From: Mark Weaver mdw1982@mdw1982.com
Hi all, I was wondering if anyone has tried and succeeded to use this controller. Its the bc4852 SATA controller. There's supposed to be support for it, but so far we've not been able to get the driver to load during the install and thus we've got no "valid drive" on which to load the OS.
Last time I checked, but I could be wrong, Broadcom's RAIDCore solutions are FRAID. They are very, very high-end FRAID designs (PCI-X, 8 channel), with lots of software IP for rebuild options, etc..., but still FRAID. I've never personally tried the cards because they aren't any cheaper than 3Ware solutions. But they claim integration tools with Linux's LVM/MD (I have no idea how these work) and should have at least ATA/SATA channel drivers for just a bunch of disks (JBOD) as with any ATA/SATA controller.
Luckily Broadcom's new IC for either PCIe x4 or PCI-X 1.0 is an intelligent, MIPS-based hardware solution for either SATA or new Serial Attached SCSI (SAS). The pricing should be similar with the controller being under $60/1K quantity, but finally bring some serious performance to the table (along with up to 768MB of battery-backed DRAM).
-- Bryan J. Smith mailto:b.j.smith@ieee.org
Raidcore is FRAID or software RAID.
Bryan J. Smith b.j.smith@ieee.org wrote:
From: Mark Weaver mdw1982@mdw1982.com
Hi all, I was wondering if anyone has tried and succeeded to use this controller. Its the bc4852 SATA controller. There's supposed to be support for it, but so far we've not been able to get the driver to load during the install and thus we've got no "valid drive" on which to load the OS.
Last time I checked, but I could be wrong, Broadcom's RAIDCore solutions are FRAID. They are very, very high-end FRAID designs (PCI-X, 8 channel), with lots of software IP for rebuild options, etc..., but still FRAID. I've never personally tried the cards because they aren't any cheaper than 3Ware solutions. But they claim integration tools with Linux's LVM/MD (I have no idea how these work) and should have at least ATA/SATA channel drivers for just a bunch of disks (JBOD) as with any ATA/SATA controller.
Luckily Broadcom's new IC for either PCIe x4 or PCI-X 1.0 is an intelligent, MIPS-based hardware solution for either SATA or new Serial Attached SCSI (SAS). The pricing should be similar with the controller being under $60/1K quantity, but finally bring some serious performance to the table (along with up to 768MB of battery-backed DRAM).
-- Bryan J. Smith mailto:b.j.smith@ieee.org
CentOS mailing list CentOS@centos.org http://lists.centos.org/mailman/listinfo/centos
William Warren wrote:
Raidcore is FRAID or software RAID.
according to Broadcom the bc4852 is indeed hardware RAID. we've been attempting to get this going with their Linux drivers, but keep getting stuck at the point where the Linux installation hangs loading the driver.
On Sat, 2005-06-25 at 21:29 +0000, Mark Weaver wrote:
according to Broadcom the bc4852 is indeed hardware RAID. we've been attempting to get this going with their Linux drivers, but keep getting stuck at the point where the Linux installation hangs loading the driver.
Ahhh, no. From their own marketing materials on the RAIDCore ...
"Fulcrum Architecture Platform-independent RAID *SOFTWARE* architecture"
"XcelCore Software RAID *SOFTWARE* stack providing highly integrated storage subsystem"
Now they put together some interesting, 4-8 channel cards. The design is to maximize throughout to a set of 4-8 ATA channels. But they feature *0* "intelligence" on the card itself -- *0* memory (other than a little extra SRAM buffer in the ATA channels). They are 100% driven by the host (i.e., the main CPU).
The "software" is designed around the ATA channel setup, and Broadcom offers extensive features (and charges piecemeal for them). These software features can actually offer more RAID levels, rebuild/resize, etc... options than most "intelligent" RAID cards. But they are still driven by the main CPU.
When the RAIDCores first came out, they were pretty impressive in the benchmarks -- until you looked at the benchmarks. I.e., 100% _read_ RAID-5 benchmarks because RAID-5 is basically just a RAID-0 when reading. When you are writing data to a RAIDCore RAID-5, or rebuilding it, the performance is absolutely horrendous because your system interconnect is busy pushing _all_ data through the CPU (the actual XOR operation is not the problem).
On Sat, 2005-06-25 at 21:41 -0500, Bryan J. Smith wrote:
Ahhh, no. From their own marketing materials on the RAIDCore ... "Fulcrum Architecture Platform-independent RAID *SOFTWARE* architecture" "XcelCore Software RAID *SOFTWARE* stack providing highly integrated storage subsystem"
The confusion arises from who Broadcom is targeting. They aren't targeting general server usage. They are targeting OEMs who want to build intelligent storage systems.
If I was designing a near-line storage server or other, _dedicated_ storage device, then Broadcom's solutions is _extremely_ flexible. I can mix'n match multiple controllers across multiple buses and use the Fulcrum software to control it all.
But when I'm designing a server who's purpose is not to be just a storage device, but an actual application, file or other server, I don't want the Fulcrum software load on my system's interconnect.
-
Bryan J. Smith -
Bryan J. Smith <b.j.smith＠ieee.org> -
Mark Weaver -
William Warren