Hi guys,

I'm asking somewhat dump question and generic one, since I'm designing new
public cloud infrastructure:

We are about to go with KVM, Advanced  zone vlan/vxlan/other isolation
method, ACS 4.4.1 or possibly revert back to 4.3. We plan on using VPC
extensively and still provide let's call it "VPS" style VMs if possible.

So:

1.  Per your experience, what is the best isolation method to be used for
Guest traffic - I'm talking here about usability of the solution,
productional one:
-- vlans - works fine, limited to theoretical maximum of 4095
-- vxlan - don't really work fine for public cloud, since default MTU of
1500 bytes is lowered on vxlan bridge/interface to be 1450 bytes so the MTU
inside VM must be also lowered...1450 bytes MTU is default/hardcoded into
iproute/cloudstack, with no option to choose larger MTU on vxlan
interface/bridge (and ask ADMIN to adjust MTU to a larger one on physical
network) - also this does not allow us to use jumbo frames, but would be a
really good thing to do.
-- GRE - I'm just evaluating/researching this


2. Another quetion - since we want to go heavily with VPC, but still want
to be able to provide let's call it "VPS" style VMs - what is the best
aproach to do so?
We already have Shared/Guest network with access to Internet - so this is
the way we acomplished single VM to be on a public IP network.
Or is it better to really dump the VPS style, and just go with normal VPC
with port forwarding to internal VM - I'm just not so clear if/how much
CloudStack was designed to support this kind of "VPS" style VMs - my
understanding is that the focus is really cloud-like/VPC functionality, and
not VPS style, at least not on Advanced zone together with VPCs - so any
advice is really welcomed.


My experience with vlans is that it works like charm, but has it's
limitations. Vxlans experience is fine if you can control MTU inside VMs -
not good for public cloud...


Again, generic questions, but I'm looking into some hints if possible and
your experience that you are wiling to share

Thanks,

-- 

Andrija Panić

Reply via email to