Permabanned
- Joined
- 13 Nov 2006
- Posts
- 5,798
My motherboard only supports raid 1 + 0. As I want raid5 is the software raid setup any good?
Software Raid Solution ??
- Thread starter Johnnytoxic
- Start date
If you're talking about RAID using Windows dynamic disks then only the server versions support the RAID5esque mode and I don't think it's possible to boot from a striped parity array.
As with any software based RAID5 solution don't expect blinding write performance, the main CPU has to do all the XOR parity calculations and they're pretty complex. You could find the write speed down around 20-30Mb/s compared to 100Mb/s+ for a RAID0 array of the same size.
As with any software based RAID5 solution don't expect blinding write performance, the main CPU has to do all the XOR parity calculations and they're pretty complex. You could find the write speed down around 20-30Mb/s compared to 100Mb/s+ for a RAID0 array of the same size.
Which OS?
You can make Windows XP do it by editing dmboot.sys, dmconfig.dll, and dmadmin.exe slightly
(a google about those files should tell you everything) but you'd need at least 4 SATA ports to do it - one to boot, and 3 Windows 'dynamic disks' on which to put the RAID 5 array.
In Linux there's no issues directly booting from a software RAID 5 array, via an initrd. I used to get 30-40MB/sec on writes, 60-70MB/sec on reads when I had a VIA-C3 1.2Ghz fileserver with Linux software RAID 5, but that was over a PCI SATA controller.
133MB/s PCI limit / 4 drives = 33MB per drive total = 16MB per direction .. considering a write requires a read to recalculate the parity, this gives a theoretical (uncached) max of 48MB/s writes, 100MB/s on reads. With onboard SATA connected via a PCI-E bus, you should get results similar to that of a dedicated RAID controller.
rpstewart: I think the bus bandwidth is generally more of an issue than CPU
power when performing the parity calculations...
Linux software raid (md) in-memory benchmark, Athlon64 3800+ :
raid5: using function: pIII_sse (6990.800 MB/sec)
raid6: int32x1 848 MB/s
raid6: int32x2 899 MB/s
raid6: int32x4 870 MB/s
raid6: int32x8 578 MB/s
raid6: mmxx1 1809 MB/s
raid6: mmxx2 3298 MB/s
raid6: sse1x1 1645 MB/s
raid6: sse1x2 2469 MB/s
raid6: sse2x1 2250 MB/s
raid6: sse2x2 2947 MB/s
You can make Windows XP do it by editing dmboot.sys, dmconfig.dll, and dmadmin.exe slightly
(a google about those files should tell you everything) but you'd need at least 4 SATA ports to do it - one to boot, and 3 Windows 'dynamic disks' on which to put the RAID 5 array.In Linux there's no issues directly booting from a software RAID 5 array, via an initrd. I used to get 30-40MB/sec on writes, 60-70MB/sec on reads when I had a VIA-C3 1.2Ghz fileserver with Linux software RAID 5, but that was over a PCI SATA controller.
133MB/s PCI limit / 4 drives = 33MB per drive total = 16MB per direction .. considering a write requires a read to recalculate the parity, this gives a theoretical (uncached) max of 48MB/s writes, 100MB/s on reads. With onboard SATA connected via a PCI-E bus, you should get results similar to that of a dedicated RAID controller.
rpstewart: I think the bus bandwidth is generally more of an issue than CPU
power when performing the parity calculations...
Linux software raid (md) in-memory benchmark, Athlon64 3800+ :
raid5: using function: pIII_sse (6990.800 MB/sec)
raid6: int32x1 848 MB/s
raid6: int32x2 899 MB/s
raid6: int32x4 870 MB/s
raid6: int32x8 578 MB/s
raid6: mmxx1 1809 MB/s
raid6: mmxx2 3298 MB/s
raid6: sse1x1 1645 MB/s
raid6: sse1x2 2469 MB/s
raid6: sse2x1 2250 MB/s
raid6: sse2x2 2947 MB/s