dnetc - are ultrasparcs better than athlons?

[Special K]

I was running the RC5 clinet on my own athlon 1.2 GHz and was also running it on a sunblade 1000 that I was logged into from my own machine, and the sunblade was beating the athlon. Are they really that much better, or is the client just a specialized thing that runs better on ultrasparcs? I thought athlons had overall better performance than ultrasparcs.

 

[Postman]

Your blade had probably two 750MHz UltraSparcIII's, hence it was faster than the Athlon. D.net client automatically detects multiple processors and fires worker threads for them.

I once tried briefly dnetc on corporate E4500 for giggles, it was fun seeing 14 dnetc processes in top

 

[Special K]

How do I tell that dnetc is running on the blade? I type ps at the command prompt and I dont see it listed there. Is there some flag I need to use?

 

[Postman]

try 'ps -u <user who started dnetc>', or simply 'top'. top should show dnetc as the 1st process since the default sort order is by CPU usage.

 

[Special K]

One other thing - when I start the client it says "automatic processor detection found 1 processor" If there are 2 processors in that box wouldn't the client say so? Or do I need to set any options to specify that there are 2? And wouldn't there have to be 2 in there for it to be able to beat the athlon? I only see one instance of dnetc when I type "top"

EDIT: Another question: If I set the option to use the disk for I/O (set the option "buffer in memory only" to no), will it save the buff-in/buff-out files on my networked drive (seems like everything I work on on one of the sunblades gets saved to my networked drive), or the ultrasparcs hard drive?

 

[ViRGE]

If the client says that it only found 1 processor, then yes, it's only using 1 processor. As for the 1 instance in top, there doesn't have to be 2 in there for Dnet to be using 2 CPUs. The Mac OS X client, for example, uses both processors with one process. The Dnet client may very well be lieing though, it's very odd to hear of a Sparc beating an Athlon. If you want to be sure of it, set the Dnet client to use 2 CPUs(Go into config with -config, then look in menu 3 I belive), and see if you get an average speed boost(1 CPU will still do it at the same speed, but the average will be twice as high, as 2 WUs will be completed instead of 1).

As for the Disk I/O thing, the client will save its buffers in the same directory as where the client is. If the client is on the network drive, then the buffers will be there, and vice versa. Either way, it's a good idea not to store the buffers in memory; sooner or later you'll crash.

 

[Postman]

<< As for the Disk I/O thing, the client will save its buffers in the same directory as where the client is. If the client is on the network drive, then the buffers will be there, and vice versa. Either way, it's a good idea not to store the buffers in memory; sooner or later you'll crash. >>



Solaris? Crash? Those words don't mix!

 

[ViRGE]

That's why I said "sooner or later". Solaris is rock-stable, but even a rock will wash away given enough water.

 

[ICXRa]

<< Solaris is rock-stable, but even a rock will was away given enough water >>



yep