Server Stations

[vtohthree]

When exactly is a server system pliable or necessary? When is the line drawn between getting an Athlon series or switching over to the Opteron(or pentium/xenon)? What is the main dividing factor? I am asking out of curiosity, since I obviously do not know. However, I am talking about a business machine, I would assume that you would not build a server station for gaming, word processing etc...

If a user needed the power to have a computer run 24/7 with the cabability to communicate with multiple computers simultaneously would it be pliable? For example, I would assume that Dell uses such a setups(enlighten me if I'm wrong) in order to accept orders world wide 24/7. Also media companies that source streaming video files, or webhosts that sell webspace for the world to view and access.

What distinguishes a server cpu from a personal desktop cpu? Especially since both now have the option for multiple cores.

I have only made speculations and assumptions, but in reality I do not know much about server stations, at all. Could some one clarify the function and need of a server system?

 

[Treyshadow]

Server CPU's and setups are really made for high availability. That is truly the number 1 distinguishing difference. Your desktop only needs enough horsepower to be available to you. Most server machines are built to be available to a certain threshold established prior to purchase (hopefully) for a given task. Most are built to stay on 24/7 for access for any user in its environment whether that be a corporate LAN, a WAN connection like web access, or a small business website.

The subsystem of a high end server system show the largest contrast between a desktop machine. Things like fiber channel arrays, SANS, and large scale backup mechanisms like the Dell Powervaults show that the data on these machines is critical and must be available.

Now many companies make low end servers, and if you look at their specs, they are nothing other than desktop machines with a good motherboard chipset. Many server duties do not require high end processors. Things like DHCP, file sharing, and print serving do not require dual opterons unless the quantity of devices being "served" is extremely immense. Also when you have lots of devices, a better method is to increase availability by having more servers. RAM and hard disk space are the 2 largest factors with a server. Input and Output of data when required is the majority of a servers duties. Databases, web sites, file storage, Infrastructure connectivity. Those are the jobs, and thus the requirements. So most servers are sized for their duties prior to purchase and many desktop machines could qualify, but as the tasks are raised, a true "server-class" machine is the logical answer.


You are correct, you would not build a server for word processing (well unless the server is a citrix / terminal services server based computing machine for 50 or more users). However you may use server class hardware for things like 3D artistry, high-end math computational designs, telemedicine, mechanical design or statistics. Companies like BOXX create machines specifically for these purposes.

 

[phisrow]

Treyshadow is right in what he says. Also, reliability is a big factor.
Desktop/light workstation gear is pretty good, for the money, these days; but it isn't terribly reliable in the face of component failure. Maybe, just maybe, it'll have hot swappable drives in a RAID setup; but that'll be it. If you lose the CPU(or even one of a couple of CPUs), some RAM, a power supply, or something of that sort, you end up having to shut down and swap the part out.
Servers often have superior options in some or all of these respects. Hot swappable redundant power supplies are pretty common, and hot swappable RAIDed drives are practically universal. As one climbs higher into the heady realm of Really Expensive, features like the ability to resist RAM and CPU failures become an option. Status indicator displays, networked managment watchdogs, and similar features designed to give early warning are also often seen in servers and rarely in anything smaller.
There are also the basic diffirences in design choice. PCI-X means server or workstation, as does ECC RAM support. A machine that costs 10,000 dollars and has 8 megabytes of embedded video RAM and no AGP is usually a server(or a decade old). At least for the UNIX crew, a BIOS with decent serial terminal support is also a good sign that the machine is designed as a server. Rackmount almost always means server, very rarely workstation.

 

[vtohthree]

Ok, some light has certainly been shed over some dark spots I had. First of all, thank you Treyshadow and phisrow for the input. However, I still do have a few more questions. Now that you mention it, I remember back when I was researching RAM, I came across ECC and Parity/non-parity support. What is it(I assume that it is considerably more expensive than workstation RAM)? Also, when you mention the term "hot-swappable"(a term that I've heard quite a bit), do you mean it is possible to replace the part while the computer is still running? Last, but not least, what kind of OS's are commonly used in server stations? I have heard of Unix and Novell, amoung others being used, and I would also assume that Windows XP can run on them as well.

 

[vtohthree]

bump

 

[Xpage]

Dell has a really good deal Here for a sever. You just have to provice the OS

 

[phisrow]

Yes, hot swappable means that a part can be replaced without taking the computer down. Usually some mixture of physical, electrical, and logical design, i.e. easy to quickly take out in one piece, not going to damage anything else if you do so, supported by redundant components so the system won't crash during the swap, and so forth. ECC and Parity are both techniques that allow RAM to resist introducing incorrect bits due to hardware error. Both are, indeed, rather more expensive, as they require more chips per effective megabyte of RAM, and they classify the buyer as someone who really, really cares and will pay for it. If memory serves, ECC RAM can identify and correct up to one bad bit per byte, while parity can identify; but not correct(I'm a touch fuzzy on the diffirence).

As for OSs, it depends on the architecture. A fair few servers are basically just x86 PCs in rackmount cases. These usually run either a server variant of Windows, Linux, and maybe one of the Unixes or BSDs. These would run ordinary WinXP entirely fine; but there is rarely much point, as XP isn't a terribly exciting server OS, and servers aren't generally very exciting desktops(usually designed for cooling over silence, rackmount over tower, etc.). This category includes the vast majority of single processer servers, except for a few PPCs and Itaniums, and similar. It also covers, though to a much smaller degree, some reasonably serious hardware. 2, 4, and 8 physical processors(which with dual core chips can mean twice as many real cores, if your wallet can take the pain) can all be had in a single box that will run ordinary x86 operating systems, although you might not like the licence fee you'll pay for a version of Windows that will talk to 16 cores.

Most of the other architectures won't run Windows, at least not very well(there is a version of NT4.0 for Alpha; but you'd be a bit silly to use it, these days, and Itanium kinda-sorta runs Windows); so they tend to have some favoured Unix variant provided by the company that pushes them, as well as running Linux and BSD to a greater or lesser extent. This includes the low end PPC servers from Apple, and their OSX, the higher end PPC stuff from IBM, with AIX, Sun's SPARC, running Solaris, HP's ALPHA, running HP-UX, TRU64, or VMS(not actually a UNIX, but another of the old-school, rock solid flavour.)

Most of these focus on supporting rather more processors than do the x86 servers, and are much more alien to the day to day x86 experience, having, often, been developed for quite a while largely independently of x86. They aren't just minor changes to what is basically a standard PC, they are a whole diffirent animal(e.g. Usually no BIOS in the traditional sense, diffirent endianness, limited support for things like PCI, etc.) There are also a few examples of weird hybrids. Usually, this happens when one of the manufacturers of a specialty server and/or supercomputer line realizes that, while x86 is a whole is a pathetic toy compared to what they have, x86 processors are brutally fast and shockingly cheap. Cray, for example, went this way. They use a custom architecture, which offers useful features that normal x86 was never designed for; but employ ordinary Opterons to do the crunching, as economies of scale and vicious competition have left those among the cheapest 64 bit CPUs.

 

[vtohthree]

Ok, I see, that really helped a lot phisrow, you sure know your servers enough to educate someone else. I appreciate it. Now, I have to decide whether or not it will be worth it for me to purchase/bulid one to accept queries and forms 24/7, along with some other networking features.