1

I need advice on which of the following options would be best, plus their pros and cons. I already have an idea but I need expert opinions and experience to validate if these are even possible and if there are better similar alternatives.

But to explain the situation first:

  • I need to hook up 80 client computers. All of them are diskless, which means they will be reading from the diskless server for the OS, software, etc.
  • There is only 1 server
  • The server has SSDs on RAID and can reach up to 16Gbps read/write
  • The server will have four (4) gigabit ports
  • The clients will rarely need to transfer data among/between themselves. Mostly they are just reading from/writing to the server

Option 1: Dedicated server port for each switch Option 1

Option 2a: Main switch connects sub-switches and the server with link aggregation Option 2a

Option 2b: Same with Option 2a except sub-switches are also link-aggregated (I'm not entirely sure of this is possible so I'm just throwing this in here) Option 2b

Now for the questions:

  • Considering the situation I mentioned above, which would be the best as to performance-reliability-cost ratio (in your opinion) and why?
  • What are the pros and cons of the options above?
  • Will I need a managed switch for this? What features should I be looking for? For Option 2a and 2b I'm guessing I need a managed switch for the "main switch" and unamanaged switches would suffice for the sub-switches but confirmation from experts would be helpful.

Thank you! :)

EDIT: Updated information

|improve this question

closed as primarily opinion-based by Ron Maupin♦ Oct 13 '18 at 1:47

Many good questions generate some degree of opinion based on expert experience, but answers to this question will tend to be almost entirely based on opinions, rather than facts, references, or specific expertise. If this question can be reworded to fit the rules in the help center, please edit the question.

  • Asking for the "best" is really soliciting primarily opinion-based answers, which is off-topic here. – Ron Maupin♦ Oct 13 '18 at 1:47
  • 1
    Just commenting I had to read it twice to understand "diskless server" meant "server with the disks" .. you might consider giving it a different name! More to the point: Zac's answer gives very good questions about priorities. I'd suggest you focus on those first, which will then drive the answer, rather than do topology first. Otherwise you'll risk the very real problem of implicitly changing priorities when criticising any given solution. – jonathanjo Oct 13 '18 at 10:26
1

Unless the server's got significant storage bandwidth it'll be your bottleneck, no matter how you connect the network.

It's hard to choose the "best" solution when you're not giving us your priorities: performance? cost? reliability? scalability? manageability? security?

LAG trunks require managed switches, so for option 2a that's the main switch and for 2b it's all switches. Managed switches are more expensive but they can increase your scalability and control significantly as well.

Option 1 is the cheapest but has the problem of having to use separate client subnets, requiring them to use the router for inter-subnet traffic.

|improve this answer
  • Updated my post. The server has SSDs on RAID 0 reaching 16Gbps read/write. Wait, so for Option 2b all switches need to be managed switches? D: – Krauxe Oct 12 '18 at 22:33
  • 1
    RAID 0 has zero redundancy for media defects - generally not suited for production. Yes, LAG trunks require configuration, requiring managed switches. – Zac67 Oct 13 '18 at 7:07
  • @Krauxe Are you building from Scratch? What's your budget? Are you thinking Nexus Switches or IOS Switches? Or non-csico switches? If It's just 80 Thin-Terms you can probably build out fairly on the cheap for an office setup, but want ot know how it integrates with the existing switch infrastructure. Is the SAN/NAS "Diskless" system only to be utilized by the 80 Thin-terms? – Ben Personick Oct 13 '18 at 17:28
  • @Krauxe Lets assume "Main Switch" means you have an old Network to integrate with, perhaps where the Thin-terms need to communicate to, and that it is at least in and of itself redundant and able to handle a simple add of new switching. That this old network may have existing vlans and IP addresses paces, but for our purposes we can assume it only had a flat network in a single vlan (if you never defied the vlan explicitly it's still there it's just the default, which if not set, is 1) – Ben Personick Oct 13 '18 at 17:30
  • @Krauxe With those assumption in mind, and the assumption that the core switches are decent, and that most of the need to access the "diskless" system will be from the thin-terms in the same segment, I would purchase 3-4 3750x 48-port Switches (alternately 3-4 and 2960X 48 port switches) along with stack modules and cables. Set up a switch stack for these switches, and plug the SAN/NAS unit into all of the switches, with a Port channel (Link Aggregation) across those ports so we can utilize the entire 4Gbit/s bandwidth, and no single switch failure can take down the connection totally.) – Ben Personick Oct 13 '18 at 17:35

Not the answer you're looking for? Browse other questions tagged or ask your own question.