nand IO speed issues on evervolv JB

[bofslime]

As it seems to be a point of debate on using the sdcard for cache/dalvik/swap in order to improve performance in evervolv JB. I tested it and found some interesting results.

Fresh format, base install evervolv 3.0.0p5 jellybean.
Internal nand:
3.85MB/s, 986.13 IOPS(4k) random read
0.29MB/s, 75.49 IOPS(4k) random write

Sandisk sd:
4.39MB/s, 1124.36 IOPS(4k) random read
0.8MB/s, 205.9 IOPS(4k) random write

ran each test a few times, pretty consistent between. My phone is horribly sluggish under evervolv-jb unless I move data, cache, swap, dalvik to the sd.


Fresh format, base install CM7.2 results:
Internal nand:
5.7 MB/s, 1465.48 IOPS(4k) random read
3.12MB/s, 799.29 IOPS(4k) random write

sdcard:
4.41 MB/s, 1229.81 IOPS(4k) random read
0.58 MB/s, 150.86 IOPS(4k) random write

There seems to be some sort of IO bottleneck on the internal nand in JB on this phone and where the sdcard ends up being faster because of it.
I don't claim to be an expert, just hoping to shed light and hopefully someone is able to identify why this is happening.

 

[tommert38]

Well, that's strange. Let's see if other people using Evervolv JB have such results.

I ran it on CM7.2 too and had the following results (average of 10 benchmarks):

Internal NAND
4.48 MB/s Sequential Read
3.18 MB/s Sequential Write
5.71 MB/s, 1463.41 IOPS(4K) Random Read
3.41 MB/s, 873.57 IOPS(4K) Random Write

16GB Class 10 ADATA µSD
18.76 MB/s Sequential Read
4.33 MB/s Sequential Write
5.14 MB/s, 1316.99 IOPS(4K) Random Read
1.05 MB/s, 270.25 IOPS(4K) Random Write

 

[bofslime]

Hrm, I might have figured it out actually.
I think I still had swappiness still set to 60 when I first ran under JB. I reverted everything this time, and I'm now seeing it at:

Internal nand:
4.95 MBs, 1268.34 IOPS(4K) Random Read
3.05 MBs, 780.85 IOPS(4K) Random Write

sdcard:
4.26M MBs, 1090.76 IOPS(4K) Random Read
0.84 MBs, 216.17 IOPS(4K) Random Write

Thought I was onto something there, but at least it would seem as though setting aggressive swap isn't helpful for speed improvements.

 

[securecrt]

Well, that's strange. Let's see if other people using Evervolv JB have such results.

I ran it on CM7.2 too and had the following results (average of 10 benchmarks):

Internal NAND
4.48 MB/s Sequential Read
3.18 MB/s Sequential Write
5.71 MB/s, 1463.41 IOPS(4K) Random Read
3.41 MB/s, 873.57 IOPS(4K) Random Write

16GB Class 10 ADATA µSD
18.76 MB/s Sequential Read
4.33 MB/s Sequential Write
5.14 MB/s, 1316.99 IOPS(4K) Random Read
1.05 MB/s, 270.25 IOPS(4K) Random Write

How did you make these test?

 

[Spierbal]

How did you make these test?

The app is called Androbench, it basically allows you select /data or /sd card and measure them both.

If you guys don't mind, desire's nand has got the following using (evervolv bravo):
5.77 MB/s, 1478 IOPS random read (4k).
3.61 MB/s, 926 IOPS random write (4k).

While my 32gb class 10 lexar sd card also (evervolv bravo) gets:
4,51MB/s, 1156 IOPS random read 4k
1.11MB/s, 285 IOPS random write 4k

 

[securecrt]

I think we should test 4k 8k 16k 32k 64k random r/w speed.

as the system is not always using 4k to r/w files.

Sent from my NexusHD2 using xda premium