Citation Context

... is a function of the quantum size, the number of threads being scheduled by the SFQ scheduler, and the share of a particular thread. Some proportional share schedulers, such as the lottery scheduler =-=[23]-=-, provide no deterministic performance bound to threads that they schedule. The guarantee given by this kind of PS scheduler is PS s, where s is the asymptotic share that the thread is guaranteed to r...

Citation Context

...es one or more virtual processors with its parent(s) and children. Virtual processors represent the potential for allocation of a physical processor, and are a generalization of scheduler activations =-=[2]-=- that support a hierarchy of arbitrary depth. The important property of the scheduler activation interface is that it permits T4suser−level kernel applications timer scheduling expirations FP hierarch...

Citation Context

... error” and has the type PSBE s, δ. For the earliest eligible virtual deadline first (EEVDF) scheduler [22], δ is the length of the scheduling quantum. For the start-time fair queuing (SFQ) scheduler =-=[8]-=-, δ is more complicated to compute: it is a function of the quantum size, the number of threads being scheduled by the SFQ scheduler, and the share of a particular thread. Some proportional share sche...

Citation Context

...rs provide uniformly slower processors to other schedulers. Any scheduler that has a run-time representation of its deadlines may use a uniformly slower processor. The constant bandwidth server (CBS) =-=[1]-=- provides the same scheduling behavior as a basic, hard reservation. Since it is deadline based, it can be scheduled by a uniformly slower processor. The earliest eligible virtual deadline first (EEVD...

Citation Context

...tees the VP to receive the reserved amount of CPU time. The value of t is chosen by the scheduler and is not made available to the application. Continuous CPU reservations (as defined by Jones et al. =-=[11]-=-) are those that make the following guarantee: given a reservation with amount x and period y, for any time t, the thread will be scheduled for x time units during the time interval [t, t + y]. A cont...

Citation Context

...ends on the particular scheduling algorithm. This guarantee is called “proportional share bounded error” and has the type PSBE s, δ. For the earliest eligible virtual deadline first (EEVDF) scheduler =-=[22]-=-, δ is the length of the scheduling quantum. For the start-time fair queuing (SFQ) scheduler [8], δ is more complicated to compute: it is a function of the quantum size, the number of threads being sc...

Citation Context

...start of the next period. Proportional Share — The PS scheduler that was implemented for HLS implements the start-time fair queuing (SFQ) [8] algorithm with a warp extension similar to the one in BVT =-=[6]-=-. When the warp of all virtual processors is zero, it behaves as an SFQ scheduler. Goyal et al. showed that an SFQ scheduler provides fair resource allocation in a hierarchical scheduling environment ...

Citation Context

... [20] uses a deadline-based real-time scheduler to provide hard real-time guarantees to tasks that it schedules. Since it is deadline-based, it can make use of a USP. The total bandwidth server (TBS) =-=[19]-=- provides soft CPU reservations: it guarantees that a minimum fraction of the total processor bandwidth will be available to entities that it schedules, but it can also take advantage of slack time in...

Citation Context

... workloads that can be placed upon general-purpose operating systems. However, it seems likely that the boundeddelay model could be usefully incorporated into HLS. Finally, CPU inheritance scheduling =-=[7]-=- supports a hierarchy of arbitrary schedulers, but does not provide any facilities for composing schedulers in a way that provides guaranteed scheduling behavior to applications. 10. Conclusions HLS p...

Citation Context

...tle effort. In fact, in our implementation it can be done without even shutting down the system. Another complex scheduler that can be easily created with HLS is one that provides CPU service classes =-=[4]-=-, which were designed to support soft real-time applications such as MPEG decoders whose worst-cast execution times 8 * RESBH 5, 33 RESBH 1, 10 T1 RESBS 5, 33 J2 RES ALL * * ALL RESBH 10, 33 J3 T2 T3 ...

Citation Context

...eduling algorithms are used) does not necessarily require information about individual deadlines at run time. The static partition and bounded-delay resource partition models, developed by Mok et al. =-=[15]-=-, provide a framework for reasoning about hierarchical CPU schedulers. Like HLS, they separate the properties of real-time schedules from the algorithms that provide them. HLS differs from these model...

Citation Context

...U reservation, whether hard or soft, basic or continuous, with amount x and period y may be converted into the guarantee PS x/y. The following theorem was motivated by an observation by Stoica et al. =-=[21]-=-, who said it is possible to provide reservation semantics using a proportional share scheduler that has bounded allocation error. Theorem 6. The guarantee PSBE s, δ can, for any y ≥ δ s , be converte...

Citation Context

...far as we know HLS is the first framework permitting reasoning about general scheduling hierarchies—those that allow user-specified schedulers at all points within the scheduling hierarchy. RED-Linux =-=[24]-=- characterizes each task as a 4-tuple consisting of priority, start time, finish time, and budget. Schedulers are functions from task tuples to effective priorities; the system always schedules the ta...

Citation Context

...sage of the terms hard and soft from the work on portable Resource Kernels [16]. Uniformly Slower Processors — A uniformly slower processor (USP) is the guarantee provided by schedulers such as BSS-I =-=[13]-=-. A USP guarantee has the form RESU r, where r is the fraction of the overall CPU bandwidth allocated to the USP. The granularity over which the reserved fraction of the CPU is to be received is not p...

Citation Context

...where it does not receive the full CPU bandwidth. Therefore, the conversion PS ↦→ PS + is justified. It was also shown that when an SFQ scheduler is scheduled by a fluctuation constrained (FC) server =-=[12]-=-, then the entities scheduled by the SFQ scheduler are also FC servers. An FC server is characterized by two parameters (s, δ). Informally, s is the average share guaranteed to a virtual processor and...

Citation Context

... of schedulers. Third, an API for writing new schedulers should be provided. This would benefit researchers and sophisticated users willing to pay the cost of developing new schedulers. Appendix B of =-=[17]-=- describes such an API. An obstacle to widespread adoption of real-time scheduling techniques in general-purpose operating systems is the fact that these systems can complicate developers’ tasks by in...

Citation Context

...ther than only certain scheduler-chosen intervals. Continuous reservations are provided by schedulers that utilize a (possibly dynamically computed) static schedule, such as Rialto [11] and Rialto/NT =-=[9]-=-, where a thread with a reservation receives its CPU time at the same offset during every period. In contrast, a basic reservation scheduler retains the freedom to schedule a task during any time inte...

Citation Context

...e trivially converted. Schedulers From the Literature — This section explains and justifies the guarantee conversions for the schedulers listed in the bottom part of Table 1. The BSS-I [13] and PShED =-=[14]-=- schedulers provide uniformly slower processors to other schedulers. Any scheduler that has a run-time representation of its deadlines may use a uniformly slower processor. The constant bandwidth serv...

Citation Context

...ts of multi-task applications in the case that all deadlines can be made dynamically available to the EDF-based scheduler at the root of the hierarchy. The Open Environment for real-time applications =-=[5]-=- also uses an EDF scheduler at the root of the hierarchy to allocate processor time to fixedbandwidth application schedulers. In addition, it supports guaranteed schedulability in the presence of glob...

Citation Context

...f f f t f t t, 3 t, 5 t PSBE f f f t, 6 f t, 6 t t t PS f f f f f f f t t NULL f f f f f f f f t ↦→ ALL RESU RESBH RESBS RESCH RESCS PSBE Table 2. Guarantee rewrite matrix The Spring operating system =-=[20]-=- uses a deadline-based real-time scheduler to provide hard real-time guarantees to tasks that it schedules. Since it is deadline-based, it can make use of a USP. The total bandwidth server (TBS) [19] ...

Citation Context

...sure is encouraging vendors to migrate functionality previously performed in dedicated hardware onto the main processor; this includes real-time tasks such as sound mixing and modem signal processing =-=[10]-=-. Of course, powerful hardware alone is not enough: to reliably run combinations of independently developed realtime applications, an operating system must effectively manage system resources such as ...

Citation Context

...rmly slower processor. Rialto [11] and Rialto/NT [9] are reservation schedulers. They could be adapted to be scheduled using a USP because they have an internal representation of their deadlines. SFS =-=[3]-=- is a multiprocessor proportional share scheduler. It does not provide bounded allocation error to entities that it schedules.sALL t t f t f t t t t RESU f t f f f f f t t RESBH f f t t f, 2 t, 1 t, 4...

Citation Context

...n complicate developers’ tasks by increasing the number of low-level APIs that they must understand and use in order to get work done. As a partial solution, we have proposed the CPU Resource Manager =-=[18]-=-, a middleware application that will permit legacy applications to benefit from real-time scheduling techniques by automatically providing real-time guarantees to applications in accordance with appli...