Ouch- We are bleeding hosts
9 Jul 2007 18:39:36 UTC
Topic 192950
(moderation:
We are down 22,000 hosts since March. I hope things get staightened out soon!
Status Graphs
Ouch- We are bleeding hosts
)
I pulled 35 hosts off this project when they dealt me an application that stuck my farm with a 30% performance penalty. I could no longer justify expending resources on a project with such whiz bang code. Then they decide to test potential fixes on the lab rats that were already being poisoned. I even went as far as replacing some of my K8 CPU's with K7's. Then, having read all the workaround posts, I just said no more. Any project that needs a code hack just to allow my rigs to maintain parity, and then tests fixes in the field has a serious issue in the code grinding department. Spent two and a half years with Einstein, but this latest "enhancement" made me change my zip code.
Regards-tweakster
I really do not come over
)
I really do not come over here very often... Only to see if it is fixed.
I do know that many projects run "Beta Sites" to test these issues... Or the Projects statement is they are a Beta Site... Then issues are resolved before affecting the larger population. It is sad, that someone did not think of this earlier...
Alternatively, some projects
)
Alternatively, some projects have a project-specific setting to accept (or not) beta/test work. That solves having to have multiple sites/projects.
Reno, NV
Team: SETI.USA
RE: ...Any project that
)
Hi tweakster,
It's good to see that you are still around and speaking your mind so eloquently :).
As you would remember, I've been known to express my frustration from time to time and to try to highlight what needs to be fixed. Having got something off my chest, I then remind myself that we are not talking about a big commercial enterprise with a well resourced R&D department and a virtually unlimited budget. We are talking about Academia where we expect brilliant people to have brilliant flashes of inspiration but never enough resources to do things the "business model" way. Once I remind myself of that, I usually decide to try to help them out by throwing a few more resources in their direction.
You are probably aware that things are steadily improving. The 4.24 app did have a short beta test before becoming the default app. Since that time my largely AMD and Intel PIII fleet, quite a few of which were still running Windows have benefitted. Of course, those that had already transitioned to Linux are not affected but the significant number that were still running Windows have shown at least a 25% gain.
So, speaking from a purely personal viewpoint, I hope to see you back here again one day, particularly when (not if) the Akos inspired optimisations make it into the app. The indications are that there may be a considerable speedup at that stage :).
Cheers,
Gary.
RE: RE: ...Any project
)
In the past couple of months appr. every 4th work-unit did not validate due to the cross-platform error. Furthermore the new Mac-PPC client is much slower than the Windows client.
As soon as they solve the cross-platform issue and release a Mac-PPC application that is on par with the Windows apps I will reallocate the 50% CPU time I switched to SETI recently.
Hi! I think there are
)
Hi!
I think there are several reasons for a decrease in participation.
In no special order:
1) Seasonal effect: Summer on the northern hemisphere (except Germany ;-)). This happens every year, as the weather gets hotter, people shut down their computers if not needed ==> should continue for a few more weeks, see the historical data over the last 365 days (source: boinc.netsoft-online.com)
2) Problems with failed validations ==> hopefully fixed or almost fixed, see today's news item
3) Some E@H hosts may have reached their disk quotas because of a bug ==> fixed by now, tidying up unused files will probably happen automatically soon.
4) In general, a higher awareness about energy costs and environmental issues ==> will affect all BOINC projects, except, perhaps, CPDN :-)
5) Upgrading to MS Vista seems to cause some unique (graphics?) problems ==> should be fixed with future releases of BOINC and/or E@H ?
6) Many people prefer smaller Workunits and/or longer deadlines ==> after fixing the validation and stability problems, optimization will be implemented and this will speed up E@H considerably.
7) The promotion effect of the "Physics Year 2005" is declining
8) Those migrating to 64 bit OSes have difficulties running the 32 bit E@H version ==> a new server side version of BOINC should make this more straightforward
So, most of the issues, at least all those that can be controlled by the E@H team, are solved or in the process of being solved. The FLOPS statistics and active hosts statistics of Einstein@Home are already showing signs of stabilizing, and hopefully the negative trend can be reversed pretty quickly.
CU
BRM
RE: Hi! I think there are
)
I have to admit that I have shut down most of my hosts for these two reasons alone. (Especially after I received my last electric bill, now that I'm having to run air-conditioning along with the computers.)
On the other hand, I did add a new dual-core 6000+ machine to the mix a couple of months ago, so it at least can make up for some of the old P-3's that I've shut down.
RE: I have to admit that I
)
Rule of thumb for summer airconditioning is that you spend just as much to get rid of the heat as you do making it in the first place. Shutting down half your boxen over the summer should see you break even in terms of computing energy costs. OTOH if you're living in arizona you might not be able to aford only the breakeven level. :)
RE: RE: I have to admit
)
Actually, it's much worse. I live in southeast Georgia, where there's not only lots of heat, but also lots of humidity. So, I can't get by with using those cheap-to-run swamp coolers like the Arizona folk can.
RE: As soon as they solve
)
I know how you feel. I pulled all my systems off Einstein around the first week of May and went 100% resources with CPDN. I gave it a try here for a month when the new app. was implimented before I couldn't take it anymore. That and waiting 6 weeks for someone to finally finish a WU that would validate and give me the credit. I don't like large WUs either. The biggest difference with CPDN is at least everyday I get some credit from the trickle up effect. Currently I'm averaging about 1550 credits daily with CPDN as opposed to einstein where if I was lucky with the new credit formula it could be anything from 0 to 1200.
I still have about 37 days left before I finish my first WU with CPDN at which time I will see where we stand here. If things here get back to "NORMAL" then I'll be bringing my systems back to einstein as each one finishes up its alotment of work for CPDN.