We have a 200 person firm with 3 offices and we are about to go virtual with ESX VMWare. What are your hardware suggestions?
Server Fault is a question and answer site for system and network administrators. It only takes a minute to sign up.Sign up to join this community
Servers: HP Proliant DL3x0 G6 or Dell PowerEdge 11th gen. Three reasons: Nehalems are built for virtual loads, they support tons of memory (up to 144G in the higher-end systems), and these servers can be configured to have a multitude of NICs (properly designed network layer requires 6 or more network connections per server). Size them for your needs.
Network: good managed GigE switches in redundant pairs; 10GigE if you need fast iSCSI or FCoE storage or want Fault Tolerance. FC switches if you prefer FC-based storage fabric.
Storage: NFS-based NAS like NetApp or iSCSI or FC SAN (HP MS2000 or Lefthand, Dell MD3000 or Equalogic, EMC Clariion - pick your poison based on the budget and features required).
Building virtual infrastructure without shared storage is a waste of time.
The answer is kind of the same for most hardware questions.
As much RAM and CPU cores as you can afford. (we're running 48gb to 100gb+ RAM per ESX box).
Also, stump up for extra NICs.
As for storage, it depends on how many hosts you're going to be running, how large they'll be, and how intensive. Local storage on the server can be fine, but I'd suggest looking at something like a NetApp device if your budget can stretch to it.
Pick good CPU's add lots of RAM and lots of NIC's - a VMware Cluster ideally has 4 for kernel\console functions, plus as many physical NICs as you need to provide the bandwidth and redundancy your VM's will need on top of that. Pick a nice modern 64bit CPU that support 2nd Gen hardware virtualization (ie support for EPT\NPT). The Xeon 5500's in the HP G6 servers\Dell 11G range are pretty much the best at the moment but the 6 core AMD Istanbul Opterons are not too shabby either -lots of cores helps significantly with scheduling multi-vCPU VM's. And spend wisely on shared storage (NAS\SAN) to get the most out of your virtualized environment.
Spend some time analyzing the workload that your servers need - specifically peak\average CPU, Memory and IO (both bandwidth and IOPS) and size the systems so that you can run everything without contention with at least one VM host down. That way your cluster can handle failures and you can keep everything maintained with minimal downtime.
There are plenty of folks who will carry out "Virtualization Readiness Assessments" for you that will give you this in a nice shiny report but if you're a reasonably competent systems admin you may be able to gather the data yourself or if you want a very good analysis of your environment Novell's PlateSpin Recon is very good but it's licensing costs $2 per server per day for the duration of your assessment, and you want to run the assessment for a nice long period 2 weeks absolute minimum, ideally longer than a month and covering all major peak load periods (e.g. month ends\full backups etc).