Date: Mon, 01 May 2000 16:25:31 -0700 From: Mike Smith <msmith@freebsd.org> To: Brad Knowles <blk@skynet.be> Cc: freebsd-stable@FreeBSD.ORG Subject: Re: How good is AMI MegaRAID support? Message-ID: <200005012325.QAA04307@mass.cdrom.com> In-Reply-To: Your message of "Tue, 02 May 2000 01:01:28 %2B0200." <v0422080ab533ba47bf86@[195.238.24.94]>
next in thread | previous in thread | raw e-mail | index | archive | help
> > This would suggest that you're hitting other limits. I seem to recall > > that you're only using one (or maybe two) host ports on the controller - > > this will be a bottleneck just for starters. > > I am using on interface on each array controller, each connected > to a dedicated host adaptor. Not really enough, if the array controller can keep up with the disks. > > The biggest killer with external RAID units is the miserable bandwidth > > that you get with just a single or even dual host ports. Fibrechannel > > can help here, but you really need to look at every part of the system > > before you start making these sort of judgements. > > The host bandwidth isn't an issue here. If it was, then setting > up a software stripe across the RAID-5 LUNs wouldn't buy me any > additional speed. I'm not coming anywhere close to what a single > SCSI controller can do. This definitely sounds like your array controllers aren't up to snuff. Especially if you're writing into or reading out of cache, you should be pegging the SCSI bus. > Besides, what I need is not bandwidth, but minimum latency. > 100GB over a 24 hour period of time equates to a little more than 4GB > per hour, about 70MB per minute, and about 1.2MB per second. Hell, > I've got some five year old ultra low-end 90MB hard drives that could > easily keep up with that data rate. SCSI is miserable for latency. Fibrechannel is better. PCI is better still. 8) > > I still haven't heard anything from anyone using embedded RAID adapters. > > Take some time with eg. an AMI MegaRAID 1600 or (once I'm done with the > > driver support) something like a 2000 series Mylex controller. > > I don't have any of those to play with. Now, if you want to send > me one or three, I'll be glad to look them over.... ;-) Wait 'till I get mine from AMI. I have a fairly tight deadline for getting these working, so I should be able to recommend them (or not) fairly quickly. > > Another very important item to bear in mind is that most controller > > vendors are just starting to wake up to real peformance issues and > > unix-like operating systems. It wasn't until I explained to Mylex, for > > example, the inherent badness in a power-of-2 stripe size in conjunction > > with the FFS disk layout that this even appeared on their radar. > > There are situations where you want to very carefully adjust the > stripe size to perfectly match the cylinder size. Joe Greco can tell > you a lot more about this than I can. Since the cylinder size is variable, this is somewhat bogus. > > Software RAID implementations like Vinum, that were designed from the > > start to address this, have some definite advantages there. > > Vinum has the distinct advantage of not having any pre-built > preferences for stripe size, etc.... I understand that a lot of > hardware RAID devices are optimized for essentially just one > configuration only, and if you change just a single parameter to be > different from the default, you're screwed. Your data may want a > very large stripe size (or maybe a very small one), but if the > controller doesn't agree, then you've got a problem. This is more or less correct. Some adapters (eg. those from ICP Vortex) allow you to select from a set of optimisation modes. > With vinum, you can select the stripe size you want/need for your > application, and while Greg will show you some charts that > demonstrate that the optimal stripe size is somewhere between 128KB > and 256KB, if your application doesn't fit the mold that Greg was > testing with, you can safely set the stripe size to be something > else, and vinum won't really care. This is what I'm trying to teach Mylex. They seem eager to learn; time alone will tell, I guess. -- \\ Give a man a fish, and you feed him for a day. \\ Mike Smith \\ Tell him he should learn how to fish himself, \\ msmith@freebsd.org \\ and he'll hate you for a lifetime. \\ msmith@cdrom.com To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-stable" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?200005012325.QAA04307>