "Mom and Pop" shops where if the server goes down for 1/2 day - it's not a major problem (Heck, I've worked at software shops like this - just keep working on what you already have out). Other places, your down for 5 minutes (of even 60 seconds) and the phones will be ringing (where I am now) - heck, with the project I'm on now, on "production day", there will be:
The Programmer (me)
The DBA
The Server Admin (or 3)
Network Infratructure Tech (or 3)
Union electricians
Consulting DBA
Microsoft Level 2 Support
ALL sitting in the server room, watching the box(es), network traffic, server processes, database states, etc for the whole day. If we have more than 15 seconds of downtime at the wrong time, heads WILL roll.
The system is spread over 2 cities, 2 or more redundant servers in each, in sepearate data centers, being fed by T1s that go by different routes to the data source (yes, you can find out the routes, and pay for different routes)
We'll use the system for 2 days, and then all the hardware gets put to other use, and we do it again in 2 years - 6 months of testing, 3-7 days/week, all for one day
-- 73 de KG2V For the Children - RKBA! "You are what you do when it counts" - the Masso
"a couple" = 3 (higher bond.)
3 days for $1500, = $500 per day.
365 * $500 = $182,500 per year IT labor budget.
If the equivalent of one full-time six-figure person spends three MORE days configuring redundant desktops instead of redundant servers, either it's a horribly big network for that one person (if each replacement takes two hours--which is a lot--that's twelves replacements per year!) or they're incompetent.
next time I post to /. I see the need for more specifics --
I am more considering purchasing parts and rolling them out as 'servers' as opposed to just using DELL GX-WHATEVERS...
This way the memory can be spec'd, the motherboard can be spec'd - etc....
I do find it a bit funny that many people say 'SERVERS HAVE SCSI DRIVES which make them reliable' - however DELL sells a slew of servers which only have IDE drives and arent setup in any mirror fashion at all -- now how can those be any more reliable that a tightly built machine - which may have a HARDWARE RAID card?
That's sad. So let me guess, they probably have better benefits than any of the high-powered "IT professionals" on the team?
Nobody seems to have talked about ram types yet. So here is what my mentor told me back when I was getting started in the business:
Non-Parity RAM, if it goes bad, starts making freaky, head-scratching, transient problems happen with various software. The machine just keeps running in this errant way, and it is up to you to notice and find a fix. That isn't a problem if you are just on a one-person/ hobbyist/ desktop kind of system.
Parity RAM, if something goes wrong, screams bloody murder and shuts the machine down. Maybe I should say the motherboard does the screaming, since you have to have a mobo that does check the RAM's parity. You get no haunting, mysterious problems to hunt down; but you have a dead machine until you fix it.
ECC RAM [error-correcting code RAM], together with a compatible motherboard, screams at you when something goes wrong, BUT KEEPS DOING ITS JOB giving you some time to get set up to fix it! There may be a 30-50% cost premium, but the reliability makes ECC a really nice thing to have in servers, if by "server" you mean no-downtime-allowed.
If no-downtime is the goal with your servers, then yeah, those hot-swappable features in PSU, CPU, and IO cards are really nice. So is UPS protection. If fast network IO or huge storage space is the goal, then of course we are talking about whole different priority sets: Then you are asking whole different questions.
--And you don't tell us what manufacturer that was. I for one want to know that particular detail, so I can watch for their stuff. Without that info, the post is vague, and not really worth anyone reading.
Cheers
This comment is aimed at "production use" -- for "test/development" (non production) machines, please disregard.
While an HP/Compaq "Proliant DL380" at around $5,000 with a 2nd CPU, redundant fans, RAID hard drives, etc. is a _lot_ more expensive than a $1,000 white box with a couple of IDE drives with software RAID, it tends to be worth it. At least in my situation.
I've used white box servers in the past, and they are fine while they work. Once something goes wrong you're sort of on your own to track down the problem, find the original vendor (and your reciept), wait several weeks for waranty repair or (more likely) purchase a new motherboard/power supply/hard drive/whatever to use while you wait for the replacement to come back.
The biggest problem is troubleshooting time. The Compaq servers have excellent integrated logs, diagnostic capabilities, and in general it's easy to isolate what's wrong and if it's a bad part, proprietary though it may be, you generally have it by the next day.
With a white box, you (or your reseller) integrate the system, which _usually_ is OK, but if you get a sometimes-flaky RAM chip, or a bad trace on a motherboard, it's easy to fix by swapping it out, but very time consuming to take the machine apart, replace your best guess, put it together, try it, then repeat. With a "real" server the quality is generally _much_ higher, so you don't get most of these flaky problems in the first place, plus the rack mount cases are designed to keep repair times minimal. You can pop open a DL380 and replace a fan without even shutting the machine down. The power supply can be swapped out without even opening the case (and has a nifty LED on it to indicate if it's got a problem), and If you spring $200 for a redundant power supply you don't even need to shut down first. Ditto for the hard drives.
What it really comes down to are 3 things:
1: Is the service provided worth anything? If not, then why bother doing a server at all? If it is important, how many days a month can the system be unavailable?
2: If the system is worth something, and downtime is not acceptable, can it be clustered (e.g.: DNS server) or is it best done as one machine (e.g.: SQL server)?
3: If it can be clustered (and will be -- many things _can_ be clustered but only at enormous expense), then using crap hardware may be acceptable as long as you have 2 or more running, the failure of 1 is tolerable. If it can't be clustered, then you really need to think about "real" server hardware, penny-wise/pound foolish, and all of that stuff.
Is this something of your own initiative or are you being pressured to do this? If the pressure is on, then you are being set up for failure. If you're forced into it get your resume polished and/or try to set someone else up to take the fall (which is unlikely to work since we're talking about $2-4,000 -- chump change for all but an extremely small business where no one else is around to be blamed).
"But actually trying to use m4 as a general-purpose langage would be deeply perverse" --ESR
Two years ago I'd have told you to save your cash and go with a desktop. I'm now two years into running my own consulting firm and my attitude has changed a bit. Here's the question you must answer: When your server goes down, how many people will be idled by the failure? How much do they cost (remember to mutliply their hourly rate by 3.0 to incoude overhead and profit, plus the shutdown/startup inefficiencies)? How long will it take to fix the server?
X * 2.5Y * Z * #Failures* / Serverlife is your breakeven point. I'd figure 1 to 3 for faiures per server life if your feeling lucky. Also, without knowing your industry, you're 3.05Y iwill probably range from $25 to $90/hr for the rabble. If your CEO is using mission critical stuff from your server (his stock quotes, for example), he's worth a zillion dollars.
So if you plan on saving $3000 by going with desktops, and with middle of the road numbers ($60/hr, 2 failures), you'll need to take less than 12.5 man-hours per failure to fix. 25 employees? That's 30 minutes to troubleshoot, replace, repair any software hiccups/reinstall drivers/regenerate the drive, reboot, and inform the company that the coffee break is over. (That last part will take about 45 minutes all by itself)
It's rarely worthwile to save money on inferior equipment, unless you have the funds/time to deal with the lower reliability. Though I'm not fully fault tolerant, I've found that it's worth while for me to have a live backup HD for my OS (refreshed about monthly), and weekly backups of my working files. I can't afford more (manpower) and I can't afford less (recreation of work).
*If your server fails, it will be because you chose desktop hardware, and you will not be able to convince a soul that a real server would also have a non-zero failure rate.
Is it just my observation, or are there way too many stupid people in the world?
Reid Maynard wonders: "I work in an environment where the dollar is more than almighty (who doesn't?). One of my cost savings plans is to replace the desktop computer with an abacus. They cost much less, the parts are readily available and/or interchangeable - as opposed to waiting for overnight proprietary or obscure parts from a vendor, and so on. -- is there something special about a desktop computer other than looking at the blinking lights and feeling special?"
-- www.globaltics.net
Political discussion for a new world
I used to use a tool from Compaq called CIM (Compaq Insight Manager). You would get warned if memory or a HD was on its way to a failure, or if a whole server is down. You could also get the utility to page or email you with trouble. Just this feature was worth its weight in gold.
...and press a desktop into server duty, the one area not to skimp would be the hard drives. CPUs, RAM, and motherboards don't typically blow out on a desktop any more often than they do on a server. I think hard drives will be the most likely point of failure.
:-)
1. Go with a mirrored (RAID-1) setup at the very mimimum. If it's anything but the VERY lightest-duty server (a fileserver for a 4-person office or something) invest in something besides consumer-oriented 7,200RPM IDE/SATA drives. Go with SCSI (it's cheaper now) or with those SATA drives that are designed for server use. Simulate drive failures so you make sure that you can recover.
1a. You may also want to experiment with power-saving features on the drives, depending on your environment. Most OSs have an option to spin down the drives after N minutes of inactivity. This will also prolong their life. Obviously there are servers where the spin-up delay would be undesirable. Depends on your situation.
1b. Make sure you have gobs of RAM for whatever task you're doing. If you start thrashing the disks and hitting swap space a lot, it's going to take a heck of a toll on the drives. By the same token, when you have so much ram that everything's running out of memory, it's a beautiful thing. Nothing like seeing a busy database server with only an occaisional flicker of HDD activity, since the entire DB fits into ram.
2. Make sure cooling is adequate. Do some stress-testing that keeps the CPU and disk usage pegged at 100% for at least 24 hours or so. Cases with 120mm fans will move more air than cases with 80mm fans. The 120mm fans will also run more quietly, and will run at lower rpms for a longer lifetime. Try disconnecting one of the fans and seeing if the server remains stable, because the fans WILL die at some point.
3. Test your backup plan. Simulate failures and restores.
Obviously, any of that advice would apply to "real" server hardware as well. You just have to do slightly more work if you're rolling your own servers because they might not come preinstalled with backup software/hardware or cooling.
OtakuBooty.com: Smart, funny, sexy nerds.
Lieberman's go-l site had some absolutely smoking hot boxes. I thought these overclocked hotrods were toys cobbled together by rich highschool and college nerds but here they are, 64bit, liquid cooled, 5 and 10 GB of ram and a host of RAID options ALL OFF THE SHELF. Also, though not strictly needed to run a server, a 92 inch display [4 video cards!].
Oh, you have budget? never mind.
well, gotta go, there's drool all over my keyboard.
SLASHDOT: news for people who can't concentrate on work or have no life at all and got tired of yelling back at the TV.
Onboard VGA is a NO-NO.
It wastes CPU cycles by stealing it memory from main memory. Buy a cheap PCI VGA card. A great improvement all the way around.
Only if you're running in graphical mode and/or using hardware 3D. Neither of which is likely on a server (unless you're using Windows, I suppose).
Besides, there are plenty of motherboards out there with onboard 8MB ATI Rage3D and similar to counter the criticism you made.
--
No protection is provided; you just pick up speed. The ACM paper, the only metric/standard that can be applied here, is pretty specific on what constitutes a RAID. They start at 1, mirrored pairs, and stop at five (non-parity, redundantly stiped). If it's not as described, it's not a RAID. The terminology has been abused at other levels, too. There is no RAID 6, just a general agreement that a hot spare is available. That said, there a lot of 'lesser name' 1U/2/U/XU servers out there with cool storage schemes. I'm sitting next to about 20 of them from HPaq, Gateway (!), Apple, and so on. Brand names are no general assurance of quality these days.
---- Teach Peace. It's Cheaper Than War.
I've run Mac OS X Server on a 350 mHz iMac that provides files for 30 clients at a time. ANYTHING can be a server.
You're naming the specs for the G5:http://www.apple.com/powermac/
Perhaps this is why there are four of them across the room, running with four HPDL360s, a Gateway, some older Compaq servers, and three ancient VA Linux servers, all humming away, to my immediate right. It's deafening, and they work 24/7.
---- Teach Peace. It's Cheaper Than War.
He's talking about VGA built into the chipset - typical examples would be Intel 810/815, etc.
These really do reduce performance; I've found a 15-20% drop in memory bandwidth to be 'normal' with such a chipset.
Obviously, motherboards where the video is on-board but not integrated into the chipset do not have this problem (such as the common ATI Rage/Rage3D/whatever config popular with server manufacturers).
So was I, in my first para.
These really do reduce performance; I've found a 15-20% drop in memory bandwidth to be 'normal' with such a chipset.
Even when running with a text console (i.e. not X, or Win32, or even fbconsole)?
--
This is what i just did.
$450 Dell sc400 server (2.8ghz, gigabit, serial ata, etc)
$60 ECC 256mb DDR400 ram (cheaper to buy at newegg)
$200 2 Serial ata 36gb 10K RPM raptor drives. run them in software raid 1 with Fedora.
$100 UPS.
Buy one more system than you need so that you always have a spare machine for parts or whatever. You wouldn't go around driving without a spare tire would you? And at prices like this, there's no excuse.
Of course if you have a few dozen servers, go with blades/rackmounts. The above is certainly not as scaleable.
I agree with all but one thing you said
not last and not least: having someone working on a server is probably most stupid idea in the whole IT. Whatever OS you use, beleive me, users will find a way to devour 98% CPU time and 99% of memory.
I think the author was talking about using a desktop type box using standrad desktop components, not actually a desktop machine on someones desk doubling as a server
"goodbye and hello, as always" ~Prince Corwin, from Zelazny's Amber series