I have a Sun M4000 connected to an EMC CX4-120 array with a write-heavy database. Writes peak at around 1200 IO/s and 12MB/s.
According to EMC, I am saturating the write cache on the EMC array.
I think the simplest solution is to move the redo logs to a DRAM based SSD. This will reduce the load on the EMC array by half and apps won't be seeing log buffer waits. Yes, the DBWR may become a bottleneck, but the apps won't be waiting for it (like they do on redo commits!)
I currently cycle through about 4 4GB redo logs, so even 20GB or so of SSD would make a big difference. Since this is short-term storage and is constantly being overwritten, Flash based SSDs probably aren't a great idea.
The M4000 doesn't have any extra drive lots, so a PCI-E card would be perfect, I could go external or move boot volumes to the EMC and free up the local drives.
Sun sells a Flash Accelerator F20 PCIe card, but that seems to be a cache for some SATA disks, not a DRAM SSD solution. Details are sketchy, it doesn't list the M4000 as supported, and I'm tired of fighting Sun's phone tree looking for human help. :(
Do others agree that a DRAM SSD is the way to go? Any hardware recommendations?
UPDATE In addition to the info in a comment below, I tried various settings for "commit_write" and it didn't make a difference.