Date: Fri, 12 Oct 2001 16:56:03 -0600 From: Scott Long <scott_long@btc.adaptec.com> To: Alexey Koptsevich <alex@astro.su.se> Cc: scsi@freebsd.org Subject: Re: Adaptec 3210S: RAID5 performance and bus throughput Message-ID: <20011012165603.D469@hollin.btc.adaptec.com> In-Reply-To: <Pine.GSO.4.10.10110122310180.12138-100000@dioscuri.astro.su.se> References: <20011011123557.A406@hollin.btc.adaptec.com> <Pine.GSO.4.10.10110122310180.12138-100000@dioscuri.astro.su.se>
next in thread | previous in thread | raw e-mail | index | archive | help
On Sat, Oct 13, 2001 at 12:37:09AM +0200, Alexey Koptsevich wrote: > > Hello, > > > > Also, how many discs can I put to a channel and avoid decreasing > > > performance? I think, if modern disk is able to transfer 30 Mb/s only 5 > > > disks can be put to one channel, so 10 disks is practical limit for > > > 2-channel Ultra160 controller. But it is only an idea, what does follow > > > from the experience? > > > > It follows the law of dimishing returns. Every disk that you add will > > increase performance my a lower factor than the previous. You also > > have to remember that performace is also based on the PCI bus and > > the CPU on the controller. If the card is only in a PCI 33MHz/32Bit > > slot, you can only get 132MB/s max (actually much less than that in > > real life). > > Since the bus is real bottleneck, I wonder what is the purpose to have > 4-channel RAID controllers, in principle? Just to connect many discs? I > mean, 4-channel controller gives 160*4=640Mb/s, but this cannot be driven > even by 64bit/66MHz bus which provides only 528Mb/s! Also, are there x86 > motherboards which provide those 64bit/66MHz? Yes, but they are more in the high-end scale and not found in very many white boxes. With 4 channels, you can create one huge array, or a whole bunch of small arrays. No, you won't be able to run at the aggregate spindle speed, but that is not the only factor that is considered important by some people. > > I also wonder what is real performance of 3210S in RAID5 mode? Are 4-5 > disks which saturate 160Mb/s on SCSI bus (assume that load is heavy) are > able to saturate those 132Mb/s on PCI bus after performing all XOR > calculations? If not, how many discs saturate 132Mb/s? RAID5 read performance will be almost identical to RAID0 read performance. Write performace will be considerably less because it involves more than just writing to one disk; if you are writing a piece of data that is less than the size of a full stripe, you will also have to read the parity disk, compute the parity, and then write out the data plus the parity. It is the price that is paid with RAID5. Also, for a 3210 card, I would contact your vendor and make sure that you have the latest firmwware. > And, as a conclusion from previous question, is there real purpose to have > 2-channel controller on 32bit/33MHz PCI (which is used on my Intel > L440GX, assuming its 66MHz PCI to be weak), using it in RAID5 mode? Two channels are great for mirrored arrays (RAID1 or RAID 0/1). You separate each side of the mirror on a different channel, and the controller can write the data in true parallel. Scott To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-scsi" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20011012165603.D469>