Root MIUI Governor / IO Performance Tests

[GrandTheftBJJ]

As my first post I'll gladly move this if it has already been discussed elsewhere. With that said, here are some benchmark tests I ran while experimenting with governor/io combinations for performance comparison. If anyone has other results or suggestions, I'm more than open to them and will include the findings in the OP.

Constants:
Device - Motorola Triumph
ROM - MIUI 2.4.20
Kernel - MIUI Speed Demon
Launcher - Go Launcher Ex
Min CPU - 24 mhz
Max CPU -2016 mhz
CPU mods using - "No Frills CPU Control" (sorry, can't post links yet)
Testing Measure - "AnTuTu Benchmark" (had FC problems w/ Quadrant)

NOTE: Before the test I randomly selected overclock, governor, and io settings to see which ones were stable to run. I only tested those that did not result in screen tears, FC of Go Launcher, jittering, reboot, etc... As mentioned in the Speed Demon Kernel thread >2 Ghz is very experimental and I could not find any combination to run stably. Before beginning the test I DID read the governor descriptions in the Speed Demon Kernel Thread and research IO descriptions on this forum. The IOs selected for testing were mostly noop and cfq based on the descriptions I found, though I did run some with sio. The governor selection was based on what I thought would be top choices; thus including: Savaged Zen, SmartassV2, InteractiveX, MinMax, Ondemand, and Smoothass. I tested several combinations that are not listed in the "Results" because they caused a reboot or FC.

Benchmark Score Reults (gov/IO):
Savaged Zen / sio - 1601
Savaged Zen / noop - 1608
Savaged Zen / cfq - 1646
InteractiveX / cfq - 2506
Default Setting - (Ondemand / sio; 245-1024 mhz) - 2562
InteractiveX / noop - 2701
Superbad / cfq - 3073
Smoothass / noop - 3109
Smoothass / sio - 3333
Superbad / noop - 3528

Future Tests:
I'll be running battery life tests of the top 5 performers in the next few days and will edit this post to show the results. I do currently have IncrediControl installed but have yet to experiment with it. However, I'll certainly include the variables when I update as well. Again, if anyone has anything to add I will edit accordingly.

 

[dsmryder]

As my first post I'll gladly move this if it has already been discussed elsewhere. With that said, here are some benchmark tests I ran while experimenting with governor/io combinations for performance comparison. If anyone has other results or suggestions, I'm more than open to them and will include the findings in the OP.

Constants:
Device - Motorola Triumph
ROM - MIUI 2.4.20
Kernel - MIUI Speed Demon
Launcher - Go Launcher Ex
Min CPU - 24 mhz
Max CPU -2016 mhz
CPU mods using - "No Frills CPU Control" (sorry, can't post links yet)
Testing Measure - "AnTuTu Benchmark" (had FC problems w/ Quadrant)

NOTE: Before the test I randomly selected overclock, governor, and io settings to see which ones were stable to run. I only tested those that did not result in screen tears, FC of Go Launcher, jittering, reboot, etc... As mentioned in the Speed Demon Kernel thread >2 Ghz is very experimental and I could not find any combination to run stably. Before beginning the test I DID read the governor descriptions in the Speed Demon Kernel Thread and research IO descriptions on this forum. The IOs selected for testing were mostly noop and cfq based on the descriptions I found, though I did run some with sio. The governor selection was based on what I thought would be top choices; thus including: Savaged Zen, SmartassV2, InteractiveX, MinMax, Ondemand, and Smoothass. I tested several combinations that are not listed in the "Results" because they caused a reboot or FC.

Benchmark Score Reults (gov/IO):
Savaged Zen / sio - 1601
Savaged Zen / noop - 1608
Savaged Zen / cfq - 1646
InteractiveX / cfq - 2506
Default Setting - (Ondemand / sio; 245-1024 mhz) - 2562
InteractiveX / noop - 2701
Superbad / cfq - 3073
Smoothass / noop - 3109
Smoothass / sio - 3333
Superbad / noop - 3528

Future Tests:
I'll be running battery life tests of the top 5 performers in the next few days and will edit this post to show the results. I do currently have IncrediControl installed but have yet to experiment with it. However, I'll certainly include the variables when I update as well. Again, if anyone has anything to add I will edit accordingly.

As a note on theese kind of tests, they don't always result in real world performance. With that said, I would look to "feel" as a contributor. With OnDemand Gov. you can set the up kick and limit how much up you get. That gave me the best benefit to my real world usage.

 

[GrandTheftBJJ]

I was trying to be a bit more objective than "feel" Im sure that some of the combinations will be close as far as the numbers go and a lot will come down to personal preference. If there was a clear-cut best, we would all be using the same thing Different situations dictate feel as well like if someone is intent on getting the most out of their battery vs optimal performance or somewhere in between. Thanks for the suggestion!

 

[GrandTheftBJJ]

As promised, I ran battery tests of the top 5 benchmark performers. The same CPU settings used were the same as the previous test (24 - 2016 mhz unless otherwise specified). I also included CPU usage stats to try to avoid skewing of the data. I recorded the percentage of time the CPU was running at its lowest, highest, and two most commonly occurring speeds. The testing time frame for each setup was about 4 hours. Prior to this process I recalibrated my battery via this process:
1) Power on, charge to 100%
2) Power off, charge to 100%
3) Power on, power off, charge to 100%
4) Boot into CWM, advanced, wipe battery stats
5) Power on, run battery completely dead
6) Leave phone off and charge to 100%

Superbad / Noop
Battery Usage: 23% in 260min
Calculated: 5.3% / hour
Est. Max Battery Life: 18 hours 50.4 min
CPU Usage:
24mhz - 17.5%
245mhz -27.3%
368mhz - 24%
2016mhz - 2.2%
Notes: Some process FC, wasn't an inconvenience though

Smoothass / SIO
Battery Usage: 58% in 257min
Calculated: 13.54% / hour
Est. Max Life: 7 hours 23.4 min
CPU Usage:
245mhz - 36.11%
576mhz - 43.42%
1024mhz - 6.47%
2015mhz - 4.11%
Notes: Random process FC, Occasional Facebook FC, 1 Reboot

Smoothass / Noop
Battery Usage: 20% in 229 min
Calculated: 5.24% / hour
Est. Max Life: 19 hours 4.8 min
CPU Usage:
24mhz - 0.03%
245 mhz - 40.1%
576mhz - 50.8%
2016mhz - 2.32%
Notes: No FCs!

Superbad / CFQ
Battery Usage: 28% in 258 min
Calculated: 6.51% / hour
Est. Max Life: 15 hours 21.6 min
CPU Usage:
24mhz - 0.05%
245mhz - 40.12%
576mhz - 34.54%
2016mhz - 1.02%
Notes: Occasional process FC, not an inconvenience

InteractiveX / Noop
Battery Usage: 33% in 242 min
Calculated: 8.18% / hour
Est. Max Life: 12 hours 13.2 min
CPU Usage:
24mhz - 80.33%
921mhz - 10.13%
1024mhz - 2.34%
2016mhz - 0.64%
Notes: 1 Reboot, Occasional FC

Ondemand / SIO (MIUI stock: 245 - 1024mhz)
Battery Usage: 16% in 273 min
Calculated: 3.52% / hour
Est. Max Life: 28 hours 13.2 min
CPU Usage:
245mhz - 15.83%
368mhz - 4.03%
921mhz - 3.95%
1024mhz - 63.96%
Notes: No FCs!

Summary
Smoothass / Noop - was the best overall in my experience. There were absolutely no FCs, it scored a 3109 benchmark (3rd on my list), and had an estimated battery life of a whopping 19+ hours. This will be my go to setup if I experience problems elsewhere.
Superbad / Noop - was the top notch performer on the list (3528 benchmark), had an excellent calculated battery life of nearly 19 hours, and if you don't mind the semi-frequent but not inconvenient android.process.xxx FC it is a sure winner. I will probably use this unless I find that in the long run the FCs do become an annoyance. In which case I'll revert to the Smoothass/Noop combination.
Ondemand / SIO - Without overclocking, this is the default setting on my MIUI ROM/Kernel. Due to the given parameters of this test its benchmark suffered a bit (2562) but it makes up the slack having no FC problems and a ridiculous estimated battery life of nearly 28 1/2 hours. When I need to get the most out of my battery, I will for sure use this setup.
Superbad / CFQ - Kind of middle-of-the-road here. It was 4th on my performance list (3073 benchmark) and an all be it good battery life of well over 15 hours, this combo did not excel in either battery life or performance. Thus making it a good but not preferred choice IMO.
Smoothass / SIO - Things just got ugly. In spite of a 3333 benchmark; this combination had a sub-par estimated battery life of about 7 1/2 hours and problems abound; such as: process FCs, app FCs, and a crash/reboot on top of it all. The only thing keeping this setup from last place is its' 2nd place benchmark. I would not recommend this combo.
InteractiveX / Noop - A benchmark of 2701 makes it rather vanilla and compared to other tested setups the battery life was less than satisfactory, especially considering that the usage stats reveal it was at its lowest clock speed for 80% of the time. Not to mention I had an almost immediate reboot (though none after that) and the somewhat expected android.process.xxx FC. I would not recommend this either.

 

[GrandTheftBJJ]

Perhaps it might be poor taste to bump my own thread, but since the release of MIUI 2.4.20v3 (working incoming calls) I've been really excited to run some performance and battery tests. The following tests were conducted using the Speed Demon 5.7 Kernel, No Frills CPU Control, Go Launcher EX, and Antutu Benchmark. So without further ado, here we go...

Benchmarks Using Antutu (Stock ROM = 2487):
*This list started with some of my personal choosing and ended with the ones that did no result in a reboot, freeze, or FC of Antutu
Governor / IO / Score / Clock
Minmax / noop / 3882 / 24 - 1901mhz
Minmax / sio / 4107 / 24 - 1901mhz
Minmax / cfq / 4277 / 24 - 1901mhz
Minmax / vr / 4110 / 24 - 1901mhz
InteractiveX / cfq / 2715 / 61 - 2016mhz
Superbad / noop / 3540 / 61 - 2016mhz
Superbad / sio / 3546 / 61 - 2016mhz
Superbad / cfq / 3969 / 61 - 2016mhz
Superabd / vr / 3581 / 61 - 2016mhz
Savaged Zen / cfq / 1769 / 61 - 2016mhz
Smoothass / noop / 2462 / 61 - 2016mhz
Smoothass / cfq / 3215 / 24 - 1901mhz
Ondemand / sio / 2468 / 245 - 1024mhz (Default for Rom/Kernel)
*The max CPU on the Minmanx was lowered for stability. Not sure why I increased the min CPU on some of the others, might have been an error to change it on my part.
** I also have Governor/IO/AnTuTu Scores on file for TG-CM7 Reloaded with default Kernel, bKernel, and with the OC Kernel. Also, I have a few with CM7-G60 using the default Kernel and theOC Kernel. However the data is incomplete as I forgot to include the CPU clock settings in my notes.

Battery Tests:
*Stock Rom for me usually ran for about 5 hours with ranges between 3 and 8 hours depending on usage.
**The purpose of this test was NOT to find the best setup with the longest battery life. It WAS to find out, which of the top performers (over 4000 benchmark) had the best battery life. I tried to keep usage typical: having screen dimmed, not having 3G or wifi on when not using, keeping home screens and widgets to a minnimum, etc.. basic battery tips.
Governor / IO / Est. Bat. Life / Test Time - Percent Bat. Used / Clock Low / Clock High / Other Most Occurring Clock
Minmax / sio / 8 hours 19.8 min / 220min - 44% / clock data not available due to accidental reboot (my ignorance's fault, not setup)
Minmax / cfq / 11 hours 6.6 min / 280min-42% / 24mhz-22.4% / 1901mhz-16.06% / 245mhz-0.43%
Minmax / vr / 13 hours 6.6 min / 236min-30% / 24mhz-14.1% / 1901mhz-60.68%
Superbad / cfq / 6 hours 41.4 min / 229min-57% / clock data not available due to freeze reboots

In regards to some lagging issues some have reported with this ROM, I did not experience this to an unbearable degree other than on the Superbad/cfq config. Minmax / cfq was bit laggy on screen rotation and keyboard. Overall winner based on Performance/Battery/UI I'd have to say taht Minmax/vr is the winner here.

Further Tests: I plan to run these same types of tests with the mOCk Kernel which has been highly regarded. I am also working on switching PRLs to post data speed and reception comparisons.

 

[JollyRoger87]

Minmax / vr / 24 - 1901mhz

You speak highly of this setup yet you show no data for the benchmark score... Any reason as to why?

 

[GrandTheftBJJ]

You speak highly of this setup yet you show no data for the benchmark score... Any reason as to why?

updated. Thanks for catching that. Must have just got lost while entering the data.

 

[Lanman1978]

Well I am on CM7 overclocked to 1.6 with smartassV2 and anticipatory and I get decent scores on Quadrant. I just found a more thorough benchmarking program called Smartbench 2012. My scores were not so hot when I ran its tests. You might want give Smartbench a try.

 

[GrandTheftBJJ]

I have heard of that and will look into it. Though the purpose of Benchmarking in the context of this thread isn't so much to compare your score to other phones', but to have the same measuring stick for different configs you're testing. With that in mind, I'd have to go back and redo all previous tests w/ the new "stick" (Smart Bench) because it would be useless trying to compare AnTuTu scores to Quadrant or Smart Bench scores. If nothing else, for consistency's sake, I'll be sticking with AnTuTu for a while.

 

[Lanman1978]

I have heard of that and will look into it. Though the purpose of Benchmarking in the context of this thread isn't so much to compare your score to other phones', but to have the same measuring stick for different configs you're testing. With that in mind, I'd have to go back and redo all previous tests w/ the new "stick" (Smart Bench) because it would be useless trying to compare AnTuTu scores to Quadrant or Smart Bench scores. If nothing else, for consistency's sake, I'll be sticking with AnTuTu for a while.

Well Smartbench 2012 is the most thorough benchmarking program I have come across and it may be useful for your purpose. Give it a try.

 

[GrandTheftBJJ]

First, Lanman1978, I used Smartbench and I agree that it is more thorough, but I am going to stick with AnTuTu for consistency so that I am always able to compare my results across the board. As I mentioned earlier, I would have to redo all my tests with Smartbench for the results to mean anything and that is a time effort I'm not willing to put forth at the moment. Feel free to run your tests and post them in here. We might find something interesting


Performance:
Device - Motorola Triumph
ROM - MIUI 2.4.20v3
Kernel - mOCk
Launcher - Go Launcher EX
CPU Mods - No Frills CPU Control
Benchmarking - AnTuTu
CPU Settings - min = 61mhz; max = 1901mhz
Governor / IO / Score
SmartassV2 / cfq / 4173
SmartassV2 / vr / 4228
SmartassV2 / sio / 4141
SmartassV2 / noop / 4157
InteractiveX / cfq / 2890
Smartass / cfq / 2860
Smartass / vr / 2782
Interactive / cfq / 3956
Minmax / cfq / 4116
Minmax / vr / 4244
Ondemand / cfq / 4234
Ondemand / vr / 4249
*I also did some tweaking with ROM Toolbox Lite from the market just to see what would happen

Ondemand / vr / 4214 (LCD Density changed from 240 to 280)
Ondemand / vr / 4059 (LCD Density Changed from 240 to 200)
Ondemand / vr / 4063 (Heap Size changed from 32 to 20)
Ondemand / vr / 3836 (Heap Size changed from 32 to 50)
Minmax / vr / 4223 (SD Boost changed from 1024 to 4096
Minmax / vr / 4013 (SD Boost changed from 1024 to 128)
*What this tells me is that the Devs did a darn good job building things as none of these tweaks had much of a (if any) positive effect. Another ROM Toolbox tweak that might help with battery life is the memory allotment to various services. However, I think we all know the pros and cons of app-killing.

Battery Use (of the three highest performance configs from above):
Device - Motorola Triumph
ROM - MIUI 2.4.20v3
Kernel - mOCk
Launcher - Go Launcher EX
CPU Mods - No Frills CPU Control
CPU Min - 61mhz
CPU Max - 1901mhz
Governor / IO / Est. Bat. Life / Test Time - % Used / % at clock high / % at low / % at other
Ondemand / vr / 17 hours 45 min / 245 min - 23 / 85.3 / 7.28 / 245mhz - 1.87
Ondemand / cfq / 19 hours 12 min / 288 min - 25 / 76.39 / 12.37 / 245mhz - 2.93
Minmax / vr / 9 hours 13.2 min / 166 min - 30% / 75.64 / 17.07 / 122mhz - 1.34


Final Note(s):
Taking into account these tests and my subjective UI appeal factor, I find that under my usage conditions MIUI 2.4.20v3 w/ mOCk set to Ondemand and VR and CPU @ 61 - 1901mhz came out on top.
If another kernel or MIUI 4 with working camera get released I will continue this thread. If anyone else wants to chime in using another Benchmark App for the same governor / IO configs please feel free to do so. Also, please post if you have tested other configs using the same tools. I'd love to Dropbox a spreadsheet with all of them just for reference.
I hope some users have found this thread useful, its the least I could do since I'm not into dev(ing) ... yet So, I'll be on to running tests with CM7.2.

 

[mattbodin]

thanks,I'll be waiting for your 7.2 results!

 

[GrandTheftBJJ]

thanks,I'll be waiting for your 7.2 results!

If you look through the MTDEV-CM7 thread you can see some of the stuff I have been trying out. I haven't posted any battery / bench comparisons because i've been pretty busy with school lately and haven't had the time to log the data into the forum. but the sampling is done for the most part.

 

[mattbodin]

OK thanks ill take a look.

 

[fusionice]

Thank you so much for performing these tests. Ondemand + cfq on mOCk is awesome. Performance + battery life.

 

[Deleted User]

I like scary gov but it has the sleep of death issues