Hello all.

First post in this forum - long time BOINC user.

I came across some threads and information relating to BOINC optimized clients and how they increased the BOINC client benchmarking results.

It seemed harmless enough so I decided to give it a try to see for myself. Sure enough, my scores more than doubled.

My client was not doing round=robin work at the time, though I was set to 10 days and only had four work units currently in queue. I am attached to QMC, SETI and BBC Climate Change.

The QMC was the earliest deadline date, but the work seemed to take twice as long as the expiration date indicated. After the optimized client was "installed", the work seemed to continue more quickly. Though some problems were reported with this client as to corrupting Work Units, this particular QMC one finished successfully.

BOINC_5.5.0_SSE2 is the client I have been using for the last day. Someone pointed out to me that this may be an "illegal" optimization, so I am posting here for an official word on the matter and not some hearsay or belly-aching from another team about points. I have found no links either where I found the optimized client or on the pages of WCG where the cheating remarks are said to originate from.

If I am told here that this is in some way cheating, I will revert back to the official release and apologize to any and all.

Thank You,
Charles

---

Optimised clients sometimes make the science application unstable.
Anyway, BOINC is moving away from using benchmarks for determining the amount of credit, so these special versions of BOINC will be a waste of time.

Climateprediction has NEVER used benchmarks; it allocates a fixed amount of credit each time a small amount of data is returned; i.e. a 'trickle'.

How fast you complete a work unit is only of importance if you're just in it to collect credits. Otherwise the reliablility of the science data is more important.

Amongst its updates the 5.5.0 client (built by a 3rd party) uses a triple multiplier, meaning any credit you request/claim (based on the benchmarks, not on the FPOPs (??)) is three times that of what the normal client would claim on the same project.

Some see this as cheating.
Some projects let their scheduler check if you are using 5.5.0 and when you do, all granted credit is automatically set to 0.00

So it's really up to you if you want to keep using it.

---

Thank You all for replying qickly.

As I suspected when this was first brought to my attention and from what I understand about various clients and Work Units, the only way cheating could be accomplished would be via some kind of multiplier in regards to points earned.



I am curious though as to my perceived impression that the Work Unit in progress during my experiment finished quicker. If there is no basis for this helping the science, I can really care less about the points. I was running the beta QMC, and work was taking almost twice what the deadline reported. This was the only project to suffer from this, BTW. It did finish quicker it seemed and suffered none of the broken Work Unit syndrome as reported by some other users.

Though I care more for the science, doesn't mean I want to loose the credit I have earned, so I will revert back to the normal setup.

Thanks again,
seafsee

---

There's no way it could affect the science app, since the science app is a seperate executable.

---

A clue may be found in the words: beta QMC
You should keep an eye on the forum for that project, and see what the project people have to say from time to time about their WUs.

Greetings,

I wanted to add to this discussion by asking a few questions, if I may.

1) Are there any *real* optimisers?

I mean by this, are there any BOINC clients which are optimised ("tweaked") to increase data processing on specific hardware (Intel, AMD, etc) and/or operating systems (different versions of Windows, "distros" of Linux, etc)?

2) Is there a list of known "false" ones?

In other words, "optimisers" which misreport the amount of WUs/data processed, result in loss of credits (as mentioned by Ageless earlier) or result in crashes?

I believe that the answers to the above questions would be of use to anyone wishing to download/run BOINC clients.

It would also be useful for teams to be able to list "good" and "bad" clients to advise those visiting their forums which ones to use and which to avoid.

Kindest regards

Dragan Glas
Team CastleCops Chief Host
http://www.castlecops.com

---

1) Are there any *real* optimisers?

I mean by this, are there any BOINC clients which are optimised ("tweaked") to increase data processing on specific hardware (Intel, AMD, etc) and/or operating systems (different versions of Windows, "distros" of Linux, etc)?

No, there are none which make the slightest difference in the speed of the science application. There are some Boinc clients which give improvements other than speed (for example, they act somewhat like an account manager, and make large numbers of PCs easier to handle).

2) Is there a list of known "false" ones?

In other words, "optimisers" which misreport the amount of WUs/data processed, result in loss of credits (as mentioned by Ageless earlier) or result in crashes?

Hard to identify, you need to look at the reported benchmarks. There is nothing stopping people building their own but using a standard version number.

---

WCG who is one still working with the median of quorum claim, found a solution to the 5.5.0 / Truxoft and other benchmark manipulation participants.

Maybe that the better way around this whole problem is what Einstein is doing: Not let either the client or the science application do the claiming, but let the application count the amount of Flops and let the server give out the preset amount of credits for that amount of Flops.

I think that even the 5.5.0 client will only get the amount of credit that a true (non-'optimized') client does.

---

Assuming that the science app or return data set hasn't been patched to give a misleading number of flops. There was talk of this as a possibility when Rosetta was discussing ways around the credit-cheat problem.

I prefer the CPDN approach, that is, give everyone the same credit for a work unit. But this only works if either the WUs are all the same size (within 10% for CPDN), or if the project can estimate the size from the starting parameters (I'm sure a statistical approach could be used on a lot of projects).

---

Greetings,

Thank you all for the replies so far.

As I thought - nay, feared! - there aren't any *real* optimisers nor lists of "good"/"bad" clients.

As you both say, MikeMarsUK and Ageless, it might be better to get away from the current method to report something more reliable, such as cpu cycles.

The only problem is that this could still be abused using multipliers, as pointed out.

Hence, the need for the above mentioned lists.

The only people who could be considered "unbiased" in deciding which clients are "good" or "bad" would be those most closely linked with the concept of BOINC.

I realise that this might well add a considerable amount of work to their schedule, but I think it's the only real solution to the problem of "inconsistent reporting" by - mostly third-party - clients.

If such (updated) lists were kept here, for example, they could be used as a standard guide for all those wishing to run clients on their hardware.

Kindest regards,

Dragan Glas
Team CastleCops Chief Host
http://www.castlecops.com

---

If such (updated) lists were kept here, for example, they could be used as a standard guide for all those wishing to run clients on their hardware.

If you or anyone else want to start a list, be my guest. I'll even link to it in my FAQs.

---

Greetings,

Ageless
Thank you for the kind offer!

One could cut the work down considerably by just considering the "official" (standard) BOINC clients - on the basis of "if it isn't official, it's bad!".

Standard NON-BOINC clients - such as Folding@Home, etc - would also be added to this list in their own section.

The only problems are:

1) Is there a list of all clients anywhere? (Whether BOINC/Non-BOINC, "standard" or "non-standard".)
2) Where is it?
3) Who'd be best-suited to examine these? - i.e., skill-set(s) needed to investigate them (coding, etc).
4) Where/Who to co-ordinate such a project?

It seems to me that there's a need for a team of suitable people to work together on such a project - preferably co-ordinated here, if only for the BOINC clients. (If not, well, we could "host" the list at our site - unless someone else wished to do so!?)

I confess I don't know enough about these clients to be one of such a team. If there are any suitable people at our site, I could let them know...

Kindest regards,

Dragan Glas
Team CastleCops Chief Host
http://www.castlecops.com

---

1. The list with official clients can be found at http://boinc.berkeley.edu/dl/?C=M;O=D
A list of unofficial clients? That's probably at http://boinc.berkeley.edu/download_other.php (and missing some).
2. Um...
3. A developer maybe? Not the main ones, but a volunteer developer?
4. I'd say to ask about that on the developers email list at http://www.ssl.berkeley.edu/mailman/listinfo/boinc_dev or the Boinc Projects email list at http://www.ssl.berkeley.edu/mailman/listinfo/boinc_projects. See if they want to know about it or help you out.

I do have a problem with adding non-BOINC clients to the list, though.
You talk about Folding... well, since Folding doesn't want to know about Boinc, do we want to know about Folding? ;)

---

Greetings,

Ageless
Apologies for the delay in replying - night-time and work got in the way! :D

Thank you for the links - they may well prove useful, on their own if not as resources for the suggested project.

I'll post to the email lists - if only to see if anyone has had a similar idea.

I understand your reticence regarding non-BOINC clients. ;)

As I said earlier, even if you only dealt with your own - i.e., BOINC - clients, that would be a considerable help. Other sites - such as ourselves - might then host a non-BOINC clients list whilst providing a link to here for the BOINC list.

Just a thought! ;)

Kindest regards,

Dragan Glas
Team CastleCops Chief Host
http://www.castlecops.com

---

Ageless
Apologies for the delay in replying - night-time and work got in the way! :D

I went to bed after posting that post. :)

As I said earlier, even if you only dealt with your own - i.e., BOINC - clients, that would be a considerable help. Other sites - such as ourselves - might then host a non-BOINC clients list whilst providing a link to here for the BOINC list.

Just a thought! ;)

I am a mere moderator of these forums. I've got two people above me that I can ask what to do on here. I don't think I need to ask this as I think it's a splendid idea.

Heck, even without the tag I would've involved in this conversation and asked anyone to post a list here. We're here to help BOINC users.

And thus when we can make a list of good, usable versions of BOINC, for all platforms available, I think that's a good idea. Preferably with direct links to said version and a small how to to install it. (Which is a crime on Linux versions, last I heard).

---

Greetings,

Ageless
Apologies for the delay in replying - night-time and work got in the way! :D

I went to bed after posting that post. :)

...And here's me feeling guilty that you've been looking at your monitor all this time! :D

I did email the DEV list on the lines of "Is anyone involved in a project to list *good* and *bad* clients?" - but got "BOINCed" (!) ... "You are not allowed to post to this mailing list..." ;)

Like yourself, I've got a number of people above me - not least Robin (Laudanski, co-founder of CastleCops and the Head of the Team CastleCops Committee). So, any real involvement on our site would need to be "OK'd" by TPTB - but I don't really see an issue: our members and visitors are just as concerned at downloading/using both stable and reputable clients. ;D

I'm sure any site would do no less to help actual and potential users.

[By the way, I was looking for a way to create/edit a signature to add our team stats to my posts like those of MikeMarsUK.

...And, no, I haven't read your FAQ. Yet! ;D]

[EDIT - OK, second topic by Jorden - that answers the stats part ... now for the "create a signature" part...]

[EDIT - 8O - That's you!?]

Kindest regards,

Dragan Glas
Team CastleCops Chief Host
http://www.castlecops.com

---

I did email the DEV list on the lines of "Is anyone involved in a project to list *good* and *bad* clients?" - but got "BOINCed" (!) ... "You are not allowed to post to this mailing list..." ;)

You need to register to that list. All of the email lists for that matter. No problem, just register. Or email me on elst93 at gmail dot com and I'll forward it.

Like yourself, I've got a number of people above me - not least Robin (Laudanski, co-founder of CastleCops and the Head of the Team CastleCops Committee). So, any real involvement on our site would need to be "OK'd" by TPTB - but I don't really see an issue: our members and visitors are just as concerned at downloading/using both stable and reputable clients. ;D

As long as the list is available for all to see without needing to register to a forum, I think it's the better solution. And then it doesn't matter who is boss.

My two bosses are hardly seen on these forums.
And in my FAQs I am boss. ;)

[EDIT - 8O - That's you!?]

That's me. :)

---

Greetings,

Ageless
I've chosen the easier of the two and forwarded it to you. If you wish to edit it or pass on the query/suggestion in your own words, please do so.

The list(s) would, of course, be publically-available - they wouldn't be much use to "passers-by" if they had to join first. ;D

Most of our site's advice is publically available, given the fact it's security-oriented - only in cases where a OP has a severe issue (hacked/rootkit, for example) do we take them into a "private" section for expert guidance out-of-sight of the public and the culprit(!)

So, how do you create a signature here - I didn't see an explanation in your FAQ, or am I missing something?!

Kindest regards,

Dragan Glas
Team CastleCops Chief Host
http://www.castlecops.com

---

James, may I call you James? My name is Jord. Your email is forwarded. :)

---