From: FUJITA Tomonori <[EMAIL PROTECTED]> Subject: [Iscsitarget-devel] Re: [ANNOUNCE] iSCSI enterprise target software Date: Tue, 01 Mar 2005 18:35:49 +0900
> The last reason is that user-space cost like memory copy. With 1Gbs > Ethernet, is is not critical. However, with 10G, it is critical, I > expect. I've been setting up 10G experimental infrastructure to > evaluate iSCSI performance. If we try to build high-performance iSCSI target software on 10Gbs Ethernet, I think that we need to implement it in the kernel space, although this topic is still in the research stage. IICR, Intel provides open-source iSCSI target software, which uses mmap() in the user space, as you suggested. However, they chose a different approach to implement iSCSI target software for 10Gbs. They modified the Linux kernel TCP stack to fully integrate iSCSI functionality with it. It is still not clear how to get the best performance out of 10GBs Ethernet, however, it seems that using just the socket interface is not good enough. So I think that we need to control all the system-resources to do it. We've not optimized our code yet. The current code simply uses the socket interface, however, moving some functionality to the TCP stack can improve the performance, although I don't plan to modify the TCP stack. A possible approach to do it is exploiting sk_data_ready(). In addition, you can find several papers (e.g., Chelsio technical papers) saying that UP kernel can achieve better throughput than SMP kernels with 10Gbs. There is some room for further improvement in the Linux kernel. - To unsubscribe from this list: send the line "unsubscribe linux-scsi" in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html
