performance - Sharing of resources on AWS -

i interested in understanding way in hardware resources (cpu, disk, network, etc.) of aws physical server shared between different applications. people have experiences inexplicable performance changes in services running on aws have attributed application sharing physical resources? if so, how did go debugging this?

in particular, interested in more complicated interactions between resources, such cpu->memory bandwidth. if run 15 vms on single machine, surely have worse performance if ran 2 vms.

perhaps more general question xen virtualization, don't know if there kind of aws magic happening under hood don't know about.

i not sure if right forum kind of question; if not, helpful if point me towards resource or forum.

amazon ec2 instances not susceptible "noisy neighbour" problems.

based upon instance type selected, ec2 instance receives cpu, memory , (for instance types) locally attached disk storage. these resources dedicated instance , not impacted other users nor other virtual machines. (an exception t1 , t2 instance types.)

specifically:

• the instance allocated number of vcpus. these provided instance , no other instance can use these vcpus (see note t1 , t2 below). ec2 instance type page defines vcpu as:

each vcpu hyperthread of intel xeon core m4, m3, c4, c3, r3, hs1, g2, i2, , d2.

• the instance allocated amount of ram. no other instance can use ram. there no oversubscription of cpu nor ram.
• the instance might allocated locally-attached disk storage, known instance store or ephemeral storage. disk storage not persist when instance stopped or terminated, store temporary data or data replicated elsewhere.
• the instance allocated network bandwidth dedicated instance. no other instance can impact network bandwidth. network performance based upon selected instance type. basically, larger instances receive more network performance.

none of above factors impacted other instances (virtual machines) running on same host.

t1 , t2 instance types

an exception above statement are:

• t1.micro instances "provide small amount of consistent cpu resources , allow increase cpu capacity in short bursts when additional cycles available".
• t2 instances provide burst capacity based upon system of cpu credits. cpu credits earned @ constant rate depending upon instance type, , these credits can used burst cpu when necessary.

for both these instance types, assume burst capacity shared between instances, possible cpu burst might impacted other instances wishing burst. t2 instances, however, make 'fair' consuming cpu credits when cpu did burst.

dedicated instances , dedicated hosts

• dedicated instances "amazon ec2 instances run in virtual private cloud (vpc) on hardware that's dedicated single customer." basically, aws account account running instances on host computer.
• a dedicated host "physical server ec2 instance capacity dedicated use. dedicated hosts allow use existing per-socket, per-core, or per-vm software licenses, including windows server, microsoft sql server, suse, linux enterprise server, , on." basically, pay entire host computer , launch individually instances on host (at no additional charge).

the use of dedicated instance or dedicated host has no impact on resources allocated each instance. receive same resources when running normal shared instance.