I tried valgrind on a testnet node with cgminer, but so far no luck finding useful clues.

Based on what I have heard so far, it seems this is related to getwork. Since a pool at most generate ~100 blocks a day, coinotron said he had to restart ppcoind once a day, I assume that ppcoind was eating >1gb a day, so this excludes the possibility the leak is related to valid blocks (each block taking about 10KB~100KB).

Now getwork does maintain a list of block candidates and a leak there could account for the rapid memory increase. I didn't change things related to allocation and delete of these block candidates and a code review there didn't reveal obvious problem either. I believe cs_main is also used to serialize getwork access so there shouldn't be concurrency issues with the block candidates.

Andy I seem to remember you were testing with some pool software earlier and also observed memory leak issue. Is there a simple way for me to set up a pool software and reproduce this situation? Can you all (Chris, coinotron, Andy) confirm my observations? How fast is ppcoind leaking memory in your production system? Is the leak speed roughly proportional to the number of getwork requests?

Thanks,

Dunno I feel there is quite a bit noise in the valgrind report. After generating ~100 blocks on my testnet node I got 3KB definitely lost, 6KB indirectly lost and 1KB possibly lost. And the stacks don't look plausible at causing GB range leak problems. It would most definitely help to have a reproducible setup where large amount of memory is known to have leaked.

It does seem proportional to the getwork requests. When my pool hashrate was greater than 100Ghash it was losing 1GB of memory a day. Now that it's low I only need to restart the ppcoind once a week or so when it reaches 4-5GB of memory.

I use getwork, getdifficulty, sendtoaddress and getmemorypool.

Does it show up if you hack the client to not check the hash on blocks and to assume every submission is correct (running on a private testnet of course). This should help find it faster.

How frequently are blocks being checkpointed currently?

Can we leave some space to allow for reorganizations to occur, say one checkpoint every 10 blocks maximum?

Sorry I haven't read the paper. Correct me if I am wrong: I keep some ppcoins in my ppcoind and keep the daemon running, after a while, the proof-of-stake magically add more coins to my wallet or what? Otherwise, how the proof-of-stake wealth is distributed?

Yes. Provided that your wallet is not encrypted.

What was the starting difficulty of PPC? Was it 1? Or 256?

Awwww wierded or scared?