I was at a UK WLCG meeting this week. One of the sponsors was Dell so they got a slot in the schedule to try to sell us high performance servers... The presentation concentrated on the performance of servers on HEP-type jobs with various operating systems. They noticed a small jump going from Scientific Linux 5 to SL6 -- but a huge leap going to Red Hat 7A (beta apparently) for both Sandy Bridge and Interlagos. So, those of you in a position to test (and recompile under) RH7A should probably do so -- and report back here, if you would be so kind.

---

Distribution itself is not that relevant. It would be more useful to know, is it 3.x kernel (or some options in it) or some of the newer libraries (glibc, etc...) scientific applications use or compiler version/optimization flags they've been compiled with.

---

Distribution itself is not that relevant. It would be more useful to know, is it 3.x kernel (or some options in it) or some of the newer libraries (glibc, etc...) scientific applications use or compiler version/optimization flags they've been compiled with.

Quite. That info is hard to find on the RH site (i.e. I haven't found it yet...).
I'll try to get the presentation slides from Dell.

---

Skildude had mentioned that Suse outperformed Kubuntu by 5% with processor tasks. I mentioned that he should try a lightweight Ubuntu (xubuntu) and compare. I felt that all the bells and whistles in that heavyweight distro could account for 5% less performance vs. Suse.

But what I did not even think to mention in that thread, was that perhaps suse's kernel was compiled/optimized in a different fashion.

And now I see something similar being mentioned about redhat.

If that's the case, I'd really love to know what optimizations can make the kernel a better crunching platform... And maybe some distros are compiling kernels that are in fact superior to others for crunching.

(I'd love to see a Linux box pull off win7 performance with crunching. It makes me sad that it doesn't work that way. (Is this 100% due to the lack of proprietary drivers and software in Linux????) Especially considering you can run Linux without a GUI, I'm surprised that a GUI-less Linux box does not outperform a windows box.)

---

#resist

One important one that we discovered after a long discussion, several years ago, is that the Linux kernel is compiled (or configured) to use the "performance" governor rather than "ondemand". Otherwise, power saving features kick in - great for the environment, lousy for crunching.

Windows vs. Linux

Apologies for going off topic but think this could be important enough.

Just clicked on that link Richard & got this.....

---

Fair warning.

The chances are that it's either (less likely) somebody's avatar image, or (more likely) an external site not under SETI's control, perhaps displayed via somebody's signature.

Two tests:

1) What does 'show details' (as in your screenshot) tell you about the source or nature of the threat?

2) If you turn off signature display in your user preferences (middle set), is the threat still detected?

Sorry Richard, didn't stop to think of those. I'll try again & see if its repeated.

Yes, its repeatable, so going to do as you suggested.

Edit:



Going to retry test with sig turned off.

Edit 2:

Tried again with both sig & avatar turned off, & it still came up with the same threat.

One important one that we discovered after a long discussion, several years ago, is that the Linux kernel is compiled (or configured) to use the "performance" governor rather than "ondemand". Otherwise, power saving features kick in - great for the environment, lousy for crunching.

Windows vs. Linux

Indeed so...

And from that, I thought an option was included in Boinc whereby you could set to run at low priority but not so low that hardware power-saving activated... "nice 5" or so should ensure that (as opposed the usual lower "nice 19" priority). ('Normal' user priority is "nice 0".)

I simply have the power saving turned off :-)


As for crunch performance, by far the biggest influence is whether or not you are running an 'optimised application'. The default Windows compiled version served out by Berkeley has long had a performance advantage over their versions as compiled for Linux. Try instead comparing the Lunatics optimised builds?

Also, the overheads from any OS itself should be very low. More significant will be what other applications might be running that will steal CPU cycles away from Boinc.


Happy fast crunchin',
Martin

---

See new freedom: Mageia Linux
Take a look for yourself: Linux Format
The Future is what We all make IT (GPLv3)

It is pretty much application-dependent. E@H applications for Linux, especially CUDA one, are significantly faster than Windows ones.

Also, I'm crunching majority of S@H work on Linux, including CUDA and am using script which raises the niceness of all CUDA-feeding processes to -10 for max. performance.

---

Tried again, & same threat, so is it possible the threat's on that page?

Sirius,
Funny, I tested with windows 7/mse and I'm not detecting the same issue...

Maybe something you already have is being triggered by something on someones sig on that thread...????

I use Firefox. I'm assuming you must be using internet explorer Sirius?

---

#resist

It is pretty much application-dependent. E@H applications for Linux, especially CUDA one, are significantly faster than Windows ones.

Also, I'm crunching majority of S@H work on Linux, including CUDA and am using script which raises the niceness of all CUDA-feeding processes to -10 for max. performance.

Thanks for that tidbit. That is good to know sir. At least then we know results definitely vary by application.

Now I'd love to know if the different processing speeds of different Linux distros are also random depending on the application, or if for example suse or redhat will typically do 5% faster than kubuntu regardless of application...

---

#resist

No, my rig died last Sat & only just done a fresh clean install with AV showing no issues.

I've done what Richard suggested & still got the threat. but saying that, I'll run a complete check again, just to be safe.

---

check your computer with malwarebytes or something similar. make sure your browsers extensions (add-ons) are up to date, especially java and flash. And perhaps you're getting the warning because someones sig/image/etc on that page links to something bad, and your browsers security settings are set different than mine. I just think it's weird no one else has an issue with that link or thread...

Man some p***** off Mod is really gonna have to clean up after us. Sirius start a new thread in windows q+a for it, we'll discuss it there.

---

#resist

Good point. Have done so.

---

[moved to other thread]

---

see the thread ...something"threat" in questions and answers

---

#resist

So how bout that linux? We need some true linux junkies to get at the facts why some distros outperform others, and whether it's across the board or just on certain apps.

---

#resist

So how bout that linux? We need some true linux junkies to get at the facts why some distros outperform others, and whether it's across the board or just on certain apps.

Me a Linux junkie?... ;-)

For any good OS, the overheads from the OS should be negligible. That's certainly the case for what I've seen, including for Windows and others.

What does make a big difference with what OS you run is what else you have running with that OS, and how well or otherwise the OS manages scheduling and resource use for a particular application... Some OSes are better than others on that count...


The one biggest performance killer is at what point does your system run out of available system RAM to hold all your executables and their data in RAM. Distros with a smaller RAM footprint will give you more spare space in RAM. However, note that 'more spare space' does not give any additional speed-up. It is purely that you need to have 'enough' instead of the 'not enough' causing a progressively ever worse slow-down.

Next in significance I'd rate how well optimised the application is for your hardware. For s@h, check out the latest from the Lunatics.

Next is whether you're running a 32bit system or a 64bit system. For more than 3GBytes system RAM, 64bits wins out nicely. Also, some applications optimise up for very much better performance with 64bits. For example ABC@home gains a x2 performance boost for 64bits over that of 32bits.

Then comes all the various 'fine tuning' tweaks such as process priorities, what processes you have running, 'power saving', and OS clock ticks interrupts rates, HDD layout or SSD use, "RAMdisks" use, caching, and deeper esoterics to squeeze out the last fractions of a percent.

(Aside: The Linux kernel has moved to a "tickless" clock for schedling events. Hence I suspect that the resolution of the system clock doesn't incur a noticeable overhead.)


One brief comment about caching... Too big a Boinc cache and you can lose many minutes of CPU time just for Boinc to manage its own cache of jobs!


Happy fast crunchin',
Martin

---

See new freedom: Mageia Linux
Take a look for yourself: Linux Format
The Future is what We all make IT (GPLv3)