Private Cloud Tender - Part One

In this article, I want to share my experience / methodology for implementing Private Cloud. Here I will give the questions that we asked during the implementation and how they were resolved.

.

We participated in a tender to develop a Private Cloud service for one large European organization. The purpose of this tender is to offer an infrastructure that provides IaaS and PaaS services to the client. Below are the key points of those. tasks:
  • major operating systems are x86-based Windows and Linux
  • “Standard” list of software for PaaS services (web, database, application server, etc.), including Oracle databases
  • it is also advisable to offer Solaris x86
  • infrastructure should be located in two data centers (distance <50 km)
  • 6 types of virtual machines (from 1 vcpu to 10 vcpu)
  • 4 services (SLA): from bronze (best effort) to platinum (RPO: realtime, RTO: 2h)
  • firm order for 1000 virtual machines, with a possible growth of up to several thousand


In Private Cloud, the hardware does not play a key role: servers, storage and more can be purchased from anyone, there are plenty of players on the market. All the highlight is in the software part, which will manage / conduct everything and everything. This is harder.

After a detailed study of those. assignments, we identified two areas: studying the hardware offer and the Private Cloud software offerings. For this, meetings were organized with manufacturers of hardware and software.

Iron

We had the following choice:
  1. to assemble by ourselves from those elements that we are used to using: HP servers, EMC and NetApp storage systems and a network from Cisco.
  2. follow the flexpod architecture
  3. use VCE Vblock (hardware and software solution)


NB: We did not consider the integrated solutions of IBM or HP, since they were our competitors in the tender.

The first option was crossed out quickly enough, because the client wanted a solution in the form of a block and the most integrated one. In addition, it was strategically important to offer the client a solution that does not tie his hands and feet to our organization. Relying on ready-made solutions, we got rid of the problem with the support and development of architecture.

In short, the key points of FlexPod and Vblock

FlexPod is an infrastructure design model (blueprint) for Cloud, consisting of Cisco UCS (compute & network) and NetApp (storage). The following image shows the standard FlexPod FCoE and 10 GbE architecture.



You will have to assemble and configure the entire infrastructure yourself, but for this there is documentation available on Cisco / NetApp sites, such as this one . The architecture that received the FlexPod certificate entitles you to a unified support service, regardless of the source of the problem: compute, network, storage. True, for one part of our employees, a single support service is absolutely not an argument.

The flexibility of the FlexPod architecture is that you can choose any suitable NetApp storage, the required number of chassis and Cisco UCS servers. You can add an internal SAN if, for example, the FCoE protocol does not satisfy you or if consolidation of SAN & LAN on one device (Cisco Nexus 5K) also does not suit you. As a virtualization, there is a choice of VMware, Hyper-V, Citrix.

Vblock is a finished product, very similar in appearance to FlexPod: Cisco UCS (compute & network) and EMC (storage), as well as including the VMware hypervisor and management tools in addition to the hardware. Unlike the standard FlexPod, Vblock always uses a separate Cisco MDS SAN switch to access storage.



All configuration and assembly is implemented by VCE. There are several Vblock series: 100, 300 and 700 (from the initial to the most productive).

image

Each series in turn consists of a number of models. Unlike FlexPod, VCE is the legal entity responsible for Vblock. In this case, the support looks more serious than for FlexPod. All architecture changes that you intend to implement must be confirmed by the VCE consortium, otherwise the Vblock certificate is lost.

NB: If we replace the NetApp storage system with EMC in the FlexPod hardware, we will almost get Vblock. What is Cisco UCS can be read here .

After meeting with Cisco / NetApp and VCE, we settled on the FlexPod solution for the following reasons:
  • FlexPod supports multiple hypervisors, unlike Vblock
  • FlexPod has more flexibility when building a hardware complex. For example, if you already have NetApp storage, adding Cisco UCS and following the Cisco / NetApp instructions will give you the FlexPod infrastructure. In the case of Vblock, this is not possible, you have to buy the entire "ammunition."
  • FlexPod infrastructure changes are not subject to such strict rules as in the case of Vblock. For example, blade servers are always added only in pairs for Vblock.
  • in order to provide a high level of service with critical RTO / RPO, we need a solution for synchronous data replication. In the case of NetApp storage, we get the built-in MetroCluster solution , which is included in all FAS 3xxx and 6xxx models. For Vblock, this is implemented through RecoverPoint or Vplex, but it costs a lot of money (SRDF / S does not count).
  • NetApp is a unified storage system, as well as a backup system (back-up), so there is no need to purchase an additional solution such as EMC Avamar.


Vblock undoubtedly has its own arguments, but in this case, the above elements were more important for us. Those who are interested will find a more detailed comparison at this link .

In the second part , I will talk about the design of FlexPod, and then about the software component of Private Cloud.

Also popular now: