DPC Latency issues with Boinc

Message boards : Number crunching : DPC Latency issues with Boinc
Message board moderation

To post messages, you must log in.

AuthorMessage
Mark Lybeck

Send message
Joined: 9 Aug 99
Posts: 245
Credit: 216,677,290
RAC: 173
Finland
Message 1766448 - Posted: 20 Feb 2016, 8:57:11 UTC

I have the following setup
ASRock WS motherboard
Boinc Manager version 7.4.42 (x64)
Running v.44 lunatics installer apps on CUDA
GPU 960.
Nvidia Driver version 347.88
OS Windows 8.0.

DPC Latency Checker V1.3.0 shows flatline of 1000us latency while GPU is loaded. However if i Open Boinc Manager Tasks tab intermittent spikes of 2000-8000us (3000us average) will occur 12 seconds after opening the tasks tab. If I select another tab the spikes will not occur. I suspect this has something to do with Boinc Drawing the task status and calculation on GPU at the same time.
ID: 1766448 · Report as offensive
Profile jason_gee
Volunteer developer
Volunteer tester
Avatar

Send message
Joined: 24 Nov 06
Posts: 7489
Credit: 91,093,184
RAC: 0
Australia
Message 1766451 - Posted: 20 Feb 2016, 9:27:13 UTC - in response to Message 1766448.  

Could be a number of issues. Knowing there's DPC spikes (1000 offset being normal reading on Win8+ with that program), I'd suggest taking a look with the more detailed LatencyMon. I'd suspect you may have a network or chipset driver issue (which should reveal itself more precisely in LatencyMon), since the gui is talking to the client via the gui RPC port, which are effectively through the tcp network stack.

I had to find modified lowlatency drivers for my Atheros chipset based Wifi adaptor, for similar effects, though other devices/drivers could be doing it.
"Living by the wisdom of computer science doesn't sound so bad after all. And unlike most advice, it's backed up by proofs." -- Algorithms to live by: The computer science of human decisions.
ID: 1766451 · Report as offensive
Mark Lybeck

Send message
Joined: 9 Aug 99
Posts: 245
Credit: 216,677,290
RAC: 173
Finland
Message 1766453 - Posted: 20 Feb 2016, 9:37:14 UTC - in response to Message 1766451.  

Here is a brief run:

OS version: Windows 8 , 6.2, build: 9200 (x64)
Hardware: ASRock, Z77 WS
CPU: GenuineIntel Intel(R) Core(TM) i7-3770K CPU @ 3.50GHz
Logical processors: 8
Processor groups: 1
RAM: 8070 MB total


_________________________________________________________________________________________________________
CPU SPEED
_________________________________________________________________________________________________________
Reported CPU speed: 3500,0 MHz
Measured CPU speed: 1084,0 MHz (approx.)

Note: reported execution times may be calculated based on a fixed reported CPU speed. Disable variable speed settings like Intel Speed Step and AMD Cool N Quiet in the BIOS setup for more accurate results.


_________________________________________________________________________________________________________
MEASURED KERNEL TIMER LATENCIES
_________________________________________________________________________________________________________
This value represents the maximum measured latency of a perodically scheduled kernel timer.

Highest measured kernel timer latency (µs): 47453,244533


_________________________________________________________________________________________________________
REPORTED ISRs
_________________________________________________________________________________________________________
Interrupt service routines are routines installed by the OS and device drivers that execute in response to a hardware interrupt signal.

Highest ISR routine execution time (µs): 143991,714857
Driver with highest ISR routine execution time: HECIx64.sys - Intel(R) Management Engine Interface, Intel Corporation

Highest reported total ISR routine time (%): 1,524011
Driver with highest ISR total time: dxgkrnl.sys - DirectX Graphics Kernel, Microsoft Corporation

Total time spent in ISRs (%) 1,857737

ISR count (execution time <250 µs): 1509873
ISR count (execution time 250-500 µs): 0
ISR count (execution time 500-999 µs): 393
ISR count (execution time 1000-1999 µs): 11
ISR count (execution time 2000-3999 µs): 11
ISR count (execution time >=4000 µs): 0


_________________________________________________________________________________________________________
REPORTED DPCs
_________________________________________________________________________________________________________
DPC routines are part of the interrupt servicing dispatch mechanism and disable the possibility for a process to utilize the CPU while it is interrupted until the DPC has finished execution.

Highest DPC routine execution time (µs): 29331,256857
Driver with highest DPC routine execution time: USBPORT.SYS - USB 1.1 & 2.0 Port Driver, Microsoft Corporation

Highest reported total DPC routine time (%): 1,040236
Driver with highest DPC total execution time: dxgkrnl.sys - DirectX Graphics Kernel, Microsoft Corporation

Total time spent in DPCs (%) 1,652977

DPC count (execution time <250 µs): 1024798
DPC count (execution time 250-500 µs): 0
DPC count (execution time 500-999 µs): 3824
DPC count (execution time 1000-1999 µs): 7
DPC count (execution time 2000-3999 µs): 3
DPC count (execution time >=4000 µs): 0
ID: 1766453 · Report as offensive
Profile jason_gee
Volunteer developer
Volunteer tester
Avatar

Send message
Joined: 24 Nov 06
Posts: 7489
Credit: 91,093,184
RAC: 0
Australia
Message 1766456 - Posted: 20 Feb 2016, 10:08:21 UTC - in response to Message 1766453.  
Last modified: 20 Feb 2016, 10:12:31 UTC

Yeah probably this one:
Highest DPC routine execution time (µs): 29331,256857
Driver with highest DPC routine execution time: USBPORT.SYS - USB 1.1 & 2.0 Port Driver, Microsoft Corporation

Or possibly a driver or device (possibly broken) attached to the USB ports(s) ... e.g. a dodgy webcam shared on the same USB controller as something else.

More than likely though, the generic USB drivers migth need chipset or motherboard specific ones. If it's an Intel chipset rather than other brand controller, then there should be somethiong like this in device manager USB controller properties (as opposed to Generic MS ones), my example, Old Intel G33 Chipset:

INTEL(R) ICH9 Family USB Universal Host Controller
, and a similar extended USB2 one etc, using different .sys files that the Generic MS one.

If not, You more than likely need to check all the Chipset devices in System, to see if the controllers are using Generic drivers instead of proper ones, the dates on devices being a giveaway. because USB interacts over PCIe/PCI, the PCI express Chipset drivers probably need manual forcing to new Intel ones as wel, since Intel's Installer doesn;t force the issue if added after OS install. (require Intel CHipset driver .zip pack to force chipset drivers that won't update, individually manually [throuig hdevice manager] as per Intel readme)
"Living by the wisdom of computer science doesn't sound so bad after all. And unlike most advice, it's backed up by proofs." -- Algorithms to live by: The computer science of human decisions.
ID: 1766456 · Report as offensive
Profile jason_gee
Volunteer developer
Volunteer tester
Avatar

Send message
Joined: 24 Nov 06
Posts: 7489
Credit: 91,093,184
RAC: 0
Australia
Message 1766457 - Posted: 20 Feb 2016, 10:17:26 UTC
Last modified: 20 Feb 2016, 10:18:34 UTC

List of system device on my chipset that I had to manually update/force to Intel ones, included:
- All USB controllers, 8 of them
- In system section, Several PCI Express related ones, IO Controller, SMBus and LPC Controllers
- Several Raid storage controller devices (on my chipset)
[Also A PCI bridge, and Firmware Hub device, whatever that does]
"Living by the wisdom of computer science doesn't sound so bad after all. And unlike most advice, it's backed up by proofs." -- Algorithms to live by: The computer science of human decisions.
ID: 1766457 · Report as offensive
Mark Lybeck

Send message
Joined: 9 Aug 99
Posts: 245
Credit: 216,677,290
RAC: 173
Finland
Message 1766458 - Posted: 20 Feb 2016, 10:24:15 UTC - in response to Message 1766457.  

I had a few driver restarts, but they are no good you have to reboot in order to recover correctly.

I recall the motherboard vendor installation package included setup for Windows 7 only. Windows 8 was a clean install from Microsoft.

I just installed intel management engine version 11.
https://downloadcenter.intel.com/search?keyword=intel+management+engine

I have swapped cards back and forth as well as even CPU's.
ID: 1766458 · Report as offensive
Profile jason_gee
Volunteer developer
Volunteer tester
Avatar

Send message
Joined: 24 Nov 06
Posts: 7489
Credit: 91,093,184
RAC: 0
Australia
Message 1766466 - Posted: 20 Feb 2016, 11:19:11 UTC - in response to Message 1766458.  

Yeah definitely will need to (manually) compare all the system device Intel chipset ones against what's in the zip packages from Intel.
(Driver properties should say Intel, rather than Microsoft, and dates newer than Generic Windows 8 release date)

There is a section In Intel's Chipset Driver Readme explaining Install After OS is already installed, which clarifies things better than I could, that the installers generally leaves things alone.

Really odd the Win7 pro retail on one machine in this house put proper Intel chipset drivers by itself, while my main Win7 development PC needed all the forcing. Don't know what determines/triggeres that, but the other one was indeed newer hardware, so perhaps it just depends on default drivers versus specific hardware.
"Living by the wisdom of computer science doesn't sound so bad after all. And unlike most advice, it's backed up by proofs." -- Algorithms to live by: The computer science of human decisions.
ID: 1766466 · Report as offensive
Mark Lybeck

Send message
Joined: 9 Aug 99
Posts: 245
Credit: 216,677,290
RAC: 173
Finland
Message 1766476 - Posted: 20 Feb 2016, 14:15:42 UTC

I noticed that on some occasions if you enable the tasks tab in Boinc, meaning it will update the task status, you can get spikes differing between 2000 - 5000us. They disapear or diminsh in frequency when you select another tab.
ID: 1766476 · Report as offensive
Profile jason_gee
Volunteer developer
Volunteer tester
Avatar

Send message
Joined: 24 Nov 06
Posts: 7489
Credit: 91,093,184
RAC: 0
Australia
Message 1766478 - Posted: 20 Feb 2016, 14:26:40 UTC - in response to Message 1766476.  

Yeah, you're noticing that some compnent use by all of those in different amounts, is stalling. One driver or piece of hardwrae, is failed.
"Living by the wisdom of computer science doesn't sound so bad after all. And unlike most advice, it's backed up by proofs." -- Algorithms to live by: The computer science of human decisions.
ID: 1766478 · Report as offensive

Message boards : Number crunching : DPC Latency issues with Boinc


 
©2024 University of California
 
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.