• Some users have recently had their accounts hijacked. It seems that the now defunct EVGA forums might have compromised your password there and seems many are using the same PW here. We would suggest you UPDATE YOUR PASSWORD and TURN ON 2FA for your account here to further secure it. None of the compromised accounts had 2FA turned on.
    Once you have enabled 2FA, your account will be updated soon to show a badge, letting other members know that you use 2FA to protect your account. This should be beneficial for everyone that uses FSFT.

SSD Read/Write simultaenously

M76

M76

[H]F Junkie
Joined
Jun 12, 2012
Messages
15,287
Do you know about any reviews out there where they test ssd performance in scenarios where writing and reading is done at the same time?

It's a very common occurence in every day use that the system writes and reads from the drive at the same time, yet I only remember seeing tests where they run read/write tests completely separately.

I purchased a number of 840 evo SSDs and their performance in real world application where writes and reads are simultaenous is appaling. About 5x slower than the 830 ssds they replaced, and about 10x slower than the 850 PRO.

I wonder if the 850 EVO will show the same performance issues in this scenario that's why I'm looking for reviews.

I'm running multi threaded processing on 12 threads where each thread is reading and writing it's own data at the same time. (about 300GB written for every 100GB read) and when I run this on 840 EVO SSDs the performance is comparable to regular HDDs.
 
Most likely a result of the terrible controller on the EVO. I don't think its capable of switching to read and write very fast or even capable of fully doing both in sequence.

As for reviews, the only site that does a lot of testing on SSD's is Anand, and their torture test does a lot, although I'm unsure if it has a sequence to write/read at once. This is a good metric that I haven't thought much about in the past. I have a bunch of 850 Pro's in my machines now and a few cheap'o SSD's in other machines I don't care much about so it never affected me.
 
I think that because it's such an unusual workload, and so hard to replicate (too many variables involved, proportions of reads and writes can be different), there wouldn't be anything of the sort.

It is strange that the performance is comparable to regular hard drives. That makes me think that the bottleneck of whatever you're doing isn't the storage, it's something else. Possibly RAM, or CPU.
 
Tsumi said:
I think that because it's such an unusual workload, and so hard to replicate (too many variables involved, proportions of reads and writes can be different), there wouldn't be anything of the sort.

It is strange that the performance is comparable to regular hard drives. That makes me think that the bottleneck of whatever you're doing isn't the storage, it's something else. Possibly RAM, or CPU.
Click to expand...

I'm not looking for the exact same workload in tests, only to have them run read/write at the same time to see the performance hit.

No the bottleneck is a hundred percent the drives. I used to have 512GB 830 ssd in the same computer, now I have 1TB 840 EVO, and the peformance hit is enormous. With the 830 the CPU occupancy was over 70% all the time, now it barely hits 20% from time to time.
 
M76 said:
I'm not looking for the exact same workload in tests, only to have them run read/write at the same time to see the performance hit.

No the bottleneck is a hundred percent the drives. I used to have 512GB 830 ssd in the same computer, now I have 1TB 840 EVO, and the peformance hit is enormous. With the 830 the CPU occupancy was over 70% all the time, now it barely hits 20% from time to time.
Click to expand...

Even that wouldn't show likely anything, because a consumer level workload of a couple simultaneous read/writes (say duplicating some large files) is easily handled. They'd really have to crank the number of threads to start finding impacts, and then it's such an edge case not many people would find it useful.

You're probably running into TurboWrite cache slowdowns. It's a couple GB SLC cache on the EVO drives that is used for all writes, which are then written out to the TLC nand during idle times. When the cache fills, performance starts to suffer. 12 threads of 100s of GBs of writes will easily do that. Interleaving reads at the same time is going to really kill it, since that kills the inherit parallelism benefits you get with SSDs when doing sustained reads or writes. It's also going to completely break the readahead caching.

In the end, what you're seeing is pretty muchy why there's a big price difference between say the 850 PRO and 850 EVO. This is where they can save a lot of costs for benefits that a small percentage of people would be able to take advantage of

Is it your own code you're running? If so, you'd probably benefit from looking at what you can do to buffer the read/writes to reduce the interleaving. We've had the same issues before here at work when we try to bump up the threads on various code that does a ton of image processing. Wasn't always the most intuitive things that fixed it, either, really depended on the details of the workload. Often, breaking out read/write code into their own single threads often made a big improvement.

Though a simple fix is to just throw another SSD at it and read from one and write to the other. $100 for another SSD is a lot cheaper than having a coder spend ages tweaking code. :)P
 
Thanks,

No it's not our own code, it's third party software. Setting up the software to read from one disk and write to another is really tricky because it's a two tier process.

Process A reads file A and writes file B then Process B takes over and reads file B to produce file C. So some of the reads/writes would still be on a single drive no matter what I do. Unless I run the two processess apart from each other.

But setting this up makes a one click process into a much more complicated process where data can easily be mixed up accidentally. But it could be a workaround for the 840 drives until I can get rid of them.
 
most good review sites like tweaktown and anandtech do that. They do 80/20 and 50/50 read write tests.

This is the 850 Pro with other drives in chart
http://www.tweaktown.com/reviews/6714/samsung-850-pro-256gb-ssd-review/index5.html

Also your work load is huge for a cheap consumer drive. You should not be using that drive for your needs. Sandisk extreme Pro is a great drive but one test of mix load shows it being very poor. It was a single test though but worth researching. Sandisk extreme pro or samsung 850 pro would be best

Whats up with all these opinions when MULTIPLE review sites have these figures yet no one before me posted a link?
 
Last edited:
You must log in or register to reply here.
Back
Top