Intel ICH software raid can do trim, but the problem is the spec doesn't allow NCQ for TRIM. Plus you are burdening the controller with the trim, so what the industry has been doing is what is called overprovision and let the ssd handle its own GC.
example:
Intel 710 - exact same as intel 320 but with 160 to 180gb usable (320gb of flash). They also throttle down the write speeds to increase longevity.
Seagate MLC - 30% over provision - increases lifespan by 2-4x and steady state performance is legit.
Samsung 840 Pro - unstable until you OP 30% - logical right? More you set aside, the better lifespan and doesn't stall/hang.
Newcomer to the market: Intel S3500 7% OP - throttled write speeds but you get more bang for buck - no idea how it does it but is would be my #1 choice now given the track record of intel prodcut.
Raid-5 is a bad idea for SSD since it is a writing to each sad for every byte change.
Raid-10 is great for linear performance but sucks with random i/o - Better to use RAID-1 for redundancy and span at the o/s level. For people that care about random i/o.
So don't worry about trim. Sata 3.1 (draft?) just introduced NCQ for trim and only a few o/s can handle raid SCSI unmap.
For now, that leaves you with intel rst/ich raid with raid-0 or raid-1 software raid which imo is not the worst way to go. SATA to SATA is the best way. SAS controllers to sata are not so hot. Espcially with sas expanders.
SAS SSD's bring a whole new game with dual ported, multi-target, PI (protection information) PI for hard and ssd drives is advanced error detection and LSI controllers support this. Similar to how ZFS check every read for errors.
Current raid controllers ie sata, just rely on the drive saying the data is accurate. If a drive has a multi-bit error that is not detected (LACK of IOEDC) or multi-bit IOECC - you literally read bad data right on it (exception ZFS or PI). Sas has advanced multi-bit ECC and IOEDC (detection of errors).
So keep it simple, use a quality drive with the right controller - and oerprovision to 30% or more. MLC drives are so damn cheap you can afford to do 192gb out of 256gb or 384/512 or 786 out of 1TB. Small price to pay to get reliability and consistent speed (source anandtech).
My experience agrees with this. Samsung 840 pro was unreliable at stock OP, but moving to 30% OP gave us reliability.
For non-linear applications we use RAID-1 only. ESXi 5.1 - with small strip sizes to match the controller on the SSD page size.
