Posts by Lee Carre

1) Message boards : Rosetta@home Science : Reasons some people avoid BOINC projects (Message 43036)
Posted 2 Jul 2007 by Lee Carre
Post:
Here is some feedback on the issue (from a company's perspective)

As I always thought, CEO is citing potential vulnerabilities problems. I know that BOINC science projects will try their best, with operating firewalled servers and signing executables on a non-Internet-connected PC, but I prefer to play it safe (especially since modern OSes make it easy).

QUOTE - source

[snip]

Anybody have any similar experiences? Any suggestions how I might politely address these concerns?"
I've had a few cases of people using the "security wild-card" (for those that don't know the joke, "it's a security risk" is used when no valid reason can be given). Even when I went into the details of the security measures, which proved their "concerns" invalid, they still wouldn't budge.

Unfortunetly companies are ruthless these days, anything that even hints at requiring time/money and isn't critically needed is pushed aside.
Even when I've approached charities and non-profits along the lines of "helping out other non-profits" etc. they're still not interested.
This has been the same with educational institutions too, such as local schools/colleges (which where I live adds up to a few thousand (modern) computers!)

As much as I'd like to see BOINC (or something similar) adopted by most of the worlds computers, I have doubts that it will be achieved because it's just another "hassle" to most people (dispite the benifit they'll directly receive from something like medical research).

Even getting people to switch from internet explorer to something better (usually firefox or opera) has proved difficult, even when my main point has been security and showing people how badly infected their computers are (usually due to IE) with malware scanning software.

I don't mean to be pesimistic, but it just seems that unless you're going to change the fundamental nature of people (as in, making them want to do it) you're going to be hard-pressed to get very far.

One way of showing this trend in history is that scientific projects have always had virtually no budget (no funding). If people really wanted to make a difference, then surely they'd donate to such projects?
But as we've seen, this hasn't happened, and even recent projects are still being run on minimal budgets to the point of reducing their effectiveness, or quality of results, which i personally find rather disapointing.

As has been stated before, we do it because we want to. For myself this is entirely true, I believe in the greater good, and the little cost to me for running DC projects is trivial compared to the huge benefit that can be gained.
But that remains my personal view, and no matter how much I believe something is a worthy cause it doesn't change other people's opinions.
2) Message boards : Rosetta@home Science : Reasons some people avoid BOINC projects (Message 43034)
Posted 2 Jul 2007 by Lee Carre
Post:
I stumbled accross this old thread and thought I'd offer a few additions

Security: How can I be sure the program isn't searching my harddrive for private information?

I found this in the BOINC Wiki: Security but it doesn't explain how to protect your data.

Someone else pointed me to Trux: BOINC Security but it didn't seem to give me the warm fuzzy that installing BOINC is a safe thing to do. Keeps talking about exposures and risks and how you might not follow all of their suggestions and create an exposure.

I'm looking for something that would actually convince an IT director that BOINC might be an OK thing to allow on a company full of machines, and that explains how THEY control how things will run so that no exposures are created. In fact, something that says "If you're gonna do a DC project, do it with BOINC! ...and here's why". Does anything like that exist?
Personally the primary thing to mention is that the user has control over which projects they run; if they don't trust one/some/many of them, they don't have to run them.
Such things can be determined by community acceptance; for example which projects are getting the most CPU time from crunchers.

As stated in the quote, if someone is really concerned about proper security (which is never an easy or simple thing) then they can "lock-down" their machine by running boinc under a highly restricted account.
There are very rarely simple or easy ways (especially not by using defaults that leave a system in a working state) to make something perfectly secure as a default installation - most security requires user-intervention and selection/customisation; eg the user deciding if something is safe or not.

Hard drive wear: I mentioned Rosetta to a friend of mine that holds a few patents in the hard disk drive field. His first thought was that a standard PC disk drive isn't built to take the wear of being on 24/7, and actually being used.

I realize I can set my preferences to put a timer on how frequently BOINC writes to disk and reduce it's use. And this probably works well for reducing power requirements too, because I can set a high value and let the disk spin down and go idle for periods of time.

...but I was wondering if anyone has outlined how to size and utilize a virtual disk to bare the brundt of the workunit IO? The virtual disk could be on a mapped drive on a server, or a ramdisk or even a memory stick or something. But sizing it properly would seem important, and then how to get BOINC to use it just for specific files that are written during WU processing. Not for all the program code and input files that are only read once.
Most disks last longer if they're left on 24/7 compared to being on for only 2 hours a day (as an example). This includes disks in home computers, due to the fundamental nature of the machanics. A disk will die from thermal-stress (mentioned later) before it will die from wear-out. I've been through plenty of disk of many different types, and never seen a disk die from normal wear-out (excepting early failures etc.) before it died from thermal-stress.

Doing as suggested and setting the power options to spin the disk down and back up now and then will greatly reduce it's life, and doesn't save any significant amount of power anyway due to the relatively huge amount of energy needed to get the spindle back up to operating speed - the energy needed to keep a disk spinning at operational RPM is pretty low in comparison.

A server disk would be restrained by the same fundamental mechanical limitations, and flash-memory (eg USB-sticks) are only capable of a finite number of writes - besides the fact that they're generally slower anyway, so you might as well just use a regular disk.

The idea of using a RAM-disk is the same as telling the BOINC core-client that it can't write to disk very often - the data is stored in volatile memory.

This basically comes down to a choice between having frequent checkpoints, and reducing disk I/O.

If a user is really concerned about it, recommend that they purchase a seperate (smaller) disk just for BOINC. But they're likely to need/want to upgrade their operating system before a disk would fail.

Overheating: Same friend also mentioned that many standard PCs have undersized heatsinks, and will tend to overheat if CPU is kept 100% busy the way these DC projects tend to do.

A resolution has been posted elsewhere, I wanted to add it here to make a more complete resource for people to reference.

You can use the ThreadMaster to turn the CPU neddle back a few notches and basically NOT be used 100% of the time, thus leaving more time for cooling air to flow through the box.

The Rosetta requirements page points out the overheating problem, but not the solution or how to determine if this will be a problem for you, or what harm (or lack thereof) occurs when your PC overheats.

I appreciate your honesty and not wanting to harm anyone's machine(s), but perhaps it could be reworded so as not to STOP you in the process you've gotten this far in to, to participate in the project. Some will see this as a red flag, and not continue... even though their PC will do just fine.
this is certainly true of cheaper machines, but machines from quality manufacturers, such as Dell, are generally pretty impervious to this problem.

The real problem is with the PC, not the DC project, but I appreciate that this doesn't help those with cooling problems. I'd advise that people in this category fix their cooling problems in preference to using some kind of process/thread throttling - eg, what if the throttling fails? or they have a really hot day or something?

Most CPUs will turn themselves off if they get excessively hot to prevent damage. So random apparent power-loss (when the supply of power isn't a problem) is usually an indicator of a cooling problem. Temperature monitoring software can be used to get an actual temperature reading to compare to the CPU tollerances detailed in it's relavent spec(s).

As for documentation wording, perhaps a disclaimer-style thing, saying that this shouldn't be a problem for the vast majority, but some machines due to low quality or non-reputable manufacturers may be affected. In which case they should seek a warrenty claim or similar, in general taking the machine back to the company they purchased it from for it to have sufficient cooling added.

My point being that machines should be able to run at 100% capacity 100% of the time without problems (besides general maintainence, such as removing dust).
If this isn't the case, then it's a problem with the machine (it doesn't really make much difference what software is causing the high usage, many demanding apps would cause the same behaviour; such as editing a large image, video/animation editing etc.
Because most "home" users are just doing simple tasks, using say 5% of the capacity (such as web browsing) then manufactures are tempeted to reduce cooling, to reduce costs, to under-cut competitors - and sadly many people will go for the "bargin" machine, dispite the fact that you usually get what you pay for.

Power consumption: I read on another thread an estimate that a crunching CPU takes about 60 watts more power than a PC that's on but idle. Has anyone seen any reasearch confirming this number??

If that's right, then the incremental cost, 24/7/365 is about $42 per year (at 8 cents per Kwh). And if you live in Minnesota (like me) half of the year the electricity will actually help cut your heating bills (although if you live in California or other tropical clim. then the A/C costs must be factored in to the cost as well).
that all really depends on what the type of machine, the type and class of processor, and a few other factors such as how it's used exactly (different types of processing are more/less effecient in varying circumstances).
Even things like power supply units make a difference; they're more efficient the closer they are to 100% capacity.
If you've got one with a huge capacity (say something ridiculous for today's computers, like 3000W) but only say 5% of it's capacity is being used (150W) then the process of conversion taking place (again, due to the fundamentals of physics and that nothing is 100% effecient) then it's going to consume more power than say a 160W PSU, which would be running at ~94%.

For today's computers, however; 60 watts is reasonable, but it stands to reason that if something is doing more "work" (in the physics sense) then it needs more "power" to do that "work" in a given amount of time (compared to not doing the "work").

As for heating/cooling the building in which the computer is housed, that's down to proper placement of the machine.
As in, if you want it to have a heating effect, then placing it more centrally will be better, to get better cooling, placing it near an outside wall with a window that can be opened would be sensible.

Again, fundamental factors such as computers can only run within certain temperatures are a reality that has to be dealt with.
If it's really a problem then the user can set up a processing schedule for the summer so that the computer only processes data when it's cool (eg, at night, or during the winter only)

PC longevity: Will leaving my PC on more of the time, and running it harder, reduce it's lifespan?

For starters... what happened with your last PC? Did it reach the end of it's life? Or just the end of it's usefulness due to the bigger better faster more effect of time?

I know that from a chipmakers point of view, that with a multi-layered circuit-board you want to leave it at a steady temp. all the time (i.e. minimize expansion and contraction due to heat, by leaving it on all the time). And, therefore, microfractures between those layers won't occur, extending useful life. But is this true for the REST of the machine?
I quote agree with the sensible statements made here.
For example old machines (even 486's) can be used for networking purposes, such as NAT routers, firewalls, DHCP/DNS servers etc. to great effect.

From a technical stand-point, leaving a computer (but not the monitor) on 24/7 will increase it's life. As for running it at 100% capacity 100% of the time, I'm not so sure; obviously in some cases you'll reach wear-out (mainly of mechanical devices, such as disks) sooner, but these have such a long life anyway (we're talking decades, even longer with modern disks - I've got some really old server disks which are still running perfectly years after they stopped being used in a real server, this is because they were treated well and were always on (as most servers are).
As for electronics, I don't think it makes any difference as long as they're at a fairly consistent temperature, and don't over-heat. I see no reason why they'd last longer sitting at say 40 degrees all their life, compared to 50-60 degrees all their life, just as long as they're within operational limits.

Just to clarify a few points, other things besides electronics will last longer if they're at a constant temperature (eg, on or off 24/7, but you can only use a computer when it's on anyway, hence "leave it on").
Disks suffer from thermal-stress far more than wear-out, so turning them on/off lots will kill them quicker too.

However, things like monitors, due to the fundamental ways they work, and inherent properties of the technologies, suffer from wear-out far more (when was the last time you saw a monitor "blow-up" as it would from thermal-stress fractures? they usually just fizzle out and die).
Monitors only have a rather limited number of "on" hours, so if you consider their "lifespan" to be how many years they're useful and working (in general, including "off" time) then the less time they're "on" the longer their lifespan will be.
So in short; monitors especially will last longer the less time they're on. I'm not saying they should be turned on/off by the second, if you're not going to use it for an hour or more, then sure, turn it off, but if it's just a few minutes, turning it on/off lots over short periods will probably kill it quicker.
I'm also not saying that people should use their monitor less, I'm just saying that due to the limited number of "on" hours these devices have, they should be turned off when not in use for a reasonable length of time (again, an hour+).

As for power saving, a monitor is one of the biggest users in most systems, so the single most significant power-saving act you can make is turning off your monitor when not in use.

Network bandwidth: But our network is on the brink now! We can't add any workload to it.

For starters, sounds like a network upgrade may be due. But, otherwise, you can control the hours of the day and/or the bandwidth BOINC will consume in the General Preferences.

Let's say we configure a whole cube farm of PCs to allow network usage only between 7PM and 7AM each day (and we bump our General Preference for getting more than .1 days of work at a time). Do you know what happens? The PCs crunch WUs all day, each of them has SOMETHING completed by day end, and ALL the PCs try to send their results at the same time! Thus if anyone IS trying to use the office network at 7:01PM, they aren't able to get anything done.

Does anyone know of a way to tell BOINC to lighten up? "Hey, look man, we got ALL NIGHT to report these results...let's wait our turn". "Wait our turn" would be a solution where the PCs are coordindated by some scheduler or other signal (and if you're doing all that, then prioritize so the PC with no work left gets on the network first, and the one with a result that's almost past due is right behind him). But short of that, is there a way to somehow Randomize when each PC tries to jump onto the network? So maybe they each begin at some random time between 7:00 and 8:30PM, rather than all jumping on RIGHT AT 7:00?


As with most resources, the demand only increases.
This is especially true of computing resources, particularly network bandwidth (and speed too, but that's a whole other discussion).
Users are always wanting more, and companies are devising ways of delivering larger media over networks, such a audio and video (think YouTube and to a lesser extent Skype; although skype is more affected by delay, and doesn't consume huge amounts of bandwidth anyway, due to using rather efficient audio codecs, but again, that's another discussion, the real point is that skype is barely possible over a 56Kbps connection, firstly due to the bandwidth, secondly due to the huge amount of delay (latency) that traditional modems added to packet transmition - that's yet another discussion).

So as stated, if you're near the capacity limit, something would've pushed you over the edge anyway, it just happens to be BOINC in this case.

One way would be to enforce, or force a staggered start, either by giving each client a window in which it's allowed to use the network, or at the network level by only allowing certain IP blocks access between certain times.
The first is preferable, because you're solving the problem at the source.
The disadvantage is that administration becomes quite time-consuming.
Letting all clients connect between say 00:00 (midnight) and 04:00 (4 am) would probably work well, it's rare that anyone will be using/needing the network (at least not internet access) and the clients would exponentially back-off and sort them selves out. Obviously the window size will need to be adjusted depending on how many clients need to connect, if we're talking hundreds/thousans, then a more sophisticated solution is probably needed.

Advanced co-ordination could be achieved by using the GUI RPC function and issuing commands to groups of clients from a "monitoring" PC (or server).
a simple way would be to allow a block of clients access by directly controlling their network access setting (changing from disabled to automatic) then after a period of time, blocking the group again (automatic back to disabled) and allowing the next group.

For the sophisticated requirements described previously, you'd need to either integrate/co-ordinate with an advanced manager (such as BoincView) or probably write your own. The protocol spec is open-source afterall.

One method of reducing high-load would simply be to limit the bandwidth BOINC is allowed to use in the preferences.

I think something that's been forgotten is that the servers have limited bandwidth too, so even if you had just one machine on a 2 meg WAN (internet) connection, it wouldn't use the full 2 megs anyway.
Also by that logic, even if you had lots of boinc clients, you could probably safely have quite a few of them using the WAN connection together.

Other more advanced things such as QoS (Quality of Service) could be implemented to reduce the priority of boinc traffic, or increase the priority of user traffic (such as HTTP for the web).
This could be done many ways, either by classifying the boinc traffic based on type/format etc. (with something similar to L7 from the linux world), or more simply by destination IP address (eg the BOINC project's servers).
The former would be prefered because it has the least points of failure. A simple example is that IP addresses can change.
3) Message boards : Rosetta@home Science : DISCUSSION of Rosetta@home Journal (3) (Message 40467)
Posted 7 May 2007 by Lee Carre
Post:
I'm starting to see some new work units using a "(beta) strand pairing" search method.
Could someone knowledgeable on the subject of "(beta) strand pairing" elaborate, in layman's terms, what the term means and how it affects the way the search is performed?


Yes--these are my work units which are testing a new idea we have had. I'll give an explanation in my journal tonight or tomorrow.

thanks very much, I look forward to reading it
4) Message boards : Rosetta@home Science : DISCUSSION of Rosetta@home Journal (3) (Message 40464)
Posted 7 May 2007 by Lee Carre
Post:
I'm starting to see some new work units using a "(beta) strand pairing" search method.
Could someone knowledgeable on the subject of "(beta) strand pairing" elaborate, in layman's terms, what the term means and how it affects the way the search is performed?
5) Message boards : Number crunching : Report Problems with Rosetta Version 5.22 (Message 18835)
Posted 17 Jun 2006 by Lee Carre
Post:
http://boinc.bakerlab.org/rosetta/result.php?resultid=24571715

i was viewing the graphics window at the time it failed incase that makes a difference
6) Message boards : Rosetta@home Science : How Rosetta works, an explanation your cat will follow (Message 18757)
Posted 16 Jun 2006 by Lee Carre
Post:
good summary, i vote that this be made a "sticky" and be linked to from the introduction material, it sums things up very nicely :)
7) Message boards : Number crunching : To Forum Moderators & Rosetta Experts...Q & A Forum Orphaned (Message 18481)
Posted 11 Jun 2006 by Lee Carre
Post:
I would be in favor of leaving the current message board topics the way that they are and just getting rid of the Q & A section. Those posts could easily fit in "Number Crunching" area.
indeed, and more people would see them

many people already post problems and issues in the NC forum anyway, on most projects, so it does make sense,

however Janus has made many many improvements to the Q&A forum, making it much easier to use
8) Message boards : Number crunching : To Forum Moderators & Rosetta Experts...Q & A Forum Orphaned (Message 17305)
Posted 29 May 2006 by Lee Carre
Post:
i'm sure you're all aware, but just incase, Janus from BURP is rewriting the whole forum architecture, recoding it properly, and the Q&A is much improved, take a look in the forum testing section on the BURP site
9) Message boards : Number crunching : rendering frames have a performance effect? (Message 17246)
Posted 27 May 2006 by Lee Carre
Post:
hi, i was just wondering, does rendering the protein graphics not eat up a tiny bit of you processing power?
ie. would it be even more effective to set the screensaver to go blank after a minute or so so that no processing power is lost to rendering the protein graphics?
well, on a single CPU/core machine, the graphics will use quite a bit of the available power, reducing the amount available for crunching
so for best speed, don't use the graphics, but if you like to see the screen saver occasionally then sure, set a timeout of a minute or 2, then go to blank screen

(i've got a pc with no screen/mouse/keyboard being controlled with remote admin tool that just runs Boinc/Rosetta all day, cos i don't use it, and was wondering if i could squeeze more out of it by turning of the graphics.....)
in that case yes, choose a blank screen saver or something
10) Message boards : Number crunching : users who run off-line are impacted by shorter deadlines (Message 17245)
Posted 27 May 2006 by Lee Carre
Post:
some further thought on short deadlines

for those that have trouble meeting them, this may have a negative impact on the project and speed at which resutls are obtained, because boinc will resend work, which will take longer, and even if the "late" result is returned, and is useful, you've got at least one other cruncher doing work that it didn't have to, reducing the overall crunching power (consider multiples of this situation)

so increasing the deadline just a small amount, might actually have a positive impact, but i don't know, the only way to really know is to try it, i'm just voicing some lateral thinking ;)
11) Message boards : Cafe Rosetta : Moderator contact thread archive (Message 17203)
Posted 27 May 2006 by Lee Carre
Post:
Q. Which ports does Rosetta@Home need open?


A. Port 80 (http) is used for most communications with the project servers. Port 443 (https) is used if the password is being exchanged . Ports 1043 and 31416 are used for local control as well as for remote control of a BOINC client. 1043 is preferred and 31416 is a fall back. Unless you are RPC'ing over the Internet don't open 1043 and 31416 in your non-local firewall specifically for BOINC.
https isn't used to exchange the password, the password isn't even sent, only a hash of the password (use Ethereal to see for yourself), rosetta using http by default, will have the hash sent on port 80

the newer boinc clients use port 31416 by default now

to communicate with project servers ports shouldn't need to be opened, unless a firewall/router is really dumb, any connection from the client-side should be allowed to pass (and it's returning data), a software firewall might prompt the user to allow boinc to access the internet/manager, but this is a software-level thing, not a port-level thing



i'll update this in the wiki as well :)
12) Message boards : Number crunching : Opening network ports (Message 17202)
Posted 27 May 2006 by Lee Carre
Post:
well, some of the info in the FAQ is wrong lol, https isn't used for password exchange (the password isn't even sent, only a hash of the password, use Ethereal and see for yourself)

unless the firewall is really stupid, or badly configured you shouldn't need to open any ports to connect to the project servers, and you should be prompted by your firewall to allow boinc.exe to listen on port 1043/31416 (depending which version of BOINC you're using)
it might prompt you to allow boinc to access the internet (but this is a program level thing, rather than a port level thing)

all communication to the project is initiated from your client, so no ports need to be open, because BOINC doesn't need to listen for incomming connectons (except from the manager, but that's different)

i know that doesn't solve the problem, but i'm trying to find out why his firewall is behaving badly, or isn't prompting him for access: the root cause, rather than the symptom(s)
13) Message boards : Number crunching : Is this for real??? (Message 17189)
Posted 26 May 2006 by Lee Carre
Post:
Here's 6 of them, this individual is scaming the credit system bigtime!!!!

[sarcasm]wonderful, it's nice to know some people are sad enough to sink that low, attempting to render the whole credit system useless (it is if people can claim what they want)[/sarcasm]

i wish rosetta would implement FlOps counting...
14) Message boards : Number crunching : Firefox Plugin!! (Message 16931)
Posted 23 May 2006 by Lee Carre
Post:
more information can be found on the SETI boards (when they're finished their weekly update, hence why i didn't post a link to the thread of the author)
15) Message boards : Number crunching : difficult target first (Message 16923)
Posted 23 May 2006 by Lee Carre
Post:
...only send one big job per dual core/dual cpu computer if that's possible.

THAT one's going to be tough. BOINC's rules aren't that complicated I don't think. But if they can establish the needed memory "PER CPU", then hopefully that achieves the same objective.

from my understanding of the system, you're correct in that it will be hard if not impossible to make such a specification
i don't even think they can specify RAM per CPU, only RAM per WU, this is yet another area in which BOINC is lacking :(
16) Message boards : Rosetta@home Science : Discussion- Proteins and Work Units (Message 15336)
Posted 2 May 2006 by Lee Carre
Post:
I would guess that the size of a small protein in its native state must be something like 50 atoms across, so the size of a protein would be a few Nanometers or more (I am not a bio science guy either, I am sure the experts will be able to give you more accurate numbers).

a little bigger, you have to take into account the size/length of the bonds (between atoms and between amino acids) as well as the fact that there are quite a few atoms per amino acid, and quite a few aminos per protein
for example a carbon-carbon bond is something like 150-250 nanometres, so an average protein is probably a few thousand nonometres across, however i don't claim to be a true expert, i'm sure one of the project scientists could give a definate answer

there are also probably more than just 50 atoms in even the simplest of proteins
17) Message boards : Number crunching : ABORT THESE JOBS (Message 14874)
Posted 28 Apr 2006 by Lee Carre
Post:
I see that, as requested, a moderator has created a thread for this


We do try ;>)

i just wanted to point it out to those who'd requested it, incase they hadn't seen it already

i must say it's very nice having a highly organised forum, i for one appreciate your efforts, whoever you might be Mr. Moderator :p
18) Message boards : Number crunching : "Recommended Minimum" hardware (Message 14859)
Posted 28 Apr 2006 by Lee Carre
Post:
as for older hardware, i've got a couple of PIII machines running a veriety of projects quite happily, most of the time, all they're doing is crunching, but i do use one for simple stuff like playing music etc. and it still never misses deadlines

PIII 600MHz
512MB RAM

PIII 500MHz
384MB RAM

the 500MHz machine is the one that's used quite often, but only for simple tasks
19) Message boards : Number crunching : ABORT THESE JOBS (Message 14858)
Posted 28 Apr 2006 by Lee Carre
Post:
I see that, as requested, a moderator has created a thread for this
20) Message boards : Number crunching : Does BOINC run on the CPU? (Message 14857)
Posted 28 Apr 2006 by Lee Carre
Post:
If I set my computer to power down my hard drive, BOINC will still run won't it?
Or how does that work? (gotta save every dime when it comes to electricity) :)

you will actually do more damage than good, a running disk doens't use much, but one that's starting and stopping all the time will use quite a lot (it takes a lot of power to get the platters spinning, but not much to keep them spinning)

also by having your hard drive spin up/down constantly you'll shorten it's life, and it'll fail sooner (as in break, and you'll loose any data on it)


Next 20



©2024 University of Washington
https://www.bakerlab.org