PrimeGrid
Please visit donation page to help the project cover running costs for this month

Toggle Menu

Join PrimeGrid

Returning Participants

Community

Leader Boards

Results

Other

drummers-lowrise
11) Message boards : Number crunching : LLR2 installed on all big LLR projects (Message 143894)
Posted 255 days ago by Pavel Atnashev
There are still some old tasks with app version v8.10. Those are not fast-DC LLR2.
12) Message boards : Number crunching : LLR2 installed on all big LLR projects (Message 143779)
Posted 259 days ago by Pavel Atnashev
I think I should also add this:

A fast DC task doesn't compute the residue of the test. It downloads the certificate and validates it by computing some value. That value makes sense ONLY to the server. The server compares it to the expected value, and if they match, the test was performed flawlessly.

We've already seen unmatched values. It was due to the original host being unstable, with lots of hardware errors. One of the errors happened in the compression stage (not protected by Gerbicz check), leading to a faulty proof. But that is extremely rare.
13) Message boards : Number crunching : LLR2 installed on all big LLR projects (Message 143775)
Posted 259 days ago by Pavel Atnashev
The thing is that you're not double-checking your own own data, you validate a certificate issued by an authority (PrimeGrid server). After the first test you're uploading proof data, which is the minimum amount of data to strictly prove you've performed the test. The server can verify it by performing some costly computation. Instead of that, the server processes the data and converts it into a certificate, essentially offloading computation to a double-checker. Each certificate has a secret part = expected value, computed from the original proof. A double-checker validates certificate by performing calculation and returning a result which should match the expected value stored on server.

This process also involves some form of encryption, so it can't be replayed without knowing server's secret key.

There could be no shortcuts in this process.
14) Message boards : Number crunching : LLR2 installed on all big LLR projects (Message 143745)
Posted 261 days ago by Pavel Atnashev
This is normal behavior. As mentioned before, we had to turn on caching of intermediate checkpoints in memory for GCW and SR5. Due to radix conversion issues of b!=2 numbers, file reading is extremely slow (for now). Checkpoints are stored in memory in ready-to-use format, so they can be accessed fast. Note that they're not needed until the end of the test (until the "compression" stage), so it's okay if they're swapped out to disk.

Please increase BOINC memory limits if you're doing SR5 or GCW.
15) Message boards : Number crunching : LLR2 installed on all big LLR projects (Message 143541)
Posted 267 days ago by Pavel Atnashev
I have a similiar simple AffinityWatcher: https://github.com/patnashev/primeUtils/tree/master/AffinityWatcher
It scans for processes of known names and binds them to configurable "nodes".
16) Message boards : Number crunching : LLR2 installed on all big LLR projects (Message 143535)
Posted 267 days ago by Pavel Atnashev
From my experience there's no need to set affinity if you run LLR on all cores, even if HT is on. Affinity helps if you target CCXs or NUMA nodes.
17) Message boards : Number crunching : LLR2 installed on all big LLR projects (Message 143532)
Posted 267 days ago by Pavel Atnashev
j.sheridan, yes, there were some specific performance problems with b!=2 code that revealed themselves with live SR5 tasks. The fix is coming, but we need to test it thoroughly to not break down everything. The changes are in the very core of the test.
18) Message boards : Number crunching : LLR2 installed on all big LLR projects (Message 143531)
Posted 267 days ago by Pavel Atnashev
Bur, looks like your problem is OS-related. Something prevents threads from running with correct affinity. Note that the process name is different now, tools trained on old LLR may work incorrectly.
19) Message boards : Number crunching : LLR2 installed on all big LLR projects (Message 143514)
Posted 267 days ago by Pavel Atnashev
Yes, SR5 has some performance issues. There are remedies for that in the pipeline.
20) Message boards : Number crunching : LLR2 installed on all big LLR projects (Message 143300)
Posted 274 days ago by Pavel Atnashev
The task performing primality test computes a proof which is uploaded to the server. The proof is converted into a certificate. If the certificate is valid, then the test was performed correctly. Certificate validation still takes some time, so it is sent to a user in a separate task. By running those short tasks you were "double-checking" someone's results.


Next 10 posts
[Return to PrimeGrid main page]
DNS Powered by DNSEXIT.COM
Copyright © 2005 - 2021 Rytis Slatkevičius (contact) and PrimeGrid community. Server load 3.56, 3.66, 3.72
Generated 16 Jun 2021 | 15:39:19 UTC