Take the 2-minute tour ×
Server Fault is a question and answer site for professional system and network administrators. It's 100% free, no registration required.

We are thinking of building our own hypervisor hardware (in high volumes) for our virtualized hosting platform, instead of buying from Dell. Our model is small virtual machines 256MB of memory, and 10-15GB of disk. So, its a lot of small virtual machines (55-60) per hypervisor.

Looking for recommendations, comments, and part numbers. We want to start with a super-micro chasis, and then get a motherboard, cpu, memory, NICs, and disks.

Any recommendations on the best 1u super-micro chassis model to use? Motherboard (prefer AMD), memory, disks, NICS, power supply?

Thinking 12 cores (2) 6 core AMD cpus, 16GB of memory. Would be nice to do 32GB of memory, and fit more virtual machines per hypervisor, but I think the IO (storage local to hypervisor) is really going to be the bottleneck, so no use in putting more memory. Disks (4), 15K rpm, SAS, 500GB in RAID 10. Need (2) gigabit ports, can be embedded on the motherboard, unless there is a real benefit to splurging on Intel NICs.

Thanks.

share|improve this question

closed as not constructive by Shane Madden, John Gardeniers, Ward, Jim B, RobM Oct 28 '11 at 9:15

As it currently stands, this question is not a good fit for our Q&A format. We expect answers to be supported by facts, references, or expertise, but this question will likely solicit debate, arguments, polling, or extended discussion. If you feel that this question can be improved and possibly reopened, visit the help center for guidance.If this question can be reworded to fit the rules in the help center, please edit the question.

    
Sorry Justin but shopping recommendations are considered off-topic on this site, as they change too fast to be useful and are subjective. –  RobM Oct 28 '11 at 9:16
add comment

1 Answer

Supermicro. They have nice servers that have 2 units, handle 4 physical computers with 2 processors each. Add to that a backend with a proper SAN (can be Nxenta or somehting) to handle all the discs on a properly scled multi 1gbit network. FInished.

I would not go beelow 64gb per server (you waste money) and no local storage. Thn start putting in a proper higher end backend for the IO - SuperMicro has cases for 72 discs in 4u nits + computer, add a proper SAN software (Nexenta, coming Windows 8 server next year) with deduplication and possibly one of the new Raid controllers from Adaptec (6805) with SSD based write caching and you get an IO monster. Local IO will NOT cut it. Point. Not with good density.

Alternatively go relaly high ram and get one of theose 24 disc SUperMicrop 2U chasses. 24 discs are a lot if you put in proper not too slow discs (and again an adaptec controller with 2-4 SSD also for write caching).

share|improve this answer
    
The problem with going with centralized storage, ins't the disk speed and number of spindles, but the physical link between hypervisors and the centralized storage. A standard 1gb cat6 connection does not provide enough throughput, its the bottleneck. Either have to bond multiple cat6, or use fiber, or 10gb. All of which are cost prohibitive. Wish 10gigabit was cheaper for nics and switches. Local storage seems to be the best way to scale storage. –  Justin Oct 28 '11 at 5:30
    
No, bundle some 1gbit connections. Or try INFINIBAND - VERY cheap if you get it on the boards. And very fast. But seriously, 1gbit handles 100mbit - THIS is something you have a hard time even getting close with hypervisors. Requires a lot of fast discs. –  TomTom Oct 28 '11 at 5:37
add comment

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