Windows VM file copy strange pattern

[tiwing]

Hi, this isn't a problem really, just more of a question why...

 

I am running a windows 10 VM on unraid 6.1.9. w10 VM has 4GB Ram and 2 cores assigned (3.06Ghz Xeon). Whenever I copy large files from the cache to a share, the pattern is fast for a while, then slow, then fast... etc. as you can see from the attached. "Fast" in this case is ~180-200 MB/s. "Slow" is 2-5 MB/s. Cache pool is twin Samsung 250GB SSDs. I have cache drive enabled for my shares and the mover runs daily - there's lots of room on the cache pool. The processors are never pinned at 100% during the process, and memory usage is between 2 and 3 GB and doesn't fluctuation when copying the files.

 

I'm trying to figure out if Windows is doing something goofy where it can't handle reading and writing to an SSD or if it's something with unraid. Thoughts?

(sorry not sure why 2 of the same image got attached)

 

Cheers.

[tiwing]

thoughts? Any thoughts? Cheers...

[testdasi]

I think it's the buffer.

[tiwing]

do you mean memory buffer, on the unraid side (write to SSD), or the VM side? Any way around it that you know of?

[monk100]

I am also running into the exact same issue as you tiwing.  I am trying to use 2x 256GB Crucial M500's and running 2 Win10 VM's. (Initially on 6.1.9, now testing with 6.2.0 b21)

I have configured them in various different ways.  1parity/1drive or 2drives or 1drive/1cache.  I have the same results.  Transfer speeds are fast and then cut right out, up/down/up/down.

I have swapped sata ports on my motherboard, no change.

Tried 1 Crucial as a cache drive, 1 Western Digital Black as a main drive.  Still get the wonky transfer speeds.

 

My only luck so far has been removing the 2 Crucial drives and putting in a 128GB Liteon SSD.  I get constant r/w speeds...speeds you'd expect running 2 vm's off 1 ssd.

 

I really wish I had read your post earlier... I just purchased 2x Samsung 850 EVO's to test. 

 

Please update your post if you find a solution. 

 

 

 

[testdasi]

Try this and see if it helps. Both IPv4 and IPv6.

 

[Frank1940]

thoughts? Any thoughts? Cheers...

 

What is happening is that any unused memory that you have allotted to unRAID is used for caching.  Where most people see the  result of this is on writing files of large files directly to the array.  When the transfer starts, the array is set to receive the file and the actual transfer begins.  The transfer is actually to the cache and then onto the array.  Writing to RAM is limited only by the speed of the Ethernet connection so it is really fast.  However, when the cache fills up, the cache can only accept data as fast as the data can be written to the drives.  So the effect that you see is a fast writes at the beginning and then slowly to the speed of the physical drives when the cache is full.  What happens at the end seems to depend on how the copy process is monitoring what is happening.  Sometimes the copy is declared completed when the last data leaves the originating computer and other times when the write operation is completed on the other end. 

[monk100]

My new Samsung Evo's shared the same strange copy pattern. 

 

I tested out testdasi's fix.  Disabling the "Large Send Offload " for IPv4 & v6  and it solved my issues.

 

Here is a link to his solution if you want to read more about it.

http://lime-technology.com/forum/index.php?topic=48563.msg466225#msg466225

 

Thank you testdasi.