How many Servers

[Harpalus]

Hey all,

ANybody know of good reference material to figure out the number of gentoo/linux servers you need and the bandwidth for file servers per user.

Thanks

[masseya]

Wow, there's a lot of variables involved in a calculation like that. Can you narrow that field down a litte? What kind of application is this network for?
_________________
if i never try anything, i never learn anything..
if i never take a risk, i stay where i am..

[Harpalus]

This is simply for file servers.
Access would be through ssh(nfs) or samba and the data would be normal office type documents.
I am wondering how many servers you need on a per megabyte of file served basis. A number of factors go into this. Each physical server has a bandwith limit and a cpu limit. Additionally you have to look at diskspace limits. Maybe usign NAS or NAs would be helpful. I should mention that this will be in a WAN environment.

[elzbal]

Can you be more specific as to your requirements? What bandwidth (MB/sec) is required per user, or simultaneously? How much data is to be stored? You mention a bandwidth limit per server... is this above and beyond the 100MBit/s (~5-10MB/s in real life) cap imposed by most NIC cards? Do you have any uptime requirements that would necessitate redundancy?

Simple file/print serving is not a processor-intensive task. For example, a low-end Pentium system could saturate a T1 line. What are your requirements beyond this?

[masseya]

There are so many ways that you could set this up that it's hard to know where to begin. elzbal is right, we need a lot of information.

1. How many users will there be?
   
2. What would their average network usage look like?
   
3. More importantly, what would their peak network usage look like?
   
4. How much are you willing to do to ensure reliability? (i.e. redundancy..)

Much of this elzbal mentioned, I just thought I would put it in numbered question format.
_________________
if i never try anything, i never learn anything..
if i never take a risk, i stay where i am..

[Harpalus]

Hey all,

Thanks for the answers, the main problem being that I don't have the details of all that information. I am trying to ballpark figures here.
Imagine your standard corporate ofice worker editing some word documents, maybe a few power point presentations, downloading and storing some pdf documents that kind of activity.

In terms of bandwidth limit I was referring to NIC bandwidth limits HOw many people of that description could I realistically serve with a single server without a large performance hit as compared to local hard disk work.

I wish I had more details to go on, but I am just trying to get an idea and was actually hoping to get pointers to methodologies that can be applied to evaluate these kind of things.

Reliability would be crucial. ON a side note how much of performace hit would you take using samba over say NFS and how much of an impact would IPsec have?

[masseya]

If reliability is a serious sticking point of yours then you'll want to have at least two servers. You'll want them located on different power grids so that the whole system isn't down in case of a failure. If all you are doing is some simple file serving, then you can serve a lot of users, easily 50+, with one system even if the users are pretty hard on disk space. A single file server and a backup are pretty common setups for small businesses. Tossing around numbers is really hard though without general usage requirements.
_________________
if i never try anything, i never learn anything..
if i never take a risk, i stay where i am..

[elzbal]

I would agree with massaya. For a typical small-to-mid office environment running in a standard 100MBit environment, I would use a single server for general usage, along with a single backup that could be swapped into place in case of failure. Failure would occur rarely in a stable production environment (between 0 and 2 times a year), and I'm sure the users would be understanding while you take 15 minutes to swap another (already replicated) server in place on rare occation.

You will probably want to replicate frequently, maybe as much as on an hourly basis. I'm not sure what mechanism you would want to use, but rsync comes to mind. I might consider getting an extra NIC for each box, connecting the 2 boxes via a crossover cable or other back-end mechanism, and modifying the routing tables so the backups run on this wire. This way, the frequent backups will not compete for bandwidth with your clients.

I doubt there is much need for multiple servers for bandwidth-to-client reasons. Your clients will probably have bandwidth limitations other than the 100MBit imposed by the NIC, for example, if they are on a WAN (I doubt you'd have more than 100MBit available to remote clients) or if they are on a hub, as opposed to a switch, at any point (which would remove any benefit of using multiple NICs... it would throttle you back to 100MBit). Besides, it probably wouldn't be that noticable anyway.

As for hardware, I would get something new... you could do all this on yesterday's Celeron, but older hardware would not be as reliable. Yesterday's Celeron is just that much closer to the end of its life (and warranty). Remember, however, that you don't need the fastest server on the block, either.

I would include a healthy chunk of memory, maybe 1GB or more, because the OS will automatically cache data from the disk into RAM, and a healthy cache will help speed performance in certain circumstances.

Of course, you should be able to maximize the amount of disk space on this server. That means getting a case with plenty of expandability. If you really need to use massive numbers of disks, you should consider SCSI (which is generally faster anyway) - an IDE bus can use 2 disks (including CDROMs and other similar devices), whereas a SCSI bus can use 15. I might consider putting the OS on a modest IDE drive (as well as using an IDE CDROM), and getting a few large SCSI disks for file sharing. However, you may have to fall back to IDE if SCSI is too cost-prohibaitve.

I hope this helps give you more things to consider...