Large information systems nowadays are required to perform on a Web-scale with millions of users. Especially in the business-to-business context, the required performance is even contractually specified in the form of service level objectives (SLOs). Often, only cloud computing platforms, which provide virtually unlimited resources and on-demand resource leasing, make information systems that meet these SLOs possible. However, the pay-per-use leasing models of the cloud computing platforms still enforce a trade-off between the achievement of SLOs on the one hand and an economical operation on the other hand. Self-adaptive systems can solve this trade-off by autonomously adapting the amount of leased resources to the actual demand. Thus, these systems accomplish an economical operation and still achieve the SLOs at the same time. However, in current practice, self-adaptive systems are developed based on experience of software engineers and rule-of-thumb. Conflicting SLOs or design deficiencies that limit the achievement of the SLOs are often only discovered in late development phases, i.e. in the testing phase or even in the operation phase. Thus, the development of self-adaptive systems is slowed down or is even at risk to fail.With SimuLizar, we provide a model-driven performance prediction method that supports software engineers in identifying design deficiencies that negatively effect the achievement of service level objectives. For that purpose, we introduce the notion of graded achievement of service level objectives, such that trade-offs between conflicting service level objectives can be revealed and solved. With our method, the achievement of services level objectives becomes predictable early at design-time. Thus, also design deficiencies can be revealed early in self-adaptive system development projects and project failures and delays can be averted.