PT4 import much slower than PT3 (comparisons results inside)

[kydechuk01]

First, I'm not want to say anything bad about pt4, just test results on a sufficiently powerful machine, so that developers can pay attention to a serious relative increase of overall importing time as compared with PT3.

Config:
CPU: Intel i5-2500K (3.3 Ghz) up to 4.0 Ghz
RAM: 16GB PC3-10600
RAID0 (stripe): 2 x 90 GB OCZ Vertex3 SSD drive

Software:
OS: Windows 7 64-bit
Postgre SQL version: 9.0.6 64-bit
PokerTracker 3: v.3.11.8 beta
PokerTracker 4: v.4.01.6 beta

Import was made to a base, moved by Windows 7 "mklink /J" command from main database folder to RAM-Disk (10 GB) (QSoft RamDisk software) to reduce to a minimum disk dependence.

Test sample: 226k PokerStars hands in 5 files (4 x 50k + 1x26k)

I made 4 passes, everytime to a clean fresh database (create a new base in PT, close PT, stop SQL, move base folder to RAMDisk, start SQL, start PT, import hands, make screenshot, delete base). In PT3 passes housekeeping time does not includes CLuster process, but I made it once for PT3 - it takes 1min 16sec on the RAMdrive and have not of great importance to the results.

So the results:

N1) PT3 import all files in a specified directory
Overall time: ~22 min (Import time: 15 min / Housekeeping time: 7 min) Screenshot 1

N2) PT3 import files one by one
Overall time: ~13 min (Import time: 6 min / Housekeeping time: 7 min) Screenshot 2

N3) PT4 import all files
Overall time: ~29 min (+ 32% time vs N1; + 123% time vs N2) Screenshot 3

N4) PT4 import files one by one.
Overall time: ~32 min (+ 45% time vs N1; + 146% time vs N2) Screenshot 4

Previously, you said in announcements to PT4 that overall import time (housekeeping while import) must be faster than PT3 (import time + housekeeping time after). But my results tells that there is an problem and leak of import rate.


About why I made 2 passes each time. For some unknown reason for me importing speed in PT3 badly degraded after the number of imported hands more than 30-50k. In these tests at the beginning of the first pass PT3 shows import speed about 900-1000 h/s, and then gradually decreases to 250 h/s. But if you import the files one by one, the speed reduces from 1,000 to only 800-850 h/s, and the rate of import of the next file starts with the maximum again. And I tried to verify this behavior in pt4, but its import time only increases with the division of files and unfortunately this method of acceleration does not work anymore (because of maintance time after each import I think).

[kraada]

We are aware of these kinds of comparisons - we have done considerable internal testing. Keeping the cache constantly up to date slows PT4 imports in a nontrivial way. Also, we import more data than we did for PT3 - the PT4 schema now includes EV information which needs to be calculated at import time and inserted in all hands that it matters. Each little thing we do like this takes extra time. But the extra flexibility in the database is well worth it in our opinions.

[kydechuk01]

Each little thing we do like this takes extra time. But the extra flexibility in the database is well worth it in our opinions.

Understand it, I think this is all very good for live import while playing. Most of regular users which are not using big databases will not notice the difference (I'm sure they did not have problems with time before too), but for everyone else (not less important customers) who uses a large databases, making massive imports, and maybe not have a fast PC, a big (1.5-2 times or more as I see) increase in time spent on imports will hit their playing time. May be you will think about new option for manual imports like "delayed cache & EV rebuild"? 10-15-20 minutes difference on just 200k hands on RAMDrive config can turn to hours and hours on ordinary HDD and bigger samples.

[kraada]

We have considered options like those you are discussing and opted not to implement them yet. We certainly may reconsider at some point in the future, especially depending on how feedback is once we have the public beta available, but I cannot make any promises.

[kydechuk01]

I think many of those whom it affects will be very wait your reconsidering 8-)

Despite it, PT4 will be the best anyway :heart:

[kraada]

Thanks for your kind words and your understanding. They really do mean a lot to us.

[PartyAnimal]

... but for everyone else (not less important customers) who uses a large databases, making massive imports, and maybe not have a fast PC, a big (1.5-2 times or more as I see) increase in time spent on imports will hit their playing time.

Does this slower time also apply while playing and auto importing? I'm concerned that the money-losing lags I experience now with PT3 HUD will become worse with PT4. I have to stop the HUD a lot of the time to stop losing money; I may be better off not using a HUD until I spend $2,000+ on a new computer! Most multi-tablers like me will prefer the fastest auto importing and HUD processing while playing, then performing the long maintenance AFTER the grinding session is over.

[kydechuk01]

PartyAnimal, I dont know - just dont know how to measure automatic importing perfomance.

Also, I think developers can implement some feature like automatic collection of perfomance statistics for each internal procedure of PT4 and sending to a cloud server to monitor it (like Windows 7 does) - of course If user allows that in the options. I'm sure this can quickly show all statistical bottlenecks of perfomance and so devs can focus on it.

Kraada, what do you think about this idea? I think this mechanism much better than asking single users to send logs everytime. And not only for beta-stage.

[WhiteRider]

You should not notice any delays during auto imports, and having maintenance done during the import improves HUD performance so actually makes things better overall.

[tarix]

The entire importing of a single hand takes almost no CPU time at all. It does require some inserts into the Postgres database so you do need to be sure that postgres is configured correctly for your needs. If you are a power user the defaults are probably not going to be sufficient.

The HUD itself makes a lot few queries than it did in PT3. You should see a significant reduction in hard disk usage in PT4.

As always if you are seeing performance issues we'll need a lot of information about your usage case. PT4 supports a lot of sites, a lot of Postgres versions, and a lot of hardware configurations and we certainly expect there are some cases that need to be optimized. When you run into this please provide us as much information about your configuration and workflow as possible. In particular the hud layout, the site, number of tables, and of course log files made while you were experiencing the issue are all extremely important.