On Sun, 27 Jan 2002, Jakob ?stergaard wrote: > On Sat, Jan 26, 2002 at 02:35:24PM -0800, Micah Anderson wrote: > > Hello, > > > > I've been trying to locate a hardware RAID that is worth the price > > that they go for. From what I have been able to find many of the > > hardware raid controllers are slower or just as CPU dependant as > > software RAID, except for the AMI Elite and Vortex ICP cards. > > It's a common misunderstanding that software RAID is CPU intensive. > > Most levels require virtually no extra work from the CPU. RAID-4 and RAID-5 > being the exceptions (because they need to calculate parity). However, the > PIII-550 next to me here can calculate parity on more than 1.3 GB of data *per > second*. Even a 100MB/sec parity calculation would eat less than 10% of the > CPU power. One one of the two CPUs... I thought recovery/resync takes quite a bit of CPU overhead until it is finished? Even though this will be a SCSI system and a SCSI controller will be needed, and this will only be RAID 0 or 0+1 (can software RAID even do RAID 10 or 0+1? I don't know if it can), I think that there is still some overhead incurred on the system by running software raid, perhaps it is slight, but on a system that is going to be a database server and needs to be as quick as possible, slimming down and trimming out the excess is always a good thing. > I'm biased against hardware RAID, because of the following reasons: > > *) If the RAID firmware has some problem, there's no way that you > can fix it, or pay anyone else to fix it. Software RAID has had > it's share of problems in the past and will meet new problems in > the future, but people can fix it, as they have in the past. Agreed, the software raid community is good. However, I might argue that the same thing could be said about the SCSI card I am going to have to get to control the drives, the motherboard BIOS, etc. There is firmware in all of this, having firmware on a hardware RAID as opposed to firmware on a SCSI card doesn't seem to me a big difference. > *) The hardware solution tends to be rediculously expensive, while typically > yielding lower performance and less flexibility. *(see end note). I totally agree, WAY over priced. So far the only hardware solutions that I've seen that have been given the kudos as far as better performance goes are the cards that are in the $800+ range, such as the Elite 1600 AMI card. I would argue that *some* cards have more flexibility than software raid, including more reporting of statistics and such, but again, you pay a LOT of these cards. > *) The only difference between "hardware raid" and "software raid", is on > which CPU the RAID calculations occur. Hardware RAID just means, you > insert an extra CPU in the chain of command (thereby increasing overall > latency), to offload the main CPU (which could easily handle those few > operations RAID requires). This made sense 10 years ago when the added > CPU (typically an i960, i486 or motorola 68-something) actually made > some significant difference in the overall CPU power in the system. Today > the CPUs on typical low/medium end hardware RAID controllers are a joke > compared to the main CPU. They still add the latency, but they do not > in practice offload the main CPU noticably. Except when you have a SCSI card, so you have to offload the drive controller functions to another CPU and then the RAID functions onto a separate CPU. I don't think it makes sense to compare the relative CPU power of a system-wide CPU to a specific task CPU. The system CPU naturally is doing quite a bit more than the RAID CPU would be, so obviously it is going to need to be much more powerful than the RAID CPU is going to need to be in order to perform its task. Although I disagree with you, I think you have actually convinced me to stick with software raid, save about $700 and get more disks. Micah - To unsubscribe from this list: send the line "unsubscribe linux-raid" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
