staff blogs

distributed.net staff keep (relatively) up-to-date logs of their activities in .plan files. These were traditionally available via finger, but we've put them on the web for easier consumption.

2014/02/25

The OGR-27 project has been completed.

Filed under: project status @ 16:09 UTC

Dear friends,

distributed.net is proud to announce the completion of OGR-27!

It is almost five years ago to the day that we began on this exciting journey. Almost 20,000 of you joined us.. without whom, it would have been impossible.

We have proven conclusively by the exhaustive search of all possible rulers that the previously predicted 27-mark ruler is indeed the most optimal one. We were confident that we would find a more optimal ruler during this search, but it was not to be.

We expected it to take us seven years to complete this awesome task but thanks to your efforts recruiting your friends and co-workers to our effort and a little help from Moore’s Law, we did it in five.

The best known ruler is 27/3-12-26-25-29-2-9-36-10-68-1-4-17-53-35-8-16-28-6-14-13-71-18-19-23-7 (length 553). Represented the other way, this is marks at positions 0 3 15 41 66 95 97 106 142 152 220 221 225 242 295 330 338 354 382 388 402 415 486 504 523 546 553.

In total, we verified 302,621,586 unique stubs (2,526 with 3-diffs, 179,120 with 4-diffs, 6,457,815 with 5-diffs and 295,982,125 with 6-diffs), requiring each to have been completed at least two times independently and with an identical node count. Additionally, due to a client implementation bug in some early clients, we required all stubs to have been verified at least once by a client greater than v2.9109.518. This delayed us by a few weeks towards the end of the project but was necessary to ensure that no work was skipped.

The shortest ruler was sent to us 11 times, with one user completing it twice. I will be writing to this user directly for some help picking lottery numbers!

dnetc-ogr27-shortest

We will be sending some distributed.net swag to the lucky winners and hope that you will join us as we move on to our next challenge!

Moo! 

2014/02/18

Getting Ready For OGR-28

Filed under: project status @ 23:23 UTC

Dear friends,

Behind the scenes, we are getting ready to move over to our newest project: OGR-28. This is to discover the most optimal Golomb Ruler with 28 marks. We anticipate that it will take about as long as OGR-27 has. It has some features in common with OGR-27: some of the packets will be very large (up to 1500 Gnodes), and there are three stub spaces. We expect that the first two stub spaces will take about 90 days to complete. Ask me about that last part in a few months time. :)

For the stats freaks among you, the number of stubs for each stubspace of OGR-28 is below:
Stubspace 1: 115676
Stubspace 2: 5823649
Stubspace 3: 518152118

It won’t be necessary for you to upgrade your client software, since OGR-27 clients are ready for OGR-28. If possible, though, I would recommend that you do have the most recent version. If you are running Windows 7 or 8 and your computer is less than four years old, that’s this one: http://http.distributed.net/pub/dcti/current-client/dnetc-win64-amd64.zip :)

If you are running something else, have a look on our download page: http://www.distributed.net/Download_clients

Once again, we will be sending some distributed.net swag to the lucky users who found the Optimal Golomb Ruler this time!

Moo! 

2014/01/22

OGR-27 project almost completed

Filed under: project status @ 09:28 UTC

We’re very close being able to declare OGR-27 finished, but we have just a few more stubs remaining before final completion. The previous stubspaces 27.1, 27.2, and 27.3 are now officially complete.  We expect the last results in 27.4 to be received some time in the next two weeks, and then hope to begin sending out the first OGR-28 stubs within a few hours after that.

As a reminder, no new dnetc client binaries or configuration will be necessary since all existing OGR-27 (OGR-NG) clients are already capable of working on OGR-28 once we begin sending out those workunits.

Look forward to us making another blog post containing some final statistics about OGR-27 within the next two weeks!

(edit: a previous version of this post had some overly optimistic dates)

2013/11/16

November update

Filed under: clients,keyservers,project status @ 01:38 UTC

Dear friends,

We have completed over 96% of the work and are now close to wrapping up OGR-27. Some of you may have noticed some difficulty getting OGR packets earlier today, due to our need to begin recycling of the remaining stubs. As I write this, there are about 20 million left that need working on. It is likely that you may experience difficulty occasionally in the coming weeks when hoping to collect OGR stubs to process. This is due to the master key server being unable to speak and chew at the same time!

If you have a long memory, you may remember a note from me about needing to complete one more verification pass on a small number of stubs, about 2.2% of the total, that had been processed by buggy client versions. Now is the time! Some of you are already working on these stubs. One of our core contributors “Stream” has calculated that it will take us about 3 weeks at our current average (400 Gnode/sec) to complete the verification of the earlier stub spaces.

Today we also mark the 4000th day that we have been working on the RC5-72 project. We passed 3% completion only a few weeks ago and continue to make steady progress.

As ever, you can catch up with us in #distributed on IRC (irc.distributed.net).

Moo!

2013/10/08

mikereed [08-Oct-2013]

Filed under: project status @ 06:18 UTC

It has been a while since I have posted any updates to my blog here. I’ve even had a few notes asking if we’re still working on our projects. I can assure you that we are.

We’re going full speed ahead on OGR. My favourite web page at the moment is http://stats.distributed.net/project/ogr_status.php?project_id=27 , which shows in detail how many stubs we have left to process until OGR-27 is complete.

At the same time, we’re still working on RC5-72; where we have nearly completed 3% of the total work space.  We released new client software over the summer which works around the incompatibility we had with AMD 7xxx series graphics cards. The keyrate history graphs for RC5-72 and OGR (http://stats.distributed.net/keyrate.php?project_id=27 and http://stats.distributed.net/keyrate.php?project_id=8) are back online after a hardware failure.

I hope to have some more updates for you very soon. :)

Moo!

2012/05/04

Finishing up RC5-72 recycling

Filed under: project status — Tags: @ 15:23 UTC

Although Earth Day was a couple of weeks ago, we’ve been diligently “recycling” RC5-72 blocks (ie: re-distributing previously issued but unreturned work assignments) since last November.  We realize that distributing these old blocks has been inconvenient for some of our users because the fragments are much smaller, causing more network and slightly lower keyrates, and we have appreciated everyone’s patience and understanding.  Fortunately, we are very close to resuming work on fresh RC5-72 subspaces and will be able to send out larger, unfragmented blocks within the next day or so!

Keep on crunching! ]:8)

2012/04/07

OGR-27 50% completion

Filed under: project status @ 23:09 UTC

Dear friends,

After three years of effort, we have passed the half-way stage of this ground-breaking project. We thank you for your continued participation.

Moo!

2011/11/27

Project status for November 2011

Filed under: project status — Tags: @ 18:26 UTC

This past week we passed two nice milestones for our current projects:
On November 21, 2011 we passed 2% completion of RC5-72. We also recently passed 1000 days of OGR-27 since starting it in February 24, 2009.

Long time participant Andrew Hime posted a nice summary of the milestone statistics to our mailing list, so I’ll avoid repeating further details and just link to his post:
http://lists.distributed.net/pipermail/rc5/2011-November/042080.html

Today we also started recycling some of the old RC5-72 subspaces. This means that we’ll be sending out blocks that had initially been distributed as long ago as 2002 but had never been fully completed by any client. Since these subspaces are already mostly complete (generally about 85% to 95% finished), we’ll be able to move through these subspaces pretty quickly. Once we finish this pass of the uncompleted blocks, we’ll resume distribution of blocks from fresh subspaces again.

As always, thanks for your participation and support! ]:8)

2011/04/11

OGR-27 client issues

Filed under: clients,project status @ 15:12 UTC

Dear friends,

We are pleased to announce that we have recently identified and fixed some bugs in our OGR client codebase. One of them causes the client to skip part of a packet in very rare cases. It can happen only in OGR-27 and above, so previous projects were not affected. Fixes are included in our updated client, v2.9109.518.

Due to the severity of this bug, we decided to change the rules for stub verification. Now not only two returned results must have the same node count, but also at least one of the results must be returned by a fixed (.518) client. Thus we will be sure that no rulers were skipped. Since we have not started the second pass of OGR-27.4 yet, this is an ideal time to introduce a validation change like this.

We’re asking you to update all of your systems to the fixed client, v2.9109.518, as soon as possible. Results sent from older clients will still be accepted and counted in stats. Updating to the latest client will help us to complete the project faster.

The updated client also recognizes the new Intel Core i3 and AMD Phenom processors, which are becoming increasingly popular among budget-conscious consumers.

You can download the updated client for all major platforms at: http://www.distributed.net/Download_clients.

We thank you for your continued participation in this ground-breaking project.

Moo!

2010/12/29

Wrapping up 2010

Filed under: project status @ 00:34 UTC

Dear friends,

With 2010 coming to an end, we thought this would be a good time to review the tremendous progress we’ve made this year, particularly on the RC5-72 front.

Modern video cards now typically include highly advanced parallel processors that can sometimes be used for general purpose computing. Although only certain types of tasks are currently suitable for GPU acceleration, we’ve found that RC5 falls happily in that category.

We began beta testing our first GPU-based clients for nVidia CUDA and ATI Stream in January and February 2009, respectively. Public testing of the GPU clients continued for about a year until we officially released both the CUDA and Stream clients in January 2010.

Adoption of the new GPU clients has been very successful throughout 2010, due in part to the folks at DNETC@Home. At this moment, about 86% of our incoming completed RC5-72 results come from Stream clients, 5% from CUDA, and 7% from traditional x86 processors.[1] Another exciting milestone is that the overall total results ever received for the RC5-72 project, results from Stream/Win32 are extremely close to surpassing those received from X86/Win32, despite the nearly 7 year lead before the official release of GPU clients.[2]

RC5-72 keyrate for 2009-2010

At the end of 2009, our overall RC5-72 project keyrate was about 250 GKeys/sec. Our current project rate is now about 1.5 TKeys/sec, or about 6 times faster than last year!

If we started RC5-56 right now (which took us 250 days in 1997), and found the key in the same place as last time, it would take us about 18 hours to complete. If we had to check the entire keyspace… it would be a little longer–perhaps 38 hours!

In the coming year, we hope to announce some other exciting developments which should continue to push the performance and awareness of distributed computing.

If you’re on Twitter, be sure to follow us on @dnetc!

Happy Moo Year! ]:8)

Older Posts »