Questions regarding a new build and Hyperthreading Questions

[DarkZenith]

I have several questions about how to setup my unRAID, I did some searching but didn't find everything I wanted to know. The parts enroute to setup my unRAID box for my Fileserver and my Kids gaming machinesare as follows.

 

Dual Intel Xeon E5-2670 8Cores 16Threads (HT) 2.6Ghz 3.3GhzTurbo

Dual Corsair H60 Rev2 Closed loop Water Coolers

Supermicro X9DRI-F Motherboard - C602 Chipset, IPMI

(16x8GB) 128GB DDR3 10800 ECC memory (Delivery was lost so it is still enroute/being replaced)

x3 Nvidia Cards (GTX 580, GTX 760, 1 New one likely to be a 1060 or 1070)

x2 SSDs (Likely a pair of Samsung 850 EVO's, 500GB)

x5 Western Digital Red 5TB NAS Drives (2 I already have, 3 are new orders)

 

The plan I have in mind is to have the unRAID run off the IPMI internal/remote system access. It will be setup as x3 VM's for my 3 kids as replacement gaming rigs for them. Also running my fileserver off this machine as well.

 

Planned config is either 4 or 5 Base Cores per kids machine, The remaining base core and all hyperthreaded cores will be setup and reserved for the fileserver/unRAID use.

The SSD's are going to be the cache drives for a storage array setup over the WD Reds. 24GB Ram per VM with the remaining 32 reserved for drive caching/unRAID use.

 

Questions I have are -

 

Regarding CPU and HyperThreading configuration

 

Has Anyone tried setting up the vm's with only base cores and assigning all Hyperthreaded cores for filesystem use? How well does this function?

 

Has anyone tried swapping Virtual Cores off between 2 VM's? I.E. VM 1 has Base cores 1-4 and virtual cores 5-8 meanwhile VM 2 has Base Cores 5-8 and Virtual Cores 1-4? I am wondering if this will essentially allow the systems to cross balance processing power. The thinking is that if VM1 isn't in use then 100% of the processing power for that core is available to the Hyperthreaded cores for use on VM2 and vice versa. when the systems are both under load then it'll balance out still. With x3 systems then run x2 virtual cores from each of the other 2 systems to balance it over all three.

 

Would it be better to just setup the system to assign x4 cores + the matching virtual cores to each system leaving the last 4 cores + virtual cores in reserve for the system?

 

With a dual cpu configuration would it be better to assign x2 cores from each processor to each vm to balance the load?

 

Lastly,

 

If I understand the processor architecture properly, each processor has x8 cores, with it being internally pairs of cores, each having 2 memory channels. Is there a way to determine and assign physical blocks of memory to each machine based on the processors that it is attached to? For example, Processor cores 2 and 3 have the pairing of memory channels 2 and 3 bound to the internal memory controller, in turn have the system assign memory range (with x8GB memory modules) 16.0GB to 31.99GB, and then for cores 4/5 use 32GB to 47.99GB range. Would there be any benefit to this? Is there any real perfomance loss if memory that is physically attached to the second processor is accessed by the first? Or is the internal bus fast enough to make this negligible?

 

For the File Storage system I am planning to run the x5 WD Red drives in a redundant raid setup through unRAID. Use the x2 SSD's for cachine access to the array as well as leaving 32-64GB of memory available another level of caching. Is this a good setup/plan to maximize the performance of the system? Can the memory be setup as a full blown write cache to queue writes as lower priority than reads to really maximize the throughput between the machines?

[heratic]

i just disabled hyper-threading. That way the VM gets the whole core. Not sure if theirs any advantage to using hyper-threading with VM's. If you choose to use HT then passing the whole cores to the VM will give the best results

[all0gIc]

Hi 

 

Did you have any results from your considerations? 

 

Cheers