JohnnyFFM / miner-burst

modded Blagominer by modding a modded version from Quibus :-), POC2
32 stars 16 forks source link

Slow read time of native PoC2 plot #8

Open valeradyomin opened 6 years ago

valeradyomin commented 6 years ago

I've converted my 2 plots (drive G:\ and I:\ on Screenshot) using Poc1to2Converter and here what I've got :

screenshot_20180621-152203_info

Native PoC2Plot reading time is slower than all my others PoC1plots. How to explain that conversion on the fly doing by miner is faster reading native?

If I change my miner config and use only these 2 plots (redlined on screenshot) I've got the same result - read time is much lower than was before the fork.

My CPU is SSE2.

JohnnyFFM commented 6 years ago

You i/o throughout is really low. Looks like USB 2.0. do you run anything in background? Like another conversion? What is you cachesize settings?

It‘s not a cpu issue, your cpu is bored and waiting for data...

valeradyomin commented 6 years ago

All my plots are located into internal drives (SATA conect). There is no other hard process which use CPU/RAM or other resourses. No any conversions. Here is the output from cmd. process.txt I've made some tests: 1).

miner.conf_PoC2.txt

config changed for mine only PoC2 plots which was converted. Here is the result 06-22-18 07-41-24 As we see the read time is not changed and far from perfect.

2). miner.conf.All-Plots.txt

This test shows all my plots in work where these 2 drives with the PoC2 plots also much slower then others PoC1 . 06-22-18 07-53-25 And another screenshow: 06-22-18 07-58-25

JohnnyFFM commented 6 years ago

Could you please try the following 2 cachesize settings: 8192 and 16384 and let me know the results? Thx!

valeradyomin commented 6 years ago

CashSize 8192 8192_2 CashSize 16384 16384

JohnnyFFM commented 6 years ago

Ok, this helped a little. The app uses more cache for the on-the-fly conversion. In mixed PoC1<>PoC2 this results in a higher priority for the PoC1 files. Increase CacheSize helped to reduce this effect. There is a couple of things you can do to further improve this:

(1) Create a new parameter in your config called CacheSize2. Set it to 262144. (2) Futher increase CacheSize.

(3) Try out a prerelase of blagominer v.1.170964. This version should restore balance on mixed PoC1<>PoC2 systems. blagominer964.zip

valeradyomin commented 6 years ago

Thanks for advise! I'll try it later and give a results here.

valeradyomin commented 6 years ago

Still bad results with blagominer v.1.170964: 06-26-18 03-20-44

With CacheSize2 262144 a litle better: 06-26-18 04-07-45 Increased CashSize not helped because read speed for drive is almost 10 Mb/s. So for now I prefer mine with just CashSize2 param. It is interesting for me is there are any way to reconvert my PoC2 plots back to PoC1? It would be a great solution for me.

JohnnyFFM commented 6 years ago

Well, there is a way back, but this can‘t be a solution. Best thing you can do is convert the other two drives as well, the mixed setup is causing issues on your setup. However, the last screenshot looks good, your poc2 files finish after 20s and the poc1 take longer. I strongly advise to convert all files to poc2 and then only optimize cachesetting(1). This will then be the same as poc1 before 502k.

valeradyomin commented 6 years ago

I've got error. Can you please explain what does it mean?

06-29-18 07-59-14

JohnnyFFM commented 6 years ago

readfile error = issue with harddisk.