Overview

Research Computing provides virtual resources through its OpenStack virtualization platform. This virtualization platform hosts a shared pool of servers where virtual machines are instantiated. The virtual machines are where researchers install their applications. Research Computing grants a free slice of these resources to all KAUST PI; this slice amounts to 32 vCPU (Virtual CPU), 128 GB of RAM and 1 TB of storage capacity. Research groups might need more resources than those allocated for free. According to our virtualization platform standard, that’s when a PI is asked to help grow the virtualization platform by leasing virtual resources granting the interested research group a larger slice of the shared pool. This document explains how a PI can lease these resources, what it means to lease these resources, lease costs, and what are the expectations for this lease.

What workloads run on OpenStack?

It really depends from the faculty group who manages what’s installed on their virtual machines. The OpenStack hypervisors are connected through 2x10G ports which means you cannot run high throughput jobs, e.g. MPI ones between them nor do these nodes have a shared fast performance storage. These nodes usually run Hadoop, Elastic Search, MySQL, Apache, PERL, Python, Machine Learning, etc.

Which resources are being leased?

Three different resources are being leased:

  • CPU
    • The underlying physical CPU is split among multiple groups, e.g. 32 physical CPU become 256 vCPU using a 4:1 ratio. This is referred to as oversubscribing. Our resource allocation model and underlying technology makes sure there is a minimum performance penalty associated with oversubscribing. The cost of one vCPU depends on underlying physical CPU model. In the current cycle we deploy CPUs from Intel Skylake Gold family and this sets the price for one vCPU to $157.
  • RAM
    • Memory is accessed in a 1:1 ratio. We don’t overcommit memory because this would cause a noticeable performance hit. Its cost is $14 for 1GB.
  • GPU
    • The underlying GPU is only accessible to one virtual machine meaning that’s it’s exclusively accessible to the group that paid for it. Its cost matches the cost of physical GPU card. At the moment, we only support NVIDIA P40 GPU cards.

How much does the lease cost?

First of all, you have to decide how many virtual resources you require from those available. Once you have decided the size of your allocation then the math is simple; we just multiply each virtual resource by its price tag.

Here’s an example. Let’s say you need 200 vCPU, 1TB of RAM and 2 NVIDIA P40 GPU. Your lease cost would be the following:

  • 200 vCPU * $157 = $31,400
  • 1024 (aka 1TB) RAM * $14 = $14,336
  • 2 P40 GPU = 2 * $5000 = $10,000 where $5000 is hypothetical cost of one NVIDIA P40 GPU card

How long does the lease last?

We lease all our virtual resources for three years. During these three years you’re guaranteed to have these leased resources available. In the following years we’ll refresh the underlying hardware using the money that has provided by all leases. You’ll have to renew your lease to access these new virtual resources.

What happens when the lease expires?

Not much really. You’ll continue to use the virtual resources you leased; we won’t kick you out. What will happen is the underlying hardware will fail or will be decommissioned after an additional 4-year grace period, whichever happens first and will not be replaced because it’s warranty/support lasts three years like your lease. This means that over time your capacity shrinks because the old hardware has been removed due to irreparable failures or because after seven years it has to be replaced.

You convinced me, how do I pay?

You provide use with your fund from where we have to transfer money to our fund for virtual resources. We’ll send you an email detailing what resources you requested asking for your permission to initiate the transfer. Once you acknowledge with a yes, our email, you’ll be able to access your virtual resources from the OpenStack administration pages; in the meantime, our IT procurement team will initiate the funds transfer.