Hi there,

 

I'm having a weird problem, before I got a cache drive I was getting about 32MB/s write.

 

I added a cache drive (Samsung 840 SSD) but when writing to a share, I only get about 40-43MB/s. I know the cache drive is being utilised because I can browse the files on the cache drive. I have also made sure to select the option to use cache disk under each share.

 

If I export the cache drive and write to it directly I get around 88MB/s.

 

What gives?

Interesting.  The 88MB/s is reasonable (a bit slower than I'd expect with an SSD and Gb network, but nevertheless not bad).

 

... and the 40-43MB/s is clearly going to the cache, as it's faster than you can write to the protected array; but this SHOULD be closer to the 88MB/s number you're seeing with direct writes to the cache.    Do you have applications using the cache drive in the background?    How much memory in your UnRAID system? 

Hi GaryCase, I've seen the SSD peak at closer to 100MB/s, but 88-90 seems closer to the average.

 

No, I don't think so. There is only really my HTPC and my desktop connected and only one is switched on at a time. My server has 4GB of RAM.

I can't really think of anything that would account for the slow cache writes.  It's clearly working, since your writes ARE faster than direct-to-the-protected-array writes;  but is also not providing near the benefit it should be.

 

Is this a new SSD?    If not, was it previously not installed in a system that supported TRIM?

 

You might want to download the Samsung "Magician" software to confirm your drive is performance-optimized;  and perhaps that it has the latest firmware (available on the same download page).

 

http://www.samsung.com/global/business/semiconductor/samsungssd/downloads.html

 

You'll have to connect the drive to a Windows PC to run these tools ... but the result may very well be worth it.

 

 

What is the CPU and MB in the system?

Wow, I completely forgot about this thread until I decided I was going to look into it again, so I thought I'd update it with where I'm at.

 

I'm still having the same problem, all though I've found something even more weird.

 

Copying from a single PC direct to the cache drive yields 85-95MB/s

Copying from a single PC to share without cache yields around 30MB/s

Copying from a single PC to a share with cache enabled yields around 42MB/s

 

(Here is the strange part)

 

Copying from multiple PCs to share with cache enabled yields around 35MB/s per PC as shown here:

 

 

These speeds were all tested with m2ts files of around 20-25GB.

 

Unraid version is 5.0-rc8a. System is an e350 + 4GB RAM.

I've noticed with my three unRAID setups I get very fast writes to the cache drive when writng directly to it. I'm using 1TB cache drives in my three unRAID setups and writing directly to the cache drive I get between 70MB/s to 100MB/s writes depending on my unRAId setup. But when writing to a share that uses the cache drive those speeds drop to between 50MB/s to 65MB/s depending on my unRAID setup.

 

Originally I thought I was having issues with my N54L unRAID but then I relaized that all my unRAID setups are slower in the same situation. Although my N54L still gives me my slowest transfer rates for some reason.

The interesting thing about these results is that it removes networking components as an issue.  Normally when you're seeing slower-than-expected networking speeds, it's either a defective switch port; a bad pair on the Cat-5/6 cable; or an improperly tuned set of networking parameters.  But since you're getting good speeds across the network with direct-to-cache writes, the network is clearly not the problem.

 

The next thing I'd suggest checking is that your shares are set to use the cache ... but in both cases you've already noted that they are !!

 

Next question is WHAT you're copying in each case.  Try copying the SAME file from the SAME source to both the cache directly and a cache-enabled share, to confirm this isn't a case where the source isn't able to "feed" the pipe at the faster speed.

 

For me, I always use the same test folder when testing copying. I have a small test folder with around 7GB or 8GB of files.

 

Otherwise i will just use a BD ISO image for a large file to test with.