Where you decide to run your applications is as important as what you run. What does your workload placement strategy look like? Home architects are very careful about their design choices. Many of their decisions, such as the best locations for load-bearing walls, support beams and other infrastructure, have long-term consequences.
Where do they put windows and skylights will deliver optimal sunlight? How do they situate bedrooms and bathrooms? What is the right density of wood, concrete and other materials required to construct safe walls, roofs and floors? Those are just the broad strokes; architects plan thousands of minor details as well, often well before raw materials are purchased.
Like their home-building counterparts, IT systems architects carefully design technology systems. Which is why workload placement has emerged as a critical strategy for governing what applications and other resources run where.
IT has grown more complex, thanks to a proliferation of environments comprised of public and private clouds, on-premises infrastructure and edge devices. IT leaders who placed assets in these locations have constructed a multicloud house without planning for the long-term impact on their organization.
For example, while it may have initially made sense to build a key business application in a certain IT environment, perhaps performance began to lag as usage grew. Maybe the goalposts for security and compliance shifted, forcing you to rethink your choice.
Whatever architectural concerns arise, where you decide to put what in your IT house can be as important as how architects design physical homes. CIOs are thinking about this a great deal, as 92% of 233 IT decision makers Dell surveyed said that they have a formal strategy for deciding where to place workloads. Half of those executed this strategy in the past year.
The Great Overcorrection
The public cloud grew rapidly, as engineers learned how easily it enabled them to launch and test new applications. Soon IT teams notched quick wins, including flexibility as they lifted and shifted existing business applications to the cloud.
Then came the overcorrection. Emboldened by the prospect of saving money while fostering greater agility as they innovated, many CIOs declared a “cloud-first” strategy. Those who were initially more measured in their adoption of cloud technology saw their colleagues migrate their entire IT estate and followed suit.
As workloads got more complex it turned out that the public cloud-first stance was not always the best fit for the business. Hasty decisions had unanticipated ramifications, either in the form of escalating costs or failed migrations.
The reasons: Workloads are unique. Each application has its own set of business requirements and benefits. Just as the home architect must carefully weigh each design choice, CIOs must be intentional about where they put their software assets.
Variations on a multicloud
Let’s consider some examples where the right workload is tied to a business outcome. Cloud environments—public or private—make sense where you get huge bursts of data traffic. Cloud technologies enable you to quickly spin up compute resources and dial them down as requirements subside.
Retail ecommerce is a classic example. For brands selling clothing, footwear and other merchandise, holiday seasonality drives peaks and valleys to web and mobile sales. Large traffic spikes in October or November through Christmas subside, then stabilize.
Or think of a digital crossword puzzle published every weekend. With most people completing these on the weekend, traffic bursts Saturday and Sunday before slowing over the remaining 5 days.
For such use cases, a public cloud that provides massive scalability may yield the desired business outcome.
Conversely, so-called “steady state” use cases—in which applications’ compute needs fluctuate little if at all—often run better on-premises, either in traditional IT infrastructure or in a private cloud. Thousands of these applications run without much deviation across business lines.
Think traditional general ledger software in ERPs. Travel and expense utilities. Software that governs data backups. Applications, such as those that monitor anomalous network traffic, often run locally for security reasons.
Other applications with disparate patterns and needs are emerging. Applications requiring minimal latency—think Internet of Things software—are moving to the edge for faster processing and cost efficacy.
In Dell’s survey, 72% of IT decision makers said performance guided their decisions to place workload, followed by data protection and security at 63% and 58%, respectively. Venues include public clouds, data centers, colocation facilities and edge environments.
Workload types vary, but 39% of respondents said they had placed data protection workloads while 35% each said they had placed ERP and CRM systems.
Diverse workloads require fungible infrastructure
There are no absolutes in determining workload placement. Well, not in the way many IT leaders think. Every software asset will have different requirements, which will influence where you decide to place them.
Just as an architect decides how to situate walls, beams, rooms and other physical infrastructure, where an IT architect places assets matters. The wrong choices can have negative consequences.
These decisions aren’t easy nor should they be made lightly, as the ramifications of poor asset placement can impact your bottom line, make your business more vulnerable or prompt you to run afoul of compliance mandates.
All diverse workloads require a flexible infrastructure that enables enterprises to move their applications and other workloads to move seamlessly across clouds, on-premises and edge venues, based on their business requirements.
As-a-Service infrastructure, which includes on-premises equipment ordered on demand, can power these workloads to meet requirements for performance and availability, as well as your needs for simplicity, agility, and control. How will you lay the foundation for your IT assets?
Keep reading: Why the Hybrid Workplace Needs a Hybrid IT Model
Learn more about our portfolio of cloud experiences delivering simplicity, agility, and control as-a-service: Dell Technologies APEX.
On the cooling-aware workload placement problemP Cremonesi, A Sansottera, S Gualandi – Workshops at the Twenty-Fifth …, 2011 – aaai.org… -AWARE WORKLOAD PLACEMENT PROBLEM, that looks for a workload placement that … a cross-interference matrix that links the workload placement to the cold air temperature. …Save Cite Cited by 7 Related articles All 5 versions
Online placement of multi-component applications in edge computing environmentsS Wang, M Zafer, KK Leung – IEEE Access, 2017 explore.ieee.org… This problem, known as the application or workload placement problem, is notoriously hard,for tree application graph placement. We jointly consider node and link assignment, and …
Virtual machine placement with two-path traffic routing for reduced congestion in data center networksR Kanagavelu, BS Lee, NTD Le, LN Mingjie… – Computer …, 2014 – Elsevier The core network oversubscription and unbalanced workload placement could lead to long-…we consider efficient placement of VMs so as to minimize network link congestion and we …
On the placement of web server replicasL Qiu, VN Padmanabhan… – … IEEE INFOCOM 2001 …, 2001 – ieeexplore.ieee.org replica placement in detail. We develop several placement algorithms that use workloadinfor… To study the effect of overlooking some network links on the placement algorithms, we …
Enabling efficient placement of virtual infrastructures in the cloudI Giurgiu, C Castillo, A Tantawi, M Steinder – ACM/IFIP/USENIX …, 2012 – Springer… placement stage has on the performance of our approach. Second, we consider a more realistic workload … apply our placement technique on generic VNIs whose VMs and links require …
Practical service placement approach for microservices architectureM Selimi, L Cerdà-Alabern… – 2017 17th IEEE/ACM …, 2017 – ieeexplore.ieee.org… For bidirectional links, we count both links in opposite direction as a single link. In summary, methods for cloud workload placement. Regarding the service placement through migration, …
Virtualknotter: Online virtual machine shuffling for congestion resolving in virtualized datacenterS Zou, X Wen, K Chen, S Huang, Y Chen, Y Liu, Y Xia… – Computer …, 2014 – Elsevier… by core network oversubscription and unbalanced workload placement. In contrast to traditional… In fact, with further inspection into the link utilization, we find the links among those racks
Achieving predictable performance through better memory controller placement in many-core CMPsD Abts, ND Enright Jerger, J Kim, D Gibson… – ACM SIGARCH …, 2009 – dl.acm.org… Lastly, we have a detailed full system simulator that allows real workloads to be applied to …placement has 33% less link contention compared to the baseline row0_7 placement used by …
Optimizing the placement of internet taps in wireless neighborhood networksR Chandra, L Qiu, K Jain… – Proceedings of the 12th …, 2004 – ieeexplore.ieee.org… In this paper, we explore the placement problem under three wireless link … link characteristics.We also extend our algorithms to provide fault tolerance and handle significant workload …
Network Slicing and Workload Placement in MegacitiesP Soumplis, P Kokkinos, D Lagos… – 2020 22nd …, 2020 – ieeexplore.ieee.org… ’ workload is appropriately offloaded. In our work, we examine mechanisms for the joint resourceallocation and the applications’ workload placement in … connections (links) between two …
Analyzing network health and congestion in dragonfly-based supercomputersA Bhatele, N Jain, Y Livnat, V Pascucci… – 2016 IEEE …, 2016 – ieeexplore.ieee.org… job workloads. In this paper, we explore the effects of job placement, parallel workloads and
… As we can see, green links have higher maximum traffic than blue links for 2D Stencil, Many-…
placement taking into account … world situation in which the link between the Edge Cloudlets …
fundamentally necessary to deploy backhaul links with adequate bandwidth to facilitate an edge …