Neo and SMP?

[TheOtherPhil]

Does it work? How efficient is it? I may donate some CPU time to the cause if the client works well and isn't buggy.

[magnav0x]

Currently the Neo algorithm generates random factors in attemp to discover the winning key. Mike Currey (the developer of Neo) is currently working with a new algorithm that will implimented in the near future. Users will have an option of using the current random algorithm or using the new algorithm. The client that was just released not even a week ago seems to be very stable. The only real draw back at the moment is no proxy support.

[PY 222]

As of right now, the NEO client has gotten pretty stable. The only problem is that it will occasionally lose workunits when you upload it to the server. This is pretty minimal and should be around 2-4% of the work you produce.

Regarding SMP, the client is pretty weird. After installing two NEO clients on two separate directories and renaming the two clients to Neo1 and Neo2, both the clients will still write to the same offline.dat file (the file that stores all the information regarding the workunits that you had processed.). This will cause some performance issues but I am not too sure how much.

Also, the NEO client will require you to set process affinity if you are running two clients to fully utilize 100% of the two CPUs (or 50% per CPU). This, I feel is pretty troublesome because if I restart my computer or shut down the client, I must always recheck the processor affinity boxes to obtain maximum crunch time.

So, what I am doing now is utilizing the NEO client on one processor and DF on the other processor. No conflicts and 100% efficiency.

But please do come join us. The more the merrier. :|party|:

[TheOtherPhil]

:thumbs: Thanks for the info....I may put a couple of the uni-processor machines on neo for a while just to see what it's like.

[PY 222]

A couple huh? Oh well, the more the merrier.... :thumbs: Maybe you can pose a more serious threat to magnav0x compared to me.

So go ahead, FORWARD....... :notworthy

[rus6]

It runs much quicker on AMD chips, then Intel... :swear:

Dual CPU support has got much better recently, it now (on ver .515 at least) creates two separate offline.dat files (one in each installed dir), and stores the file pointers separately, that mark the clients position in the file when sending units back to the server.

However there still only seems to be one registry setting for both instances so the settings that are stored in the registry like how often to send units, whether to connect to the network and number of packets queued are common to both. so it shows the total number of packets queued not how many are queued for each client...

Other then that it works fine on my dual box :D

HTH