Dear all, After switching to newer backplane revisions we have a more stable system, but far from perfect. We are using the following backplane from supermicro:
BPN-SAS-836EL2 3U SAS / SATA Expander Backplane with dual LSI SASX28 Expander Chips in combination with 1x AOC-USAS-L8I (based on Lsi 1068E) HBA Per system we use 12-14 x Western digital RE3 1002FBYS SATA disks + 1 x ssd mlc + 1x ssd slc OS: Nexenta Core Platform 3.0.1 (snv 134) After upgrading the LSI controller to firmware 1.30.00 the ssd's give less errors in the console. [b]Problem is that this setup still gives SMP PHY control link reset issues under heavy load (scrub etc) [/b] Supermicro's support is worthless, so we discussed this problem with western digital. After a couple of weeks we received the following answer: [i]Firstly from discovery info, PHY identifier 0x18 is destination SAS address ending 9C:5A:58 – a virtualized address for a SATA device. A write command tag 0x1C was sent earlier (off the top of the trace but it was WFPDMA queued LBA 0x13F69E39 for 0x100 sectors sent around 3.6ms before the trigger point). The drive signals it’s ready for transfer and the host begins to send data. [b]http://www.boeri.be/trace.jpeg[/b] Part way through the transfer, the drive begins sending HOLDS to the host. This continues for around 2.2ms until the HOST sends an SMP PHY control link reset request for the PHY identifier 0x18 (associated with the SATA device that is sending HOLDs) causing the expander to send BREAK. I’ve just shown host -> expander link 3->4 in the above – traffic is continuing on links 1->2 to another device. The HOST sends the SMP PHY control request via links 1->2 so you don’t see it above but its around 40us before the expander responds with the BREAK. So it seems to me there’s a possibility that if a drive gets in a state where it needs to send HOLDS for a period of time, the host may have some kind of timeout – do you have a view as to why this would be happening? [/i] Has someone the same experiences ? Is there really a problem with the mpt_sas driver? Can we switch to adaptec hba's, are they better? I think the newer storagetek hba's are based on the adaptec 5805 ? Point to point sata links are not an ideal solution for us, neither are sas drives. We like to hold the backplane solution. Thanks. -- This message posted from opensolaris.org _______________________________________________ storage-discuss mailing list storage-discuss@opensolaris.org http://mail.opensolaris.org/mailman/listinfo/storage-discuss
