Date: Mon, 15 Nov 2010 21:13:51 -0600 Subject: Re: OT: VM slicing and dicing From: mysidia@gmail.com To: brandon.kim@brandontek.com CC: nanog@nanog.org
On Tue, Nov 9, 2010 at 10:17 AM, Brandon Kim <brandon.kim@brandontek.com> wrote:
I'm not looking for companies that offer this service, but the actual software engines that allow you to create VM's on the fly. So a customer goes to your website and says I want Win2008 with 8gigs of RAM and 120gigs of HDD. Just like custom configuring a new PC.
How about I send you some terms to search for, using your favorite search engine... Multi-Tenant Hosting > Cloud Computing > IaaS / HaaS (Infrastructure as a Service) > Self-Service Provisioning Because the question is so vague, I think you need more research. If you read the documentation of portal software, you should be able to tell to what extent it would be "turn key"
Before looking too closely at any offering... some things to think about are.. How would you go about handling virtual networks and access to them? Will you want one shared network (with requisite Layer 2 security minefield), or will your portal of choice somehow decide to permission and make certain LANs available to certain users' VMs?
There will be security and performance considerations that some portal software programs allow you to answer, and some do not. So you need to decide the hard requirements for security, management flexibility, UI attractiveness/ease of use, functionality for the end user, resource management, and price :)
Different portals have different options, so define requirements first. A Multi-Tenant IaaS environment (meaning different users sharing pieces of metal, storage, etc) brings in some complexity.
Think about how will the resources be balanced? E.g. Will you have a
place workloads on its own, or rely on some outside system like vmware DRS. Will the portal implement and enforce resource SLAs for Network latency/loss, limit the number of VMs per NIC or per datastore, Memory, CPU and provide I/O response delay assurances, or will machines be left underutilized / overutilized, because the portal is bad at optimizing placement on
servers, or bad at avoiding overcommit?
For an IaaS provider, underutilization eventually means you are eating more kW*h than necessary, and overutilization could be immediately detrimental.
The different major virtualization software vendors each have their own Self-Service Provisioning solutions, and there are some third party
Most are for Enterprise internal self-provisioning; Hosting providers might have special requirements like "integrated user signups and billing" and "no license restriction against provisioning for outside users". I would expect these to be more expensive, or include monthly
1 GiGE switches at a minimum; some vendors (e.g., arista) have low cost 48 port 1000/10000 switches. Cisco's UCS system uses 8 10 GiGE uplinks where the servers (running a hypervisor kernel) plug into a chassis backplane with 2 10 GiGE connectors each, that mux 10 GiGE and 4/8/16 GiG FC over the combined 80 Gig uplinks. Think about latency, not just bandwidth. 100 Mb is 100 times slower in serialization/deserialization of bits on/off the wire. Also, do you really want the cable management issues associated with multiples of 48 copper cables from servers to top-of-rack switches? -----Original Message----- From: Brandon Kim [mailto:brandon.kim@brandontek.com] Sent: Tuesday, November 16, 2010 5:04 AM To: mysidia@gmail.com Cc: nanog group Subject: RE: OT: VM slicing and dicing Thanks for the suggestions James! One of the issues I had, (which is why I turned to NANOG) was that I wasn't entirely sure what keywords to search for!! So thank you for that. All of the criteria's you brought up are valid and I will add them to the list of things to consider. It's awfully difficult to figure out who can do what as it's just not possible to test all the different vendors out there unless you have a large R&D team and a lot of time. I think we are on the same page as far as what "We" think I need. But just to clarify. 1) We'd like to be able to have a web portal where new or existing clients could request servers of all types: windows, linux etc... Configure what it is that they need and in some amount of time, the VM's are provisioned. They receive some kind of email confirming that their new provisioned server is available. 2) Backend - Since we haven't invested much time into the backend, we're open to all possibilities. It doesn't need to be VMware at all. Xen seems to be extremely popular. 3) Licensing - Of course this will be all unique to each vendor but the more complicated the licensing, the more it's a turn off and difficult to keep track of. Not to "plug". But so far OnApp's pricing is very straightforward. 4) Multi-Tenant - Absolutely needs to support this. I don't expect anyone here to do research for me, but I assume that being a network operator, many of us would have some input and clearly I've received great feedback. I've been in touch with numerous vendors that were given to me from this thread and I can't wait to demo/try their products.... One question I do have for any that actually read through this entire email (haha) is about the physical network switch. Is there a case for the switch, especially in today's high density environment to go with 1GIG switches as the minimum? It seems pretty obvious but I'm wondering if it's really a necessity? Can anyone on this list argue that 10/100 will be suffice? Thanks again! Brandon portal physical programs. per-user fees.
Offhand I recall Virtuozzo [perhaps the oldest?], Enomaly / Enomalism, enStratus, MS Dynamic Datacenter Kits which are a framework, VMware vCloud Express through the VSPP, Citrix XCP, Eucalyptus, as interesting by no means exhaustive.
-- -JH