Hey guys,

I'm experiencing a very weird issue. I've had this for a while and have not been able to figure it out until now. I have a Samsung 850 EVO SSD which normally gets 500mb/s read and write speed. But mine gets 200-250mb/s write speed. I noticed that in Safe mode, I get fine speeds which made me think it has to be a program. So after closing all active programs 1 by 1 and running benchmarks, I realized that the culprit is Boinc? Makes no sense. Boinc is completely suspended, not using any CPU or disk according to task manager, yet my write speeds are halved. Anyone have any idea why this could be?

With BOINC https://i.imgur.com/u6Gttty.png

Without BOINC https://i.imgur.com/IKgCChw.png

I did this back and forth and it's consistent. I updated Boinc to latest version too.

Speccy link http://speccy.piriform.com/results/Jf5TDt47qESIt6uLJVFwquy

Not straight into core of this problem perhaps, but does that Samsung SSD have the latest firmware?

How have you set the "Leave non-GPU tasks in memory while suspended" option in Boinc Manager computation settings? How big chunk of the total RAM are the running/suspended tasks using in case they are in memory?

Not straight into core of this problem perhaps, but does that Samsung SSD have the latest firmware?

How have you set the "Leave non-GPU tasks in memory while suspended" option in Boinc Manager computation settings? How big chunk of the total RAM are the running/suspended tasks using in case they are in memory?

Hi,
I do have latest firmware yes. It does seem that fully uninstalling Boinc (and removing all related folders) and reinstalling has fixed the issue. No idea why. I had most settings default, and I definitely didn't touch that setting you mentioned, which is off. As you can see in the screenshot I posted, Boinc was using no RAM at all.

Not straight into core of this problem perhaps, but does that Samsung SSD have the latest firmware?

How have you set the "Leave non-GPU tasks in memory while suspended" option in Boinc Manager computation settings? How big chunk of the total RAM are the running/suspended tasks using in case they are in memory?

Hi,
I do have latest firmware yes. It does seem that fully uninstalling Boinc (and removing all related folders) and reinstalling has fixed the issue. No idea why. I had most settings default, and I definitely didn't touch that setting you mentioned, which is off. As you can see in the screenshot I posted, Boinc was using no RAM at all.

That's boincmge.exe, not boinc.exe and boinctray.exe and not the science tasks which may include virtual machine(s). Sometimes the VM's themselves won't stop even though the task they are running stops. I have also seen a science app, forget which project, that refused to stop because the author did not include a checkpoint method so the task could be stopped and restarted. It ran to end end even if the manager was quit.

The "Leave non-GPU tasks in memory while suspended" is important. Swapped onto disk the task or VM can't run. Also if in RAM that can mean the system can't find enough free buffer space to buffer the I/O to disk and has to fall back onto a slower method.

If it comes back you can look into Performance Monitor and click the disk tab, sort by i/o to see what process is using the disk.

Not straight into core of this problem perhaps, but does that Samsung SSD have the latest firmware?

How have you set the "Leave non-GPU tasks in memory while suspended" option in Boinc Manager computation settings? How big chunk of the total RAM are the running/suspended tasks using in case they are in memory?

Hi,
I do have latest firmware yes. It does seem that fully uninstalling Boinc (and removing all related folders) and reinstalling has fixed the issue. No idea why. I had most settings default, and I definitely didn't touch that setting you mentioned, which is off. As you can see in the screenshot I posted, Boinc was using no RAM at all.

That's boincmge.exe, not boinc.exe and boinctray.exe and not the science tasks which may include virtual machine(s). Sometimes the VM's themselves won't stop even though the task they are running stops. I have also seen a science app, forget which project, that refused to stop because the author did not include a checkpoint method so the task could be stopped and restarted. It ran to end end even if the manager was quit.

The "Leave non-GPU tasks in memory while suspended" is important. Swapped onto disk the task or VM can't run. Also if in RAM that can mean the system can't find enough free buffer space to buffer the I/O to disk and has to fall back onto a slower method.

If it comes back you can look into Performance Monitor and click the disk tab, sort by i/o to see what process is using the disk.

Thanks. No other tasks were using any memory, and I did check disk tab. No disk usage at all. This issue would occur after a fresh reboot so Boinc didn't even get a chance to run yet because i keep it suspended when idle. Very weird issue :/

Not sure if this applies to you, but I was able to use the location property to move ProgramData\BOINC to my large "D" drive. I have only a 256gb SSD but a 2TB D drive. This put all the data writes to the D.

I did not use LOCATION to move the executables. Instead I installed boinc to the D first and then moved the boinc data afterwards.

Possibly the problem is the swap or paging file used for virtual memory. I dont think that LOCATION can be used to move that file. You might try increasing the amount of memory to %90.
I show 8GB of pagefile and 16MB of swapfile on my SSD. I allow %90 of memory for BOINC and %75 of the page file. I am not sure what the swapfile is good for. I normally see only the hibernationo file (13gb) and the pagefile (8gb). With

02/13/2018 11:41 PM 13,693,136,896 hiberfil.sys
02/15/2018 09:23 AM 8,623,251,456 pagefile.sys
02/13/2018 11:10 PM <DIR> ProgramData
02/13/2018 11:41 PM 16,777,216 swapfile.sys

Looking at "properties" for a single LHC vbox64 work unit one sees the following
Virtual memory size: 85mb
Working set size: 2400mb
slot: 12

Directory of D:\ProgramData\Boinc\slots\12
...
02/15/2018 07:40 AM 3,248,488,448 vm_image.vdi

That virtual image feeds the working set of pages. This all happens on your SSD unless you can move the data to another disk drive.

just my 2c and I am no expert of m$oft internals, but it could be that suspending BOINC leaves fragmented pages in memory and normal writes to/from the ssd are causing page faults. After a while one would think it would clear up.

[EDIT] your 512mb speed may be sequential and random access is probably lower. This is another guess of course.