1

This is a difficult question to ask, since it depends on a number of assumptions. But let's assume we have 100 workstations that need to connect to a network attached storage device. What is the maximum throughput a NAS could deliver to those workstations, if each has a 1Gb/sec connection? I'm attempting to size a NAS for an installation with 100 workstations and some scientific equipment (say, equivalent to 20 more workstations). Vendors have said, "you tell me what you need." I need to connect 100 workstations to some system that will serve home directories to workstations, a few servers that will run virtual machines for such applications as license and web servers--but let's focus on what a NAS device could possibly deliver, given 1 Gb/sec connections. If the NAS is providing 20gb/sec, then I suppose at most 20 workstations at any time are going to transferring files at the maximum rate.

Related questions are what are typical file I/O requirements of a workstation, say in a scientific computing environment?

Improve this question
2
  • 5
    You probably need a SAN, or at least a proper NAS server for that workload, but either way, there's no way to give a proper answer without much, much more information... that I doubt you'll be able to provide. You should really get some vendors in to assess your environment and provide a product that will fit your needs.
    – HopelessN00b
    Jun 16, 2014 at 11:16
  • How long is a piece of string? Workstations typically pull nearly zero - 1gigabit down, 10g up and you are fine. Calculation computing? Dependw aht this means ;)
    – TomTom
    Jun 16, 2014 at 11:20

2 Answers 2

Reset to default
1

You need to some data collection. This can take basically take 4 forms that I can think of. A combination of them might be best, but individually I'm listing them from what I consider to be worst to best.

  1. Ask the vendor (or a set of vendors) and base it on what they tell you. They should all have calculators for guessing this sort of load. They might even people able to loan you a SAN to try it out (but you will likely have to spending significant capital with them for this to happen). This is similar to asking here on Server Fault, but vendors should have more significant sample sets than what you get back from a couple of our users.
  2. Back of the envelope: Figure out what sort of tasks people do in your environment with files in their home directory. Do they load big datasets, spreadsheets, actually load programs stored in their home directory? Based on that, come up with hypothetical profile for a typical computer (the amount of data they are loading, assume it is bursty, and the multiply that by the number of computers, and spread over the work day using something like a poisson distribution).
  3. Gather some sample data from a few peoples workstations throughout the day. This depends on the operating system, but you can use tools like sar in Linux environments and Perfmon in windows environments. Then take this profile and extrapolate using a similar method as above.
  4. Gather the data from all workstations. This of course, would be the best.
Improve this answer
4
  • You could also assume their entire workstation is hosted on the SAN, and pretend you are building a Virtual Desktop Infrastructure (VDI) and use some estimators like blogs.citrix.com/2010/10/31/… .
    – Kyle Brandt
    Jun 16, 2014 at 12:13
  • THe problem with this is that this may end up factors too high.
    – TomTom
    Jun 16, 2014 at 12:29
  • The VDI reference is very helpful--very close to what is needed. I'm assuming that all my connections are maxed out: all 100 workstations need 1gb/sec, a 20gb/sec link to another facility is fully utilized. This is in the absence of numbers. The workstations don't exist yet, so I am assuming worst case. I'll vote this as the acceptable answer.
    – Christian Marks
    Jun 16, 2014 at 22:00
  • Ya, just keep what TomTom said in mind, VDI is likely highballing it. But when you lack information, and you have the budget, better too much than too less.
    – Kyle Brandt
    Jun 16, 2014 at 22:16
3

Storage performance isn't always about throughput...

So to reality...

Today, I would likely build a NAS solution for the client set you have with dual bonded 10GbE connections to a pair of cross-stacked switches (or a chassis switch). It's not like you have that many options...

NAS --- 2 x 10GbE ---> switch --- 1GbE ---> computing workstations

This is based on the following factors:

  • What's generally available in the NAS landscape (you won't normally see anything greater than 1 x 10GbE in commercial solutions)
  • What's affordable for most organizations (10GbE ports are common among current switch offerings today)
  • There's going to be a certain amount of oversubscription. (need to size for the worst-case conditions)
  • You're not running 10GbE to the workstations.

At the NAS level, you should be concerned with:

  • Size of typical data working set.
  • Storage IOPS needed on the backend.
  • Capacity and room for growth/expansion.
  • Tiering/caching.
  • Resiliency (RAID, etc.)
  • Backups.
Improve this answer

Not the answer you're looking for? Browse other questions tagged .