Hi,
In preparation for a Connect LAS16 hacking sessions, I've been trying to compare schedfreq with schedutil with EAS on Hikey running Android 4.4. In the document below you can view some preliminary results of what I found. Please excuse the brevity; I'm sure way more comments would be needed to make the document more understandable, but I hope it's still useful and I wanted to share it as soon as possible. Please don't hesitate to ask for more information or clarifications (best is by commenting directly on the document, I think).
https://docs.google.com/document/d/1tMb9yfJgaZmVANbbhTWwjLQLA2v5tj-_aTLXDeSu...
tl;dr;
- merging pain wasn't so bad - schedutil relatively close to schedfreq and interactive (even if high percentiles seems to be quite off): perf is lower, but saving some energy - schedutil driven by WALT generally improves figures (but not that much) - it remains to see the amount of work required to put schedtune on top of schedutil
Best,
- Juri
Hi Juri,
In the figure of "SCHEDFREQ governor", i noticed that there is one big drop (from 1000 to 200) for cpu capacity when the util_avg is maximum.
and no such big drop for other governor, is this one bug for sched_freq governor?
On 09/16/2016 09:15 AM, Juri Lelli wrote:
Hi,
In preparation for a Connect LAS16 hacking sessions, I've been trying to compare schedfreq with schedutil with EAS on Hikey running Android 4.4. In the document below you can view some preliminary results of what I found. Please excuse the brevity; I'm sure way more comments would be needed to make the document more understandable, but I hope it's still useful and I wanted to share it as soon as possible. Please don't hesitate to ask for more information or clarifications (best is by commenting directly on the document, I think).
https://docs.google.com/document/d/1tMb9yfJgaZmVANbbhTWwjLQLA2v5tj-_aTLXDeSu...
tl;dr;
- merging pain wasn't so bad
- schedutil relatively close to schedfreq and interactive (even if high percentiles seems to be quite off): perf is lower, but saving some energy
- schedutil driven by WALT generally improves figures (but not that much)
- it remains to see the amount of work required to put schedtune on top of schedutil
Best,
- Juri
eas-dev mailing list eas-dev@lists.linaro.org https://lists.linaro.org/mailman/listinfo/eas-dev
IMPORTANT NOTICE: The contents of this email and any attachments are confidential and may also be privileged. If you are not the intended recipient, please notify the sender immediately and do not disclose the contents to any other person, use it for any purpose, or store or copy the information in any medium. Thank you.
Hi,
On 18/09/16 00:17, hungyh wrote:
Hi Juri,
In the figure of "SCHEDFREQ governor", i noticed that there is one big drop (from 1000 to 200) for cpu capacity when the util_avg is maximum.
Right. First thing to consider is that the util_avg depicted there is PELT utilization signal, while SCHEDFREQ is driven by WALT utilization signal in this case. I'll try to over impose WALT signal and update the plot.
and no such big drop for other governor, is this one bug for sched_freq governor?
I don't think so. It might just be a plotting artifact (we usually have a few of them in this kind of plots). But, I'll collect more traces and see if I can replicate it.
Best,
- Juri
On 09/16/2016 09:15 AM, Juri Lelli wrote:
Hi,
In preparation for a Connect LAS16 hacking sessions, I've been trying to compare schedfreq with schedutil with EAS on Hikey running Android 4.4. In the document below you can view some preliminary results of what I found. Please excuse the brevity; I'm sure way more comments would be needed to make the document more understandable, but I hope it's still useful and I wanted to share it as soon as possible. Please don't hesitate to ask for more information or clarifications (best is by commenting directly on the document, I think).
https://docs.google.com/document/d/1tMb9yfJgaZmVANbbhTWwjLQLA2v5tj-_aTLXDeSu...
tl;dr;
- merging pain wasn't so bad
- schedutil relatively close to schedfreq and interactive (even if high percentiles seems to be quite off): perf is lower, but saving some energy
- schedutil driven by WALT generally improves figures (but not that much)
- it remains to see the amount of work required to put schedtune on top of schedutil
Best,
- Juri
eas-dev mailing list eas-dev@lists.linaro.org https://lists.linaro.org/mailman/listinfo/eas-dev
On 09/16/2016 09:15 AM, Juri Lelli wrote:
Hi,
In preparation for a Connect LAS16 hacking sessions, I've been trying to compare schedfreq with schedutil with EAS on Hikey running Android 4.4. In the document below you can view some preliminary results of what I found. Please excuse the brevity; I'm sure way more comments would be needed to make the document more understandable, but I hope it's still useful and I wanted to share it as soon as possible. Please don't hesitate to ask for more information or clarifications (best is by commenting directly on the document, I think).
https://docs.google.com/document/d/1tMb9yfJgaZmVANbbhTWwjLQLA2v5tj-_aTLXDeSu...
tl;dr;
- merging pain wasn't so bad
- schedutil relatively close to schedfreq and interactive (even if high percentiles seems to be quite off): perf is lower, but saving some energy
When you say "interactive" what version of the interactive governor are you using?
-Saravana
-- Qualcomm Innovation Center, Inc. The Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum, a Linux Foundation Collaborative Project
Hi,
On 19/09/16 14:54, Saravana Kannan wrote:
On 09/16/2016 09:15 AM, Juri Lelli wrote:
Hi,
In preparation for a Connect LAS16 hacking sessions, I've been trying to compare schedfreq with schedutil with EAS on Hikey running Android 4.4. In the document below you can view some preliminary results of what I found. Please excuse the brevity; I'm sure way more comments would be needed to make the document more understandable, but I hope it's still useful and I wanted to share it as soon as possible. Please don't hesitate to ask for more information or clarifications (best is by commenting directly on the document, I think).
https://docs.google.com/document/d/1tMb9yfJgaZmVANbbhTWwjLQLA2v5tj-_aTLXDeSu...
tl;dr;
- merging pain wasn't so bad
- schedutil relatively close to schedfreq and interactive (even if high percentiles seems to be quite off): perf is lower, but saving some energy
When you say "interactive" what version of the interactive governor are you using?
The one that's in the Android 4.4 Hikey build.
Best,
- Juri
On 09/20/2016 02:09 AM, Juri Lelli wrote:
Hi,
On 19/09/16 14:54, Saravana Kannan wrote:
On 09/16/2016 09:15 AM, Juri Lelli wrote:
Hi,
In preparation for a Connect LAS16 hacking sessions, I've been trying to compare schedfreq with schedutil with EAS on Hikey running Android 4.4. In the document below you can view some preliminary results of what I found. Please excuse the brevity; I'm sure way more comments would be needed to make the document more understandable, but I hope it's still useful and I wanted to share it as soon as possible. Please don't hesitate to ask for more information or clarifications (best is by commenting directly on the document, I think).
https://docs.google.com/document/d/1tMb9yfJgaZmVANbbhTWwjLQLA2v5tj-_aTLXDeSu...
tl;dr;
- merging pain wasn't so bad
- schedutil relatively close to schedfreq and interactive (even if high percentiles seems to be quite off): perf is lower, but saving some energy
When you say "interactive" what version of the interactive governor are you using?
The one that's in the Android 4.4 Hikey build.
Hmmm... AFAIK, that one isn't going to get any benefit from WALT. The one that ships with MSM chips is quite a bit different and works closely with WALT. Not saying we'll have to port it over and collect numbers, but this doesn't give a good picture of the interactive governor that ships on a lot of phones.
Put another way, if there are some good parts to interactive (like the tunables it provides) that we'd eventually want to encourage adding to sched-util, it won't be made apparent from this test.
Thanks, Saravana
-- Qualcomm Innovation Center, Inc. The Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum, a Linux Foundation Collaborative Project
On 20/09/16 13:02, Saravana Kannan wrote:
On 09/20/2016 02:09 AM, Juri Lelli wrote:
Hi,
On 19/09/16 14:54, Saravana Kannan wrote:
On 09/16/2016 09:15 AM, Juri Lelli wrote:
Hi,
In preparation for a Connect LAS16 hacking sessions, I've been trying to compare schedfreq with schedutil with EAS on Hikey running Android 4.4. In the document below you can view some preliminary results of what I found. Please excuse the brevity; I'm sure way more comments would be needed to make the document more understandable, but I hope it's still useful and I wanted to share it as soon as possible. Please don't hesitate to ask for more information or clarifications (best is by commenting directly on the document, I think).
https://docs.google.com/document/d/1tMb9yfJgaZmVANbbhTWwjLQLA2v5tj-_aTLXDeSu...
tl;dr;
- merging pain wasn't so bad
- schedutil relatively close to schedfreq and interactive (even if high percentiles seems to be quite off): perf is lower, but saving some energy
When you say "interactive" what version of the interactive governor are you using?
The one that's in the Android 4.4 Hikey build.
Hmmm... AFAIK, that one isn't going to get any benefit from WALT. The one that ships with MSM chips is quite a bit different and works closely with WALT. Not saying we'll have to port it over and collect numbers, but this doesn't give a good picture of the interactive governor that ships on a lot of phones.
Put another way, if there are some good parts to interactive (like the tunables it provides) that we'd eventually want to encourage adding to sched-util, it won't be made apparent from this test.
Right. And I guess that is similarly true for schedfreq (as it is not coupled with schedtune and userspace hints in these tests).
However, I still believe these numbers are somewhat helpful in understanding how far are the "bare" governors from each other in terms of performance and power. We shouldn't use them to drive any conclusive decisions of course but, IMHO they are a good start at comparing things more fairly.
Best,
- Juri
-
hungyh -
Juri Lelli -
Saravana Kannan