Solaris server box: Supermicro Atom board (http://www.supermicro.com/products/motherboard/ATOM/945/X7SLA.cfm?typ=H)
Clients (Windows XP): Supermicro GreenCreek board with dual Intel NICs on ESB2 (http://www.supermicro.com/products/motherboard/Xeon1333/5000X/X7DAE.cfm) Gigabyte P55A-UD4P with RTL8111D NIC ((http://www.gigabyte.com.tw/Products/Motherboard/Products_Spec.aspx?ClassValue=Motherboard&ProductID=3238&ProductName=GA-P55A-UD4P) All systems connected to a an unmanaged Gb switch. The first time it failed, I was running the two NICs on the Solaris box in aggregation mode (yes, I know it will cause with multicast but I was trying to decide whether it's worth getting a managed switch). The Solaris box was blasting data in two iperf sessions to the two windows boxes. Then, I removed the aggregation and started testing a single NIC on the Solaris box. I had the two other boxes copying files from the Solaris box to the their local filesystems. Since SMB limits the data rate for a single transfer, I started 2-3 copies from each windows box. Total BW out of the solaris box peaked at 600-700 Mbps. All the file copies involved the same files, so the Solaris box didn't hit the disks very hard at all. I tried both the -3 and -6 driver versions from masa with the same results. A few minutes into the transfer the Realtek NIC goes silent. As I had mentioned the mcast version worked fine but the iperf throughput from the box was limited to 150 Mbps. When the driver works, performance really rocks. Lower utilization that the PCI Intel MT1000 NIC I am using for now and higher throughput hitting more than 900 Mbps with iperf. Is there any way to recognize that the NIC has hung and restart with a brief delay? Or maybe can we afford to trigger trasmission whenever we get an interrupt? -- This message posted from opensolaris.org _______________________________________________ opensolaris-discuss mailing list opensolaris-discuss@opensolaris.org