2015-11-29 7:46 GMT-05:00 Yu-An(Victor) Chen <chen...@usc.edu>: > Hi Meng, >
Hi, > > So I will rewrite my setup here again, but this time I shorten the period > and budget for RTDS like you suggested: > Nice! :-) > > ----------------------------------------------------------------------------------------------------------------------------- > (I like this line, BTW. :-D) > > > for xen-credit : 2vms (both vm are given 8 vCPUs) sharing 8 cores (cpu > 0-7) using credit scheduler(both with weight of 800 and capacity of 400) > for xen-rtds: 2 vms (both vm are given 8 vCPUs) sharing 8 cores (cpu0-7) > using RTDS (both with period of 4000(4ms) and budget of 2000(2ms))) > In both setup, dom0 is using 1 core from cpu 8-15 > > In both setup: > > I loaded VM2 with constant running task with total utilization of 4 cores. > and in VM1 I run iterations of tasks of total utilization rate of 1 cores, > 2 cores, 3 cores, 4 cores, and then record their schedulbility. > > ----------------------------------------------------------------------------------------------------------------------------- > > So st_jobs_stats for the missed deadline jobs are: > > trial #1 composed of 2 tasks: total tasks utilization rate = 1 > > (period, exe, deadline)=(21ms,12.023ms,21ms) -> miss all deadline > (period, exe, deadline)=(100ms,37.985ms,100ms) -> no miss > yes, this is the information I need and we can solve the mystery now... Let's look at this task: (period, exe, deadline)=(21ms,12.023ms,21ms) -> miss all deadline Its utilization is 12.023 / 21 ~= 0.5614; Your VCPU utilization is only 2ms / 4ms = 0.5 So even when this task is pinned to one VCPU, it will still miss deadline because it has only one thread. :-) If you use one VCPU with 100% utilization, you should not see the deadline miss from this task. Or at least, it won't have all jobs missing deadline. So basically, it is because your configuration is incorrect. That's also why you need to read some literatures in real-time scheduling theory. ;-) > > trial #2 composed of 2 tasks: total tasks utilization rate = 1 > > (period, exe, deadline)=(68ms,40.685ms,68ms) -> miss all deadline > (period, exe, deadline)=(70ms,28.118ms,70ms) -> no miss > > trial #3 composed of 2 tasks: total tasks utilization rate = 1 > > (period, exe, deadline)=(16ms,11.613ms,16ms) -> miss all deadline > (period, exe, deadline)=(46ms,12.612ms,46ms) -> no miss > > I do notice that for within the task that misses deadline, the completion > time get progressively longer, > for example: for trial #3, a snapshot of the task st_jobs_stats tells me > that > the completion time of the job is 79ms and then 87ms, and then 95ms > That's carry-over workload. It is common in the overloaded situation under real-time scheduling. > > ok, I will look into the reference you provide, I just started my research > in the rt fields, there is still a lot for me to learn. Thank you again for > reply! > Yeah, you need to read and learn the real-time scheduling theory, especially you are doing research in RT field. :-D > > Thank you! > Hope it helps. :-) Best, Meng > > On Sat, Nov 28, 2015 at 7:09 AM, Meng Xu <xumengpa...@gmail.com> wrote: > >> 2015-11-28 7:20 GMT-05:00 Yu-An(Victor) Chen <chen...@usc.edu>: >> > Hi Meng, >> > >> > Thank you so much for being this patience. >> > >> > So a task set is composed of a collection of real-time tasks, and each >> > real-time task is a sequence of jobs that are released periodically... >> All >> > jobs are periodic, where each job Ti is defined by a period (and >> deadline) >> > pi and a worse-case execution time ei, with pi ≥ ei ≥ 0 and pi, ei ∈ >> > integers. Each job is comprised of a number of iterations of floating >> point >> > operations during each job. This is based on the base task.c provided >> with >> > the LITMUSRT userspace library. >> >> I knew this information. Basically, I did the experiment with this >> kind of configuration before. What I want to know is what is the range >> of period, execution and deadline you have for the taskset and when a >> taskset is unschedulable under your experiment, what is the >> st_jobs_stats result from the LITMUS which will show which job of >> which task misses deadline. If you try to service a task with period = >> 100, exe = 50, deadline = 100, specificed as (100, 50, 100), with a >> VCPU with period = 100 and budget = 100, you will never schedule this >> task. The reason is because the VCPU can be unavailable when task is >> released. >> A theoretical analysis can be found in this paper "Periodic Resource >> Model for Compositional RealTime Guarantees" >> >> http://repository.upenn.edu/cgi/viewcontent.cgi?article=1033&context=cis_reports >> >> BTW, I'm assuming you are familiar with the schedulability test for >> real time systems since you are talking about schedulability. If my >> assumption is wrong, you should have a look at the survey paper A >> survey of hard real-time scheduling for multiprocessor systems >> (http://dl.acm.org/citation.cfm?id=1978814). This will at least tell >> you when you should expect a taskset is schedulable in theory. If a >> taskset is claimed unschedulabled in theory, it is very likely >> unscheduble in practice. In addition, it will tell you taskset >> utilization is not the only factor that affects the schedulablity. >> Other factors, such as the highest taks utilization in a taskset, >> scheduling algorithm, task period relation, can affect schedulability >> also. >> >> Best, >> >> Meng >> >> >> > >> > >> > On Fri, Nov 27, 2015 at 4:17 PM, Meng Xu <xumengpa...@gmail.com> wrote: >> >> >> >> 2015-11-27 14:50 GMT-05:00 Yu-An(Victor) Chen <chen...@usc.edu>: >> >> > Hi Dario & Meng, >> >> > >> >> > Thanks for your analysis! >> >> > >> >> > VM1 and VM2 both are given 8 vCPUs and sharing physical CPU 0-7. So >> in >> >> > theory,"VM1 can get the services of 400%" >> >> > And yes, Dario, your explanation about the task utilization is >> correct. >> >> > >> >> > So the resource configuration as I mentioned before is: >> >> > >> >> > for xen-credit : 2vms (both vm are given 8 vCPUs) sharing 8 cores >> (cpu >> >> > 0-7) >> >> > using credit scheduler(both with weight of 800 and capacity of 400) >> >> > for xen-rtds: 2 vms (both vm are given 8 vCPUs) sharing 8 cores >> (cpu0-7) >> >> > using RTDS (both with period of 10000 and budget of 5000) >> >> > In both setup, dom0 is using 1 core from cpu 8-15 >> >> > >> >> > In both setup: >> >> > >> >> > I loaded VM2 with constant running task with total utilization of 4 >> >> > cores. >> >> > and in VM1 I run iterations of tasks of total utilization rate of 1 >> >> > cores, 2 >> >> > cores, 3 cores, 4 cores, and then record their schedulbility. >> >> > >> >> > Attached is the result plot. >> >> > >> >> > >> >> > I have tried with the newest litmust-rt, and rtxen is still >> performing >> >> > poorly. >> >> >> >> What is the characteristics of tasks you generated? When a taskset >> >> miss ddl., which task inside miss deadline? >> >> >> >> Meng >> >> >> >> >> >> > >> >> > Thank you both very much again, if there is any unclear part, please >> >> > lemme >> >> > know, thx! >> >> > >> >> > Victor >> >> > >> >> > >> >> > >> >> > On Fri, Nov 27, 2015 at 9:41 AM, Meng Xu <xumengpa...@gmail.com> >> wrote: >> >> >> >> >> >> 2015-11-27 12:23 GMT-05:00 Dario Faggioli < >> dario.faggi...@citrix.com>: >> >> >> > On Fri, 2015-11-27 at 08:36 -0800, Yu-An(Victor) Chen wrote: >> >> >> >> Hi Dario, >> >> >> >> >> >> >> > Hi, >> >> >> > >> >> >> >> Thanks for the reply! >> >> >> >> >> >> >> > You're welcome. :-) >> >> >> > >> >> >> > I'm adding Meng to Cc... >> >> >> > >> >> >> >> >> >> Thanks! :-) >> >> >> >> >> >> >> My goal for the experiment is to show that xen rtds scheduler is >> >> >> >> better than credit scheduler when it comes to real time tasks. >> >> >> >> so my set up is: >> >> >> >> >> >> >> >> for xen-credit : 2vms sharing 8 cores (cpu 0-7) using credit >> >> >> >> scheduler(both with weight of 800 and capacity of 400) >> >> >> >> >> >> So you set up 400% cpu cap for each VM. In other words, each VM will >> >> >> have computation capacity almost equal to 4 cores. Because VCPUs are >> >> >> also scheduled, the four-core capacity is not equal to 4 physical >> core >> >> >> in bare metal, because the resource supplied to tasks from VCPUs >> also >> >> >> depend on the scheduling pattern (which affect the resource supply >> >> >> pattern) of the VCPUs. >> >> >> >> >> >> >> for xen-rtds: 2 vms sharing 8 cores (cpu0-7) using RTDS (both >> with >> >> >> >> period of 10000 and budget of 5000) >> >> >> >> >> >> How many VCPUs for each VM? If each VM has 4 VCPU, each VM has only >> >> >> 200% CPU capacity, which is only half compared to the configuration >> >> >> you made for credit scheduler. >> >> >> >> >> >> >> in both setup, dom0 is using 1 core from cpu 8-15 >> >> >> >> >> >> Do you have some quick evaluation report (similar to the evaluation >> >> >> section in academic papers) that describe how you did the >> experiments, >> >> >> so that we can have a better guess on where goes wrong. >> >> >> >> >> >> Right now, I'm guessing that: the resource configured for each VM >> >> >> under credit and rtds schedulers are not the same, and it is >> possible >> >> >> that some parameters are not configured correctly. >> >> >> >> >> >> Another thing is that: >> >> >> credit scheduler is work conserving, while RTDS is not. >> >> >> So under the under-loaded situation, you will see credit scheduler >> may >> >> >> work better because it try to use as much resource as it could. You >> >> >> can make the comparision more failrly by setting the cap for credit >> >> >> scheduler as you did, and running some background VM or tasks to >> >> >> consume the idle resource. >> >> >> >> >> >> Meng >> >> > >> >> > >> >> >> >> >> >> >> >> -- >> >> >> >> >> >> ----------- >> >> Meng Xu >> >> PhD Student in Computer and Information Science >> >> University of Pennsylvania >> >> http://www.cis.upenn.edu/~mengxu/ >> > >> > >> >> >> >> -- >> >> >> ----------- >> Meng Xu >> PhD Student in Computer and Information Science >> University of Pennsylvania >> http://www.cis.upenn.edu/~mengxu/ >> > > -- ----------- Meng Xu PhD Student in Computer and Information Science University of Pennsylvania http://www.cis.upenn.edu/~mengxu/
_______________________________________________ Xen-devel mailing list Xen-devel@lists.xen.org http://lists.xen.org/xen-devel
