SILI driver update - 10TB test succeeds

Matthew Dillon dillon at apollo.backplane.com
Sun Jun 28 12:40:39 PDT 2009


    I ran a test on a 10 terrabyte HAMMER filesystem over the last few days.
    I filled it up with copies of our iso-images directory, and I am now
    validating the data in the filesystem (which takes just as long as
    writing it out did in the first place).  However, spot checks indicate
    everything succeeded without corruption.

    The test used a Silicon Image 3132 connected to a Port Multiplier with
    5 x 2TB hard drives, in a simple striped configuration (no RAID).

    The SI 3132 board appears to have about a 150 MB/sec bandwidth limit
    per SATA port, despite the phy negotiating at 3 GBit/sec.  In otherwords,
    it isn't a high-end board but it does support FIS-based switching so
    it seems to be quite reasonable for most typical production workloads.

    --

    Numerous bug fixes have also gone into the 3132 driver.  I'm still working
    on some issues when hot-inserting a drive into a PM which is active...
    it may or may not be possible to work around them (the Sili chip has a
    lot of hardware chipset bugs that have to be worked around, but despite
    that the driver has been quite stable).

    I will repeat the 10TB test with the AHCI driver next week.

					-Matt
					Matthew Dillon 
					<dillon at backplane.com>





More information about the Kernel mailing list