Hey guys,
I apologize if this has been asked over & over... but I'm a little frustrated here.
Basically I'm seeing absolutely horrible random write results on both of my nexus devices (VZ GNex toro & N10). Bascially the numbers I get don't even show up on the charts (1 or less MB/s on Random Write... around 40 IOPS).
I have done various factory resets, freshly installed my custom ROMs, etc. (I have also run the lagfix trim operation that is under discussion on these boards recently), and have not really seen much difference.
Should I chock these results up to something with my current ROM / kernel that the ROM is using? (I'm running Jelly Belly 12.1 on both devices atm for what its worth).
In general, things have been "ok" performance wise, but I'd really like to know why I'm getting such abysmal I/O performance...
Thanks!
Related
Hello I am pretty new to these forums and the whole Android thing in general so excuse my ignorance if it appears this way .I am a bit of a tweaker at heart , always messing at PCs , Overclocking etc as you can imagine , so I am getting to grips recently with this phone.
so far i have seen my quadrant scores go from high 800s into 2500 as of tonight, tho i have also seen other models scores increase dramatically with an upgrade to froyo - android 2.2 .
what I am asking is this , can I expect another massive jump in performance when froyo is released along with all the speed fixes and hacks that we have today , or are we only extracting the froyo improves to use with the existing 2.1 software . in other words does the hardware have anymore left to give . Hope this is not long winded
Stephen
well, the improvement in quadrant scores for galaxy s comes atm from the lag fixes that you can find here. with froyo properly optimised, we should see some more points, since jit will bring some more performance to the table.
many thanks from your reply , so you think a reflash from 2.1 with all the fixes will be beaten by 2.2 stock
Depends on how Samsung manages to get past the i/o problems. If the stock 2.2 will have the same problems that 2.1 has, then there's no way in the world a stock 2.2 will beat a lagfixed 2.1, since the problem of lag and stalling comes from the slow write capabilities on the internal memory. Even if the processor flies, when the system tries to write something it can't very fast. I had similar problems with a first gen SSD, all was great untill the system needed to write small files on it, then ... stop ).
For sure, a lagfixed 2.2 will beat a lagfixed 2.1
I've always been curious about other file systems rather than rfs on OneNAND devices and their performances.
For example, supercurio, a famous developer of voodoo lagfix/color/sound, formatted /dbdata on i9000 Galaxy S (which is also OneNAND) and reports better performance, although I can't find any evidence.
Since rfs is basically FAT with POSIX and some other stuffs, other "advanced" file systems will do better performances-- this is what I thought. Yes there is no lag that occurs on other Galaxy S on Epic 4G, and it's because our OneNAND is physically superfast NAND flash with SLC NAND, NOR logic and SRAM buffer. Much faster than moviNAND(of Galaxy S) or other NAND flash. That's all. That doesn't mean changing file system from rfs to another will do nothing. With rfs the performance is good enough already. The question is does other file systems increase the performance.
(and Yes quadrant is a terrible benchmark software. It does not give any criterion about its benchmark, and boosts too much points on I/O.)
Deceived by Quantum Rom's quadrant tmpfs trick, I decided to install VIPERrom's ext2 "lagfix" (although I don't want to call it "lag fix" because there is no lag on our Epic 4G) and run RL benchmark of SQLite performance. Since SQLite is sometimes quite relevant to real usage in our Android life, especially for database-using applications, please don't call it mere benchmark. Anyway, the result gives:
In order of
1000 INSERTs /
25000 INSERTs in a transaction /
25000 INSERTs into an indexed table in a transaction /
100 SELECTs without an index /
100 SELECTs on a string comparison /
Creating an index /
5000 SELECTs with an index /
1000 UPDATEs without an index /
25000 UPDATEs with an index /
INSERTs from a SELECT /
DELETE without an index /
DELETE with an index /
DROP TABLE /
Overall
Normal, rfs, DK05 FroYo:
11.738 / 2.91 / 2.903 / 0.096 / 0.076 / 1.095 / 2.442 / 6.618 / 6.908 / 2.524 /
2.991 / 2.651 / 1.998 / Overall 44.95
VIPER ext2 patched, DK17 VIPERrom FroYo:
0.971 / 2.758 / 2.745 / 0.091 / 0.073 / 0.989 / 2.488 / 6.749 / 6.951 / 1.68 /
1.21 / 1.227 / 0.308 / Overall 28.24
(For comparison:
Normal, rfs, Korean M110S Galaxy S, SK22 FroYo (moviNAND)
81.281 / 5.857 / 4.107 / 0.251 / 0.117 / 1.221 / 2.26 / 7.181 / 13.065 / 4.359 / 7.754 / 10.481 / 11.135 / Overall 149.069)
The answer is obvious: ext2 increases I/O performance.
The new questions are:
1) is ext3 or ext4 better than ext2?
2) is it safe to use ext2 on OneNAND devices?
(there are several debates about ext's on moviNAND as well.)
Forgot to mention: it is not possible to support ext3 or 4 at the moment for FroYo now, since there is no source code released yet. (which is of course thing; there is no official froyo yet lol)
chocoberry said:
Forgot to mention: it is not possible to support ext3 or 4 at the moment for FroYo now, since there is no source code released yet. (which is of course thing; there is no official froyo yet lol)
Click to expand...
Click to collapse
If you had been following the Quantum Rom thread, you would have seen that the "Lagfix" was nothing more than a joke to prove a point about Quadrant, and other epic roms using a lagfix.
Froyo is not complete yet, there is no source, there is still debugging code, and it is not optimized for release. This will explain all low scores. Also; issuing a ton of sqlite queries hardly amounts to real life performance on a phone, and once again it is all I/O. (As sqlite is stored in files.) (And yes, I am aware that we use sqlite to store data. But we don't do thousands of transactions in an operation... ever) The same thing happens when you loop an ext FS over RFS, you get inflated values because of the ram buffer between RAM and ext filesystem.
When froyo goes official; I will try my hand at porting a ext4 fs to the epic (Minus journaling, as it kills performance); but Samsung is making improvements to rfs, and the results are getting better. However, you need to realize that I/O performance is not all that matters on a phone, and regardless there are other aspects that will throttle you anyway.
Dameon87 said:
you need to realize that I/O performance is not all that matters on a phone, and regardless there are other aspects that will throttle you anyway.
Click to expand...
Click to collapse
Yes you are true. For epic it's not the I/O which makes bottleneck (not like other SGS), obviously, because it is already on top of all Android devices even with rfs, thanks to OneNAND flash.
But the purpose of thread and benchmark is to test "if" the file system change will increase I/O performance or not on OneNAND devices (which now runs on Sammy's propriety driver). And my conclusion is that it "does" increase performance. That's all. I know that you are a ROM developer (and I'm a Quantum ROM user also) who hears tons of useless quadrant things, but you don't have to do that to me. You overinterpreted the data; I don't want to emphasize or imply anything.
chocoberry said:
Yes you are true. For epic it's not the I/O which makes bottleneck (not like other SGS), obviously, because it is already on top of all Android devices even with rfs, thanks to OneNAND flash.
But the purpose of thread and benchmark is to test "if" the file system change will increase I/O performance or not on OneNAND devices (which now runs on Sammy's propriety driver). And my conclusion is that it "does" increase performance. That's all. I know that you are a ROM developer (and I'm a Quantum ROM user also) who hears tons of useless quadrant things, but you don't have to do that to me. You overinterpreted the data; I don't want to emphasize or imply anything.
Click to expand...
Click to collapse
Oh no. I wasn't trying to come off as an ass at all. I was just giving some information.
We will not know any concrete evidence until official 2.2 is released as sammy has been working on rfs.
Sent from my SPH-D700 using Tapatalk
As I said before..the lagfix would make the phone into a nice database server lol
But part of deploying a database has always been choosing the right file system...:/
Its not like you won't get any increase at all in real life performance...I'd probably guess at around 1% or so..but obviously not enough to justify what people perceive Quadrant shows...hence if your not using Quadrant Pro..really no point of using Quadrant much :/
Dameon87 said:
When froyo goes official; I will try my hand at porting a ext4 fs to the epic (Minus journaling, as it kills performance); but Samsung is making improvements to rfs, and the results are getting better. However, you need to realize that I/O performance is not all that matters on a phone, and regardless there are other aspects that will throttle you anyway.
Click to expand...
Click to collapse
isnt ext2 the fast one, ext 3 the data secure one, and ext4 the best of both worlds? when you removing journaling isnt that basically going to just drop it down to ext2?
I did a factory reset (latest AOSP w/ Motley kernel) and installed nothing but Androbench (didn't even add my Google account) and got these Androbench results:
Squential Read: 24.3 MB/s
Sequential Write: 1.41 MB/s
Random Read: 7.81 MB/s
Random Write 0.23 MB/s
SQLite Insert: 18.92 TPS / 15.85 sec
SQLite Update: 23.45 TPS / 12.79 sec
SQLite Delete: 22.67 TPS / 13.23 sec
Browser: 2830 msec
Market: 891.25 msec
Camera: 1404 msec
Camcorder: 1323.75 msec
Pretty abysmal overall, random write especially falls off the charts altogether. This is really sad since my N7 used to run great but recently has degraded significantly with lots of lag and ANRs.
Is it safe to say I need to RMA this device due to bad storage?
Have you tried using a different ROM? Try one like Paradigm or Paranoid Android and see if there's a difference. When I ran quadrant on some ROMs I got around 2,000 consistently for I/O but when I ran it on some others (eg. Paradigm Android, BAKED Blackbean) I scored 7,000-8,000 consistently (I also used _motley's kernel).
android1234567 said:
Have you tried using a different ROM? Try one like Paradigm or Paranoid Android and see if there's a difference. When I ran quadrant on some ROMs I got around 2,000 consistently for I/O but when I ran it on some others (eg. Paradigm Android, BAKED Blackbean) I scored 7,000-8,000 consistently (I also used _motley's kernel).
Click to expand...
Click to collapse
Thanks for the suggestion. I get ~1,200 for I/O in Quadrant normally.
I installed BAMF Paradigm and my scores got much, much worse (about 1/3rd as fast across the board). I tried installing Trinity kernel on top of Paradigm and it's much better (roughly 2x standard). Unfortunately my random write is the same, which is absurdly, brokenly low.
However my Quadrant score is now ~7000... I just don't know what to believe any more :crying:
What I/O scheduler are you using?
how much space do you have left in your storage?
Using the default scheduler in any given case. And I wiped my SD Card so there's plenty of room left.
I just un-rooted and installed the factory image - the I/O performormance has largely jumped back to normal now, so I'm at a complete loss as to what's going on I'm going to take it and run with stock for a while I think.
Thanks so much for the help all!
So i had read about the problems with the tf300 and decided to see if they had been resolved with the latest color red and white models released recently. Sadly my results are not good. The quadrant i/o scores started around 4000 on ics and dropped to around 1100 on Jellybean. I would love to test some custom roms before taking it back to bestbuy but I'm afraid to get stuck with this.
I wanted to stick with android but there is no ideal purchase in the 10" size range like my Nexus 7.
Has anyone compared the hardware to the nexus 7? I just ran a side by side comparison and the io score is the only difference. They literally run exactly the same otherwise. The reason i ask is if the io hardware is the same there might be hope in the future for a software fix.
I'm sure they will be fixed with a firmware update, my tablet has been running great and no lockup or anything. The only time i notice lockups are when i torrent (anime downloads) on it and sometimes downloading to the internal card, actually scratch that it had some major lag and weirdness when i first got it (upgraded to JB without thinking) but nothing a factory reset and wiping cache/dalvik would not solve. So if i was you maybe do a factory reset and then wipe cache/dalvik using TWRP. If you don't have custom recovery then at least factory rest.
Not a huge deal for me because i use an external card for everything anyways and torrents can be done on my laptop. It is a problem for sure but not enough for me to dumb it for the Nexus 7 because this is the only tablet with a full keyboard and double battery.
Hope
I've heard from other forums too tha with jellybean the i/o score droped to around 1100.
I use a custom rom (Hydro 4.1 from seanzscreams based on Official Jb) and my Tf300 scores around 5400 i/o at quadrant.
So i suggest you to read carefully and try to install a custom rom .
So I have been trying to decide if the micro-stutter I get with my LG G2 V980 is just my hyper sensitivity or if my phone has more than other people's, since many people say it is super smooth. I rooted my phone the other day so that I could side load the tethering app and change the CPU governer from ondemand to interactive using Trickster MOD. I did not benchmark my phone before this, which I should have, but subjectively my does appear faster.
However, I still have some micro-stutter. Not too bad, but there none the less. This got me thinking of TRIM. I know that a new phone that has most of its memory free should not need to be trimmed yet. But for ****s and grins I downloaded lagfix and ran it anyway. It is really too early to tell if it made much difference, but the little I have used my phone since then have been promising.
And this got me thinking - what if some people had trim issues with the phones when they got them? We don't really know what LG has done with some of these phones before we got them, right? I downloaded AndroBench form the play store and ran a microbenchmark. From what I understand, the Random Write speeds are the ones that usually cause a ton of lag/micro stutter in phones. I did not run it before I ran lagfix, but I thought I would post my results now and ask others what they are getting. Especially people that think they have 'stutter' issues. Can you run an AndroBench micro benchmark and post your results?
Mine is attached. Verizon, stock cleanrom, rooted, interactive governor, daily scheduled lagfix.
Edit: Ran the test 2 more times, and got 7 MB/s and then 14 MB/s sequential write.
My results are terrible compared to yours :S but i have no stutter or lag problems, so im a bit confused here.
Btw im on g.lwarnes stock rom v1.06 and his 3.01 beta 2 kernel, also ive used the lagfix and it didnt matter at all.
Cant upload a screenshot so the results are seq read 85k seq write 9.79k random read 12.75 and random write 0.6
Sometimes g2 slow down a bit but nothing to worry about