What are the Pros & Cons of Software RAID vs Hardware RAID of 4xSSD(512Gig) in RAID10

Just recently I read an article by one of Godaddy's engineers about this very topic: Learning to Deal with Learning

On their hardware (Dell PERC cards) battery learning cycle happens every 90 days, but no way to know when exactly it'll happen, ie during peak or off-peak hours.

They talked about different solutions:

• Outright disable Battery Learning. Problem with this option is that you won't know the status of your battery and how long and how much it can hold charge, so in the case of outage you can risk data loss.

• Use different hardware. Some controllers have 2 batteries and flip between them during such learning cycles. Additionally, there are RAID controllers(such as Dell H710) that do not need batteries but instead use non-volatile NVRAM to store uncommitted data.

• Force write-back(caching) regardless of the status of your batteries. Like the 1st solution, you are risking data loss.

Ultimately, they setup crons for off-peak hours that monitor for the next learn cycle and if it is within the next 24 hours, they force it to happen immediately. That way they keep the benefit of exercising batteries yet without running it at peak-usage times.

I think TRIM is overrated for the most part. Yes, it can help, if you have a drive that has a heavy random workload (which is mostly the case when you're doing drive benchmarks, or for some other real-world workloads), however most people just don't have those workloads.

Here's an interesting poster from SOSP 11, which was looking at a 'selective TRIM' mechanism but also shows a clear comparison between TRIM on and off. TRIM can have a significant negative impact on your drive performance, and while it does show significant benefit in the long-running postmark test, if this usage pattern doesn't match yours then you can't compare them.

Anandtech covered the Samsung 830, and did look at the effect of TRIM here, however I think you should take most of the tests like the ones mentioned there with a grain of salt. Your workload is very unlikely to be 'torture the drive in the worst possible way for a set period of time and then do my normal job', and so the actual impact of your workload on the drives performance is much harder to understand, as is the benefit of TRIM.

Finally, bear in mind that these SSDs can sustain somewhere north of two orders of magnitude more random IOPS than a spinning drive. Even if you lose half their random performance to TRIM, you still have a system that is fantastically faster than spinning drives.

Personally, I'd ignore the lack of TRIM and go with the hardware RAID controller. I'd also manually overprovision a bit more space on the drive (partition off a bit of space and leave it unused) as overprovisioning helps both workload and endurance. And if it turns out that your workload is a very heavy random-write workload, to be honest, I'd go for better drives anyway :)