No stickies on servers? Boatloads of newb questions

[Dustmann]

Ok, first off, I don't even know if this is the correct forum. Sorry for all of the questions, I'm sure someone here can answer some or at least point me to a helpful source of info. All I end up with on a google search is hacker-proofing servers and saftey measures, nothing really on what a server is.

What are servers for? And why do I need one? Maybe even what IS a server? My girlfriend's sister and her boyfriend have a computer business that repairs/builds computers and also is an ISP. They have expensive servers that require insurance, but I don't know what for? Also at my college, there are a few expensive servers. Is this a different server than I would have on my home network? (at least I've heard mention of a server at home) What is the difference between an ISP server, a college server, a web server and a home network server?

I was told that there is a server for email, for blackboard, for the library, etc etc. And that they have huge multiple psus, cpus and TBs of storage. But no one has told me what exactly it is that they do. Do they house all of the main programs that all the computers use, like a big FTP filesharing thing? (My knowledge on ftp is also severely limited )

[mbossman2]

There are 2 main kinds of servers:

File (Storage) servers
Application servers

In both cases (and this is one of the differentiations between servers and workstations), they run specialized versions of the operating system. These specialized versions are designed to deal with multiple simultaneous users, runing mulitple simultaneous tasks with minimal preformance impacts.

Other than the O/S, servers differ from workstations in that servers have:
Increased memory - the more people, doing more things, requires more memory. I have seen servers running 64GB of RAM
Increased storage - as you pointed out, terrabyte storage is not uncommon
Minimal video capabilities - servers and their applications are not, generally, video intensive, so the 256mb SDRAM card running the latest and greatest video accelerator and chipset is a waste of money.
High performance network adaptor(s) - the standard linksys PCI 10/100 ethernet NIC is not going to cut it here (too processor intensive, s-l-o-w and copper rj45 only). Server NICs are much higher performance cards. They have specialized processing engines on board (takes the load off the CPU(s)) and are available in several different kinds of connectors (rj45, multimode fiber, single mode fiber and the like).
Multiple CPUs - like the memory above, many people, processing many things requires more processing power. dual, quad and up to 16 CPU boxes are available.
and the biggie
Redundancy (aka high availability, H/A, features) - Because servers are the central point in a network, down time must be minimized. So H/A servers incorporate some cool features:
Multiple power supplies - one goes pop, the other kicks in
RAID Arrays - multiple disks that can, in some cases, rebuild 1 of there array members in case of failure
Multiple NICs - ditto like power supplies
Hotswap capabilities - if a drive goes bad or nic card goes south or a power supply emits the death smoke, they can be replaced without shutting down the system. Pop the drive cage open, pull the drive and plug in a new one, NIC card? turn the power off to the slot, out with the bad, in with the good etc.
Integrated management - some servers have secondary computers on board that allow the netowrk managers to go in and reboot the main server remotely or troubleshoot an unresponsive server thru a "PC on a card".
Battery backups both internal and external - internal for the RAID controllers to allowed for a controlled multi site shutdown of the RAID array. regardless of the status of the external power and backup systems.

Now, file servers are just that: they are centralized storage devices for many of the files on the network. their O/S and internal design is optimized for the fast identification, retrieval (or storage) and transmission of requested files

Application servers are servers that are dedicated to one (or more) applications - Web services, database applications, email, collaboration, voice services (phones and voice mail), video servers (storage and streaming), etc. their O/S's and internal architecture are optimized for what ever application is being run and that can vary quite a bit depending upon your specific application. A video streaming server will be set up quite a bit differently than, say, an email server.

In a good network, you will have a mix of both: a server acting on/running database queries against another file type server, another running/controlling the video stream of the video file stored on another server, a 3rd running your web services touching all of the above. then you mix in email and security services (AAA - authentication, accounting and authorization) and you have quite a few things going on simultaneously and all of that requires processing power, memory capacity and storage.

Oh yeah and lets not forget the network infrastructure to support all of this, with the same level of performance (probably better for growth and scalability) and redundancy so the networkinghardware is not the cause of downtime or congestion.

Most of what we talk about here is the small sides of the network (residential, consumer based networks and small business networks) but there is, as you have seen just a glimpse from your family ISP.

For the consumer/small business it is quite possible to run a big beefy PC as a "server" - the demands on it, are generally pretty low and a regular computer (with the networkable O/S installed and running) is fine (and cost effective). But as you grow and scale, the demands upon the server (traffic, storage, response times) have usually grown and the device(s) must grow along with those demands, otherwise the network is slow, unresponsive and unreliable.

Servers can range from $1000 (sometimes less) up to the "how much you got?" range. so insurance on the higher end ones is quite normal. But there is another kind of insurance that you don't see: loss of revenue insurance. a server goes down, despite your best efforts and preparations, you can buy a policy to cover the revenue that is lost (as in the case your ISP), and the cost on that is commensurate with the amount of redundancy that you have in place.

HTH

[Byte 2.0]

mbossman2, you almost left me spinning on this. I am going to sticky this for now and see about keeping this around.

[Dustmann]

mbossman2 - Thanks for the well organized informative post! You answered many of my questions. I do have a few more now after absorbing some of that.

So then a workstation is the individual PC that I would find in a computer lab, or say my home pc? How do the servers communicate with each other in a multi server setup? Are they part of the network like a workstation, connected via say a gigabyte ethernet port?

How does the network infrastructure affect the servers? If a whole network is setup using 100mbps cabling/switches/etc, and the servers are capable of say a 1000mbps, does that 1000 get divided up among the 100mbps cabling, or is the whole system limited to a speed of 100mbps. I am just making up some numbers from what I see on my computer and (what I think would be) a typical network, is this childsplay for a real server?

[mbossman2]

workstation (in this case) = normal PC.

How does the network infrastructure affect the servers?

directly affect the performance of the server? no, but remember a server provides a service and the experience of using that service (either positively or negatively) is affected by the whole data system: Server, network and workstation.

If the request for the service is delayed/garbled/lost by any piece of the system, the whole experience suffers. For example, using your example above: all PC's (100mbps) and connected to a server (1000mbps), the maximum number of PC's that could simultaneously access the server (all things being equal) is 20 (20 x 100mbps = 2 gbps = a full duplex 1gbps connection). What happens if more than 20 attempt simultaneous access? the traffic gets buffered (hopefully) and held until there is available bandwidth to forward the request. Now scale that to hundreds (or thousands) of users across multiple locations in multiple countries.

How do the servers communicate with each other in a multi server setup? Are they part of the network like a workstation, connected via say a gigabyte ethernet port?

servers are connected to the network just like any other network device. It is the how and the where that is the important part.

to fully explain this requires a quick lesson on network architecture design (Don't worry, it sounds worse than it is):

there are 3 general areas of a network:
Access (or edge) layer
Distribution layer
Core layer

the access layer is where 'edge' devices are connected to the network (mainly PC's and printers)
distribution layer is where the commonly accessed network resources (servers, internet routers, firewalls, wireless access points etc)
core layer is where all the heavy networking decisions occur (overall network policy enforcement and the like - don't worry we won't play here, this is where the big boys play and the discussions here are the kind where people don't even understand the question, much less be able to formulate the answer)

Now, as above, servers will happily sit in the distribution layer and the switching gear here is a tad different than what exists on the edge. The switches here are more powerful, internally faster and, more importantly for your question, have higher transport speeds. On the access layer, 100mbps is king, with 1000mbps going upstream to the distribution layer. In the distribution layer, multiple 1000mbps are bonded together for interswitch links and the backplane speed of the switches here are very very fast (ranging from 10gbps up to 720gbps). Now remember when I said that a server may have more than 1 NIC? well here, a server most likely does. 1 NIC for PC access and 1 for server to server connections. so servers connected to the same switch, leveraging the big backplane can tennis information back and forth between themselves at a much higher speed than PC to server or PC to PC connections can support. (it's fun pulling back the curtain here and seeing Oz).


A network is a system and process and it all must work together to provide what is requested to the user in a smooth and timely manner.

Now, there are ways to have fun and tweak the whole system to make this happen quicker and more efficiently. These tweaks, BTW, are some of the huge differences between 'SOHO' gear and 'Enterprise' gear. The needs of a 'small' network providing basic file/print/internet sharing services are completely different than those of a 'large' network providing, in addition to file/print/internet sharing, things like load balanced content delivery, voice/video content, telephony services, web based applications ranging from online ordering systems to employee expense reporting to meshed video conferencing.

So the servers, the switches, the routers, the security, the mobility access, the design and implementation of the all of the above are critical to meeting the specific needs of the network owner and the users of the network (both internal and external).