Damn...Sparcs suck at RC5

[Goi]

I'm currently cracking RC5 with a C566@850 and a C433@488 that does about 2.3MKeys/s and 1.3MKeys/s average respectively. I was trying out the UltraSparc client on my college computers(right now I'm telnetting into 3 workstations and running 3 separate clients) but I just went over to the distributed.net's client speed page and noticed that UltraSparcs, in fact all RISC stations suck at RC5 compared to PPCs and x86 machines. That kinda sux...

 

[Assimilator1]

But they do very nicely at SETI 🙂 ,fancy joining our SETI team too?

I like your sig BTW LOL 😉

 

[Joe O]

Goi, The SPARCS will FLY at SETI!


Oh Assimilator beat me to it while I was logging on.

 

[Goi]

Hehe...sorry, but I remember seeing in the login screen MOTD...it said specifically that running SETI was prohibited, but didn't say anything about RC5 which was why I even tried. Even though I've graduated and don't really care if they cancel my account(its gonna be cancelled sooner or later anyway), I still don't wanna be in their black list, if ya know what I mean 🙂

 

[Goi]

BTW, does anyone know how to flush work while its cracking? I'm running this remotely on Solaris so I can't right click->flushwork 🙁

 

[nukefarmer]

try opening another telnet session and do 'dnetc -flush'. That should start another client which flushes the buffers and then exits again..

 

[Goi]

Thanks...that works...
Another question. If I telnet into multipler machines but they all use the same AFS/IFS space, do I needto create separate directories and copy dnetc and dnetc.ini into each directory for each telnet window, or can I run them on different machines in the same directory? Will the buff-in and buff-out files get intefered with each process?

btw they're really slow, the fastest speed I've seen so far on the ultrasparcs are slightly less than 300KKeys/s. Even my C433 does 1.1MKeys/s

 

[imported_Thunder]

Goi,

I've not even checked into this, but I thought I'd throw it out as a suggestion. If they have a D.Net client for Solaris that is capable of doing OGR, then you might want to at least see if they perform better at that.

Since you're prohibited from doing SETI and doing RC5 on a Sparc is almost a waste of time, it's at least a 3rd option to consider.

-Brian

 

[br0wn]

I don't want to ruin your fun, but
you need to get permission to run rc5 btw
(otherwise you are risking the team to get
disqualified).

It seems that they don't want ppls to run SETI,
I think it applies to any distributed projects
(they can't list all projects right).

 

[imported_Thunder]

br0wn,

I'm not completely positive on this, but you might be mistaken in your assumption that no SETI= no distributed projects. The reason is quite simply this; The SETI client utilizes FAR more resources on most CPU/OS's than any other distributed computing project that I know of.

You'll have to have some of the S@H experts here chime in on this to confirm it, but I would imagine that the problem lies in the fact that the SETI client may degrade the performance of other processes running on the Sparc, whereas most other distributed clients will not. I have to admit that I don't follow the S@H developements that closely, so I may be totally off base here, but that's my guess.

-Brian

 

[Goi]

hmmn, ok...I'll look into that...

Anyway, if I crack OGR instead of RC5, will the blocks still go towards Team Anandtech? Are they 2 completely different projects or 1 and the same? I don't really get the difference between RC5, OGR, DES and CSC.

 

[ViRGE]

Goi, all the work will still count towards TA. All OGR is, is another project with its own stats page, it still uses Dnet's equipment and master DB to decide what team gets the blocks.🙂

 

[imported_Thunder]

Goi,

DES & CSC were older encryption cracking projects that are done (at least for now). As far as RC5 and OGR, they are completely different projects, but Team AnandTech has folks contributing to both.

There's absolutely nothing you need to do as far as changing your ID or team affiliation in order for your work to count for the TA OGR project. i.e. If you're cracking for TA on RC5 then you're automatically on the OGR team as soon as any work is submitted.

It does appear to me that the current Solaris 2.x/UltraSparc client should be capable of doing OGR (ver 2.8009.460). However, like I said earlier, I have no idea how well it will perform. Just to give you an idea if your performance is good though, I can tell you that a P6 class core CPU @ 500Mhz gets about 3.4-3.5 Meganodes/sec. with a 2.8010.463b client.

For an explanation of the actual project goals The D.Net OGR project page makes for a good 10 minute read to get you up to speed. 🙂

I hope this helps,

-Brian

 

[Goi]

Thanks Thunder...I did a benchmark of OGR on the UltraSPARCs and only got like 800Knodes/s. Looks like its still slow...

 

[Lord Demios]

Goi, as an old admin of a University Unix lab, the one thing that really made us mad was when people ran D projects, but didn't NICE them to 19. I would make sure you do that. That way you only get a warnning about not being able to run them.

LD

 

[nexus9]

Sparcs really fly with floating point math - projects like seti and ECDL (which is over, unfortunately). They aren't very good on integer stuff like RC5 and OGR. BUT, if you're not gonna run SETI, anything you can give to the OGR team would be GREATLY appreciated!

-Nexus9

 

[Goi]

The default dnetc NICE value is already 19... 🙂
Anyway, I actually got 800+KKeys/s RC5 rate on ultrasparcs that don't have any other processors running...not too shabby compared to the 300+ I previously got.