Can I get some opinions on this -- CCSIO Benchmark?

[apankrat]

Hi fellas,

I'd like to get your esteemed opinions on my little Xmas side project - https://ccsiobench.com

CCSIO stands for Cold-Cache Sequential IO

It's a little benchmark that tries to find the best way read/write large files at a given location, be it a local volume, a virtual mount or a remote share

You basically tell it "C:\Foo" and it tells you "Max read speed is 2144.3 MBps if using 8 x 2MB buffers in direct IO mode."

The way it works is that it goes through a list of (io-buffer-size, io-buffer-count, io-mode) combinations, measures bulk throughput for each and tallies up the results.

Homepage has exact benchmark description and there's a longer, more technical description over at Fundamentals of Fast Bulk IO page.

The back story of this is that I've been building myself a machine and considered various ways to settle in - with a VM, with a portable USB drive, with a TrueCrypt container, etc. - and, being a mature adult that I am, wanted to check what performance penalties I'd be looking at for each option. I found some benchmarking apps (CrystalDiskMark, HDTune Pro, etc), but all of them required manually changing the buffer size/count between the runs and none of them allowed testing with different IO modes. So as per usual I just did a teenage eye-roll and wrote what I wanted. You know the drill.

[apankrat]

Evidently, I can't

[skwire]

It looks great and I'd be happy to run it on my gear for you.  =]

[apankrat]

I'd very much love that, thanks.

Give me a day. I'm adding a screenshot sharing option.

[apankrat]

> Give me a day.

That worked out to be a pretty long day.

1. Added an option for viewing a complete text report and saving it. Will be adding an option of loading and viewing these next.

2. Added an option for Quick and Extended tests. Press Alt to show the top menu.



3. Added an option for sharing screenshots.



The output is an Imgur-style page with just the image, e.g. https://ccsiobench.com/s/mpFja.png

If you got some interesting devices, let's compare some screenshots, shall we?

[Shades]

While using a different computer in my LAN to re-encode files on my work system I thought to look at your software. While there isn't something inside that would generate spectacular results (Samsung EVO 850 SSD 120GByte in combination with a bog standard Seagate Baracuda 1 TByte on a Core2Duo PC of at least 8 years old running Windows Server 2012 R2), I just thought to have a peek.

The files to encode were stored on the 1 TByte disk, so I started the test on the SSD. Shouldn't be too big of a deal as the PC with files wasn't doing anything else (no tasks in the background, no websites open...it wasn't doing anything else than serving files from a different disk than the one being tested. After looking at the test results, I wanted to check (on the LAN PC itself) how far along the re-encoding progressed. Everything failed during/after the test. Share settings didn't change, ACL on files/folder didn't change, yet it wasn't possible to write anything anymore.

Must say that the software looks great and is responsive. The rest of the system remained responsive too during the time it took to complete the test. The version I tested doesn't have any way of storing the test results, making it little more than a informative exercise. With little to say, I didn't take the time to respond and kinda forgot about this thread   

Now I see screenshots showing me that there are options to store generated results, which is great. That makes this tool much more useful to keep track of performance over time. Or in different systems. Or both.

Still, that doesn't remedy the fact regarding the boring results my system is capable of generating, so I won't bore you with those.

[Deozaan]

Screenshots don't seem to work very well on my Windows 10 system:



Here's what the window actually looks like for me:

[apankrat]

Screenshots don't seem to work very well on my Windows 10 system:

Whoops, thanks. It was an issue with bmp-to-png conversion on the server side. Fixed now.

That 10x write speed boost in the buffered mode though... W10 is being very lazy with cache flushing. I'll add a UI option for force-flushing write cache after each test and an option to run each test longer. I know that the former brings w. performance right down to the unbuffered / read levels.

[apankrat]

Everything failed during/after the test. Share settings didn't change, ACL on files/folder didn't change, yet it wasn't possible to write anything anymore.

Interesting. During the test the IO is fully saturated, but the system should recover immediately after it. Also, testing shouldn't trash existing IO cache, at least in theory. When it uses cached reading/writing, it specifies so-called "sequential" access, which should, according to Microsoft, cause Cache Manager to aggressively recycle cache pages for the file in question. It should still do some caching, but as the app moves to from one file chunk to the next, Cache Manager should reuse the cache page for the latter.

Must say that the software looks great and is responsive.

Thanks, very kind :)

Still, that doesn't remedy the fact regarding the boring results my system is capable of generating, so I won't bore you with those.

I think it's still be interesting to look at what an older drive IO profile is like.
I have a couple of ancient refurbished SAS drives with lots of mileage on them, will be torturing them a bit next week.