Rc5 question(s) - blocks, work units, packets, crashing, and more

[Dee67]

Ok, firstly.. When the client says like:

5 Rc5 pckets are in buff-out.rc5

let's say I crash (hard lock, must reboot) at that very second, do I lose those 5 packets? or are they saved and come back when I restart the client?
-------
ALSO:
Dnet stats go by "blocks", the client speaks of work units, packets and keys.... The question is.. WHAT equals a block? or is there no work unit or packet to block ratio?
--------
ALSO:
I skimmed the faq for acceptable ways to recruit. I couldn't really find anything.. Is posting ONE message in a non-distributed computing area off limits? Or is there some faq I'm missing where there is such info?


If we lose first place I'm going to pull all my hair out :|

 

[micron]

<< let's say I crash (hard lock, must reboot) at that very second, do I lose those 5 packets? or are they saved and come back when I restart the client? >>


Whatever is in you buff-out when your system locks-up is saved . But if you have 99% of a packet completed and your system freezes you lose that packet.


<< Dnet stats go by &quot;blocks&quot;, the client speaks of work units, packets and keys.... The question is.. WHAT equals a block? or is there no work unit or packet to block ratio? >>


A work unit is a block.


<< I skimmed the faq for acceptable ways to recruit. I couldn't really find anything.. Is posting ONE message in a non-distributed computing area off limits? Or is there some faq I'm missing where there is such info? >>


Not sure on this one.

 

[vss1980]

Anything in a buff-out file is safe. Only a HDD error at crashing can cause a problem there. The same goes for your buff-in file.

However, when the client starts a new packet, it effectively loads that into memory and removes that packet from the buff-in file. If your computer resets/locks/melts/explodes/etc., then that particular packet is lost. There is a way around that, but it usually ins't necessary to implement unless you really wanted to. If so, look to using a checkpoint file.

Next: Packets are basically the clients terminology for a collection of work units (WU for short). Mostly work units are distributed in packets as this is better for taking load off of a network than having to send each work unit individually. Packets can contain from 1 to 32 WU's.
When Dnet count Blocks, they are counting work units.

Next: Posting messages about distributed computing in forums that are not related to it, is generally frowned upon by the administrators and other members. Best thing is to put something about it in your sig.

Hope that all helps.

EDIT: Ack, Micron beat me by a minute.

 

[Russ]

As long as you have checkpoint enabled, you won't lose any work in progress. Check your ini file for the following:

[buffers]
checkpoint-filename=cp

A WU and a block are the same thing. A packet is the container that holds the blocks. This container can have anywhere from 1 to 32 blocks inside of it. A single block packet would be 2^28, a 32 block packet, 2^33.

Keys is the smallest unit of work and is contained inside of the blocks. Because modern computers are so fast, it was set up this way to have human understandable numbers.

Russ, NCNE

 

[Polo]

1 WU = 2^28 keys
1 block = 1 packet = from 1 to 32 WUs.

From my old 2.7106.436 GUI client :

[Jan 16 19:07:12 UTC] Sent 1 block (6*2^28 keys) to server.

This block counts for 6 WUs in the stats.

The packets in the buff-out.rc5 file are not lost. If you don't want to lose your current packet then insert a line like this in your DNETC.INI

[buffers]
checkpoint-filename=cp.rc5


[EDIT]I'm too slow at typing...[/EDIT]

and restart your client.

 

[TOOCOOL]

Not wanting to confuse the issue but on the new clients it refers to
stats units instead of work units

 

[Dee67]

you guys are awesome.. can always count on this bunch.. Post my question, take a shower, come back everything is solved..

I only wish ALL my problems could be handled with you guys

Thanks for the quick and helpful responses

 

[RaySun2Be]

I highly recommend using the checkpoint feature, it will save the work unit currently being worked on, and if there is a system crash, will start working on the current block again from the last checkpoint save. A block is a terrible thing to waste. (especially one that's 99% complete!)

 

[JWMiddleton]

Hmmm....thanks for the checkpoint tip! I have had that problem.

 

[hydrobum]

One question here....what's the maximum run-work-limit set to?
I posted a previous thread but it wasn't answered that well that I understood. He said leaving it blank like &quot;removing run-work-limit&quot; from the ini file to let the cow run continuously??

 

[ViRGE]

Just go into the client config, and erase the &quot;shutdown after completed this many packets&quot; option; then it should crack until you manually shut it down.