Performance Issues After SSD RAID Refresh

Wow- did not realize that the Mushkin had that big of a problem with performance. Even so, should its performance be so degraded that 5 year old OCZ bottom-of-the-barrel-for-2012 drives can beat it? When I got this set, I had just assumed that even the weakest drives today were better than the budget drives of yesteryear- and the budget drives I had were plenty for my use. Also, my write numbers are wayyyy weaker than the linked review in all cases- ranging from 1/3 performance to 1/15 performance. Maybe there is something else going on here?

^ Excellent overview of a lot of stuff. I did not want to include all of the details of why I went the direction I did so that the question was not so verbose, but since we are looking at alternative solutions I will elaborate a bit. Right now, I have my OS on the SSD array. I also have project source code. Some of that project source gets uploaded to Git periodically. I also store a few repos for private projects on the array without backing up to Git (private repos on Git cost $, so unless a business is footing the bill I tend to avoid). In addition, I keep various data sets that are generally subsets of larger data sets (or other dev data sets of varying sizes) on the array when in use (scripts move data in/out depending on the project I am working on).

Is redundancy important here? For the data, generally not so much. For the source code, I really do not want to lose any newly written sections if at all possible. Because backups are periodic (I do nightly backups in addition to this to a HDD; the HDD also has a bootable image in case something like the RAID card gets corrupted) they do not necessarily capture this marginal source code, hence the RAID array. In the past, it has come in handy. When I first did the OCZ array, one of the original SSD sets died very soon after use. I just stuck my spare in and got the other drive RMA'd. Except for a ~1/3 loss of performance for a few hours, all was fine and I did not lose uptime.

Given all of this, real-time redundancy is more a less a hard requirement of mine. RAID 10 does this well and is common in real-world database applications. RAID 5 or 6 tends to work well also, but used to need more sophisticated RAID cards to do so. Not sure if that is the case anymore, but I did not want to upgrade my RAID if not necessary. But, RAID is done here primarily for redundancy and not for performance- though I would think SSD RAID would improve performance for things like the various SQL and NoSQL DMBS's I use in development as well as the manual parallel file system accesses some of my applications perform. A single SSD is sufficient to get the performance numbers I need to do testing. That said, RAID corrupting performance to the extent happening here is not really acceptable.

I had looked at NVMe solutions as well, but my understanding is that there might be some compatibility problems with my platform. IIRC, the X79 has weird PCIE 2.0/3.0 issues, as Intel was just making the transition when produced. Additionally, I am working on a Windows 7 OS (sometimes in a Linux distro in a VM) and I thought there may be some problems with Windows 7 and NVMe. As a final nail in the NVMe coffin for my particular application, I could not find a lot of info about running these types of drives in RAID. It looks like there may be some (perhaps very expensive) NVMe cards with RAID built in somehow, and of course there is software RAID (yuck).

With regard to the SSD filling issue, does this refer to TRIM? I do not know too much about this problem, but when I built the old OCZ RAID array I was worried about this. After my numbers for that initial array were pretty good, I figured that this was not a problem (or, perhaps, was not a problem for this particular RAID controller?). I was also operating on the assumption that the TRIM issues were being fixed back in 2012 and that these days there was not much problem left. Since it has been suggested that this aspect may be problematic, is it feasible that this could lead to performance degradation to an order of 1/15? I did run tests before and after cloning the OS onto the new array and did not notice significantly different numbers, but maybe something done on RAID initialization congested the drive?

When we get down to it, what particularly confuses me about this is just that the old OCZ drives (I looked at the exact model, they are OCZ Agility 3 series 120 gb) perform so much better than these brand-new drives. Looking at various posts elsewhere about this RAID controller and SSDs, it appears that this controller may just be picky about what drives in particular are being used. Apparently, it has issues with write parallelism, but again it seems strange that these issues may just hit the Mushkin drives and not the OCZ drives. Maybe the old OCZ drives have a more sophisticated controller than the new Mushkin drives?