3. PERCEPCIÓN SOBRE LA FORMACIÓN EN INVESTIGACIÓN
3.2 PERCEPCIÓN DE LOS DOCENTES DE CONTADURÍA PÚBLICA
3.2.5 Participación en eventos académicos
positional analysis
We first introduce well-regulated VCPUs before we discuss how to remove abstraction overhead in compositional analysis for the vC2M systems.
Well-regulated VCPUs
Definition 6.1. A VCPU has well-regulated resource supply patterniff the resource
supply patterns in each of its periods are the same. We call such a VCPU as a well-regulated VCPU.
In other words, if a well-regulated VCPU V Pi does (or does not) execute at t1, the VCPU will (or will not) execute at t1 +k·Πi, where Πi is the VCPU’s period
and k ∈N.
Theorem 6.2. VCPUs are well-regulated VCPUs if they satisfy the following con-
ditions: (1) they use periodic server mechanism to manage their budgets; (2) their periods are harmonic and their release offsets are the same; (3) they are scheduled under EDF scheduling and are schedulable; (4) the scheduler uses a deterministic priority-tie breaking policy: for two VCPUs with the same deadline, the VCPU with a smaller period has higher priority; if they still have the same priority, the VCPU with a smaller index has higher priority.
Proof. We will prove each VCPU satisfying the conditions has a well-regulated re-
source supply.
Because of the condition (1) and (2), the EDF scheduling becomes the fixed- priority scheduling: a VCPU with a smaller period and a smaller index always has higher priority.
For the highest-priority VCPU V P1, it always executes immediately when it starts a new period. So it has a well-regulated resource supply pattern.
For the second-priority VCPU V P2, we will prove it has a well-regulated resource supply pattern by contradiction. Supposet1is the first time when the VCPU supplies differently at t1 and t2 = t1 +k ·Π2, where k ≥ 1. It has two cases: the VCPU executes at t1 but does not executes at t2 and vice versa. In the first case, because the VCPU still has budget at t1 and t1 is the first time when the VCPU’s supply pattern changes, the VCPU should also have budget att2. So the fact that the VCPU
V P2 does not execute at t2 is because the highest-priority VCPU V P1 executes at
t2. The fact that the VCPU V P2 executes at t1 indicates that the highest-priority
VCPU V P1 does not execute at t1. Because VCPUs are harmonic and V P2 has
larger period than V P1, we have Π2 = m·Π1, where m ∈ N and m ≥ 1. Since
t2 =t1+k·Π2 =t1+k·m·Π1, the VCPUV P1 should have the same resource supply at t1 and t2, which contradicts to the observation that the V P1 supplies differently at these two time points. So the first case does not occur. Similarly, we can prove that the second case does not happen either. So it is proved.
By induction, we can prove each VCPU has a well-regulated resource supply.
VCPUs in vC2M can be configured to well-regulated VCPUs with minor modifi-
cations to Xen RTDS scheduler.
The Xen RTDS scheduler, which is used in vC2M for scheduling VCPUs, does
not satisfy the condition (4) in Theorem 6.2: the scheduler breaks the priority-tie in an arbitrary order. We add the deterministic priority-tie breaking policy into the Xen RTDS scheduler with 20 lines of change. With the modified RTDS scheduler, the vC2M system satisfies the condition (4).
The Xen RTDS scheduler uses EDF scheduling. By ensuring the total utilization of VCPUs on each core is no larger than 1,vC2M satisfies the condition (3).
The release offsets of VCPUs under the Xen RTDS scheduler are always 0. By configuring VCPUs’ periods to be harmonic, vC2M satisfies the condition (2).
Although VCPUs are implemented as deferrable server under the Xen RTDS
scheduler, vC2M can run a background CPU-intensive task on each VCPU to turn
them to periodic servers as in [69]. The condition (1) is satisfied. Abstraction overhead-free analysis
We use properties of well-regulated VCPUs to remove the abstraction overhead for the analysis of vC2M.
Lemma 6.3. A well-regulated VCPU V Pi with a PRM model Ωi = (Πi,Θi) always
provides Θi time in any time interval with length Πi.
Proof. We consider a time interval [t1, t1+ Πi] for the VCPUV Pi. If t1 is the start
of a period of the VCPU V Pi, the VCPU provides Θi time in the time interval,
according to the VCPU’s definition.
If t1 is in the middle of a period, we define t0 as the start of the period where ti
resides and let t1 = t0+η. We split the time interval [t1, t1 + Πi] to two intervals [t0+η, t0+ Πi]and(t0+ Πi, t0+η+ Πi]. Because the VCPU’s resource supply repeats
in each period, the resource supply in the second interval (t0 + Πi, t0 +η+ Πi] is
the same with the interval (t0, t0 +η]. By combining the interval (t0, t0 +η] and the interval[t0+η, t0+ Πi], we get the interval[t0, t0+ Πi], which provides Θi time.
The VCPU provides the same amount of time in the time interval [t0, t0+ Πi] and [t1, t1+ Πi]. It is proved.
Lemma 6.4. The resource supply bound function of a well-regulated VCPUV Pi with
a PRM model Ωi = (Πi,Θi) is sbfV Pi(t) = x·Θi+ max{t−x·Πi−y,0}, t≥Πi−Θi 0, otherwise where x=bΠt ic and y= Πi−Θi.
Proof. According to Lemma 6.3, a well-regulated VCPU V Pi always provides Θi
time for any time interval with length Πi. Let the start time of t as the start time
of a hypothetical resource supply period of V Pi, which has length Πi but does not
necessarily overlap with the period of V Pi. The worst-case resource supply occurs
when the VCPU services its resource as late as possible in its hypothetical resource supply period.
When t < Πi − Θi, the VCPU provides 0 time, according to the worst-case
When t≥Πi−Θi, there arebΠticfull hypothetical periods of V Pi, each of which
provides Θi time. In the last partial hypothetical period oft, the VCPU provides no
resource for Πi−Θi time before it provides continuous resource for the rest of the
period. Therefore, the VCPU provides max{t− b t
Πic ·Πi−(Πi−Θi),0} in the last
hypothetical period. In total, the VCPU provides b t
Πic ·Θi + max{t− b
t
Πic ·Πi−
(Πi−Θi),0} resource when t >Πi−Θi.
With the tighter resource supply bound function for well-regulated VCPUs, we can remove the abstraction overhead in computing a VCPU’s parameters from its tasks’ resource demand.
Theorem 6.5. A harmonic task sets τi ={τi1, ...τin}, where τ
j i = (p j i, e j i), is schedu-
lable under EDF on a well-regulated VCPU V Pi with a PRM modelΩi = (Πi,Θi) if
Πi = minτj i∈τip j i and Θi Πi = P τij∈τi eji pji.
Proof. We first prove a widget to be used later: for any value aand k ≥1, k· bakc ≤
bac. Let a = k · m + r, where m is the largest possible integer for which r is
nonnegative. We have bac ≥ k ·m. Because ka = m+ kr, we get m = bakc, and
k·m=k· bakc. So bac ≥k· bakc.
Recall that the task setτi is schedulable if it satisfies Theorem 6.1. We now prove
that the dbf of τi is always no larger than the sbf of the VCPU V Pi with the PRM
model Ωi.
Because the task set is harmonic and Πi = minτj i∈τip j i, each task τ j i’s period is pji =kij·Πi, where k j i ∈N, and WCET is e j i =u j i ·p j i =u j i ·k j i ·Πi, where u j i = eji pji.
The dbf of the task set τi isdbfτi(t) = P τij∈τib t pjic ·e j i = P τij∈τib t kji·Πic · uji·kij·Πji.
According to the proved widget above,dbfτi(t)≤ P τij∈τib t Πic·u j i·Πi =bΠtic· P τij∈τiu j i· Πi =bΠtic ·ΘΠii ·Πi =bΠtic ·Θi.
The sbf of the well-regulated VCPU V Pi with the PRM model Ωi satisfies
sbfV Pi(t)≥ b
t
Πic ·Θi ≥dbfτi(t). It is proved.
VCPUs, we can compute each VCPU’s parameters with Theorem 6.5. We remove the abstraction overhead because each VCPU’s bandwidth is equal to its tasks’ total utilization.