Performance issues with NAS, and AV, etc

[adiamond]

Below is my post to the NAS manufacturer (Netgear, Model ReadyNAS 104).  This issue seems to concern the interactions of multiple products so I'm obliged to post on all the forums.

===

I have 4 drives in a Raid 5 config. I have been encountering very slow speeds with smaller files.  So, I conducted a test using a directory of 164 small files  totalling about 200KB [sic].    The tests just copies this directory from my NAS to a place on my SSD.  The target PC is a fast windows 10 machine. I wanted to see if Avast AV and Malwarebytes (MB) were effecting the time to copy the dir.  Here are some results:

 

1. AV and MB - about 90 seconds
2. AV and MB up but all real time protections turned off  - 67 seconds
3. AV, No MB (MB actually closed/exited) - 4.5 seconds
4. No AV (disabled),  No MB (closed) - 1.5 seconds
5. No AV (disabled)  and MB (all on) - 17 seconds 

So this tells us that

1. Without AV or MB (shut down completely) - the baseine speed is 1.5 seconds.
2. Just MB is 17 seconds, > 10x longer than the baseline - MB gives it the biggest hit and oddly that hit is mostly still there if its up but real time protections are disabled
3. Just AV is 4.5 seconds - 3x longer than baseline
4. MV+AV is 90 seconds - 20x longer than baseline. The speed is closer to a multiplicative effect of the MB and AV hits versus addative.

At this pont, I imagine a person reading this is thinkin, "Uh huh, what's that have to do with  a NAS?"  Here are the corresponding times copied from a local standard HD

1. AV and MB - about 0.3 seconds
2. not done
3. AV, No MB (MB actually closed/exited) - 0.17 seconds
4. No AV (disabled),  No MB (closed) - 0.17 seconds
5. No AV (disabled)  and MB (all on) - 0.3 seconds 

So when a local HD is the source ( the NAS isn't the source):

1. When AV and MB are on it's 300x faster than with the NAS - this is the big deal.
2. The effect of the AV is negligable (vs with the NAS it made the copy 3 times slower)
3. without either, the basline is 5x faster without the NAS (which, given the small file sizes might be OK relatively and probably not that big a deal in absolute times (an acceptable time waiting for tasks)

 

Obviously, I'll ping the MB and Avast but as I'm using free Avast that's unlikley plus this 3 product issue is a recipe for finger pointing.  MB is obviouly the bigger deal but why it's so lethal with AV when the NAS in particular is involved is the big bugaboo.  

 

EDIT

One more interesting item:

When I do a Properties on the copied directory on the SSD vs the NAS, they both report 200K for the files but the SSD rports 420KB actually occupied on disk while the NAS reports 10.2MB!  That's seems a bit much of a disparity.  These files are about 1K actual each which for the NAS comes out to be each 1K file taking about 64K (well, 62K) of disk space.

 

[adiamond]

FYI, the test still takes about 90 seconds even if I exclude the NAS from MB, or should I say that root dir of the mapped drive letter that's mapped to a directory of the NAS that includes my test directory?!  

So, given the terrible test results for MB w/r to the NAS even when the shields are turned off,  the only way for me to get decent NAS performance on small files is to exit MB completely!

[dcollins]

Out of curiosity, what are you using to copy files? Just windows explorer?

[adiamond]

No.  Oddly enough a copy routine in matlab (an engineering algorithm dev. program/IDE).  I imagine it's compiled C or C++ calling the Windows API.  If it really matters, I could watch it in procmon and see what it actually is doing but in the end its results seem to track the relative times of how other apps that access a lot of files run (w/r to the AV, MB).  In fact, the reason I started doing this investigation was that Matlab was unusable off my NAS for the same reasons and then I started evaluating a cloud backup program crashplan which also got murdered by AV and MB when backup up off the NAS.

Edited May 26, 2017 by adiamond

[dcollins]

We've seen some situations like this. Does explorer have any issues?

[adiamond]

I haven't forgotten about this.  I replaced a slower drive in my NAS and it takes days to resync. 

[adiamond]

I haven't forgotten about this.  I replaced a slower drive in my NAS and it takes days to resync. 

 

So, I replaced a slow drive in my nas with a faster one and updated its firmware.  

 

Updated firmware

New fast HD to replace 5400RPM WD in NAS

Y:\work\* added to avast excluisions (It was already excluded.  Not sure if this was before or after the last timings)

Updated MB

MB exclusions Y:\ is excluded (It was already excluded.  Not sure if this was before or after the last timings)

 

 

>> mkdir('C:\matlabtest\git00'); tic; copyfile('Y:\work\.git\objects\00','C:\matlabtest\git00'); toc - this is from inside the matlab command prompt. copyfile is a builtin library routine (not written in matlab) but I don't know what it calls.

Warning: Directory already exists.

Elapsed time is 4.619630 seconds. - AV on, no MB

 

>> mkdir('C:\matlabtest\git00'); tic; copyfile('Y:\work\.git\objects\00','C:\matlabtest\git00'); toc

Warning: Directory already exists.

Elapsed time is 66.644988 seconds. - AV+MB - so here we see that MB really hammers this.

Same after all MB shields turned off! - and it doesn't matter if you turn all the shields off.

 

>> tic; system('copy Y:\work\.git\objects\00 C:\matlabtest\git00doscopy\'); toc -from within matlab, I'm launching a "dos/cmd" process to do the copying.

Elapsed time is 27.997095 seconds. AV+MB - much faster then with the copyfile routine.  Go figure.  Still, much slow with MB

About 23 seconds drag and dropping using windows explorer (AV+MB) - timing done by watching a clock.  No accurate but roughly though a bit faster than using "copy" it's much slower than anything without MB

 

-- With no MB the dragging and dropping the dir from within windows explorer was too fast time manually time (by the time I pushed the start button on the timer, after dropping the source folder into the destination, the copying had already finished. - without MB