Ugo Bellavance <ugob at camo-route.com> wrote: > Anyone tried netcell products NetCell's RAID-XL is basically a fixed 32-bit (2+1 drive) and 64-bit (4+1 drive) [S]ATA RAID-3 implementation. Before I talk more about how RAID-XL works, let's talk RAID-3. RAID-3 and RAID-4 use striped data with dedicated parity, whereas RAID-5 uses striped data and parity. RAID-4 and 5 stripe large blocks, so multiple access could be possible (reducing latency), whereas RAID-3 writes sectors immediate and [virtually] parallel to all disks. RAID-4 and RAID-5 are used in servers where lots of random I/O is occurring. RAID-4 is most ideal for large reads and writes, where latency on the parity disk is not a concern (e.g., NetApp often uses RAID-4 because NFS v3 blocks are typically 32KB). RAID-5 is most ideal for lots of smaller, random RAID reads and writes, because the parity is also striped. RAID-3 is most ideal for desktops with limited processes writing to/from the disk. RAID-3 attempts to push as much data parallel to the disks directly. The NetCell approach does this by using 2 ATA disks (16-bit wide data each), plus 1 parity, for it's 32-bit cards, and 4 data disks (again, 16-bit wide data each), plus 1 parity, for it's 64-bit cards. XOR (parity) is calculated in-line by the on-board microcontroller in real-time, and written to the parity disk, and has up to 128MB of DRAM for buffering as necessary. I would _not_ use the card for a traditional server. But for a desktop, workstation or media server where you want burst disk access for only a few processes, it's RAID-XL is most ideal -- especially the 5-disc, 64-bit card. I have the SR5000 myself. > on Centos 3 or 4? Only updated Fedora Core 3/4 kernels (read on). > They claim driverless operation under WinXP, Well, Windows XP has some serious "identity crisis" issues when it comes to ATA -- especially through different service packs and hotfixes. I've documented those in my various notes on XP, postings, etc... In a nutshell, for NetCell, you need a hotfixes after SP2, and there are some recommended for pre-SP2 as well. > but in linux > the only doc that is related with CentOS would be the Fedora > core 2 (http://www.netcell.com/support/Fedora_Core2_readme.txt) > and requires a kernel rebuild. In Linux, the PCI ID and ATA DMA support is added to the ATA driver. This is why it requires a kernel rebuild. As of kernel 2.6.12.3 (IIRC -- could be 2.6.13.2?), the NetCell products are now supported in the stock ATA driver. Since the NetCell is designed for desktops and direct I/O, it doesn't make sense to use a traditonal SCSI driver. There are no real services, queuing, etc... It's really best as a "dumb block ATA device" that is written to and read from directly. That's why it was designed to show itself as a 1 or 2 device ATA channel. > I'd like to know if it would be possible to install > CentOS directly on drives on such a card, without needing a > build box... You'd have to rebuild the installer with a newer kernel like 2.6.12.3. I had tried to patch the ATA support in before, and it was a bit of a PITA. But I am running my SR5000 card in a Fedora Core 4 test system, using five (5) Seagate 7200.8 200GB drives. Thing moves data like I've _never_ seen. I'm using my SR5000 for a prototype multimedia server in my house. > http://www.tomshardware.com/2003/11/28/kill_scsi_ii/ Yes, and Tom's showing off single process benchmarks -- like most enthusiast sites. Look at them as viable for desktop-only evaluation. Start throwing a few other processes at it and you'll quickly appreciate why 3Ware, Areca, LSI and other cards are better for servers. ;-> -- Bryan J. Smith | Sent from Yahoo Mail mailto:b.j.smith at ieee.org | (please excuse any http://thebs413.blogspot.com/ | missing headers)