Re: iscsi on debian lenny and netxtreme II cards
On 09/07/2009 06:31 AM, Angel L. Mateo wrote: > El vie, 04-09-2009 a las 10:46 -0500, Mike Christie escribió: > >> You cannot use this transport with the upstream open-iscsi tools. The >> 2.6.31 kernel will have a bnx2i driver for that card, but you will still >> need a special broadcom daemon to use the offload capabilites of the >> card. Broadcom and other offload guys like Chelsio are working on >> merging a common daemon/lib into the open-iscsi package right now. >> > But, open-iscsi reports me that transport: > > felis305:~# iscsiadm -m iface > default tcp,default,default,unknown > iser iser,default,default,unknown > bnx2i bnx2i,default,default,unknown > That should be removed now. It got in there by accident. At one point Broadcom was hoping that you could just use the normal netdev tools for network setup, and so for iscsi setup you could just tell it to use bnx2i. It did not work out, so we have to set params like the ip address. --~--~-~--~~~---~--~~ You received this message because you are subscribed to the Google Groups "open-iscsi" group. To post to this group, send email to open-iscsi@googlegroups.com To unsubscribe from this group, send email to open-iscsi+unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/open-iscsi -~--~~~~--~~--~--~---
Re: iscsi on debian lenny and netxtreme II cards
El vie, 04-09-2009 a las 10:46 -0500, Mike Christie escribió: > Can you try different IO schedulers > > echo noop > /sys/block/sdXYZ/queue/scheduler > > And for your IO test what are you using and what IO sizes are used? For > a good throughput you want lots of IO (around the queue depth of the > target which is probably around 128 or 256) and larger IOs (around 64 or > 128K). > I am running tests with dd, trying different block sizes: lynx0:~# dd if=/dev/zero of=/var/LISTAS/kk bs=16k count=10k 10240+0 records in 10240+0 records out 167772160 bytes (168 MB) copied, 2,37322 s, 70,7 MB/s lynx0:~# dd if=/dev/zero of=/var/LISTAS3/kk bs=16k count=10k 10240+0 records in 10240+0 records out 167772160 bytes (168 MB) copied, 2,42447 s, 69,2 MB/s The first is a NFS filesystem and the second an iSCSI one. But this time, performance is very similar. When I sent this firs message, iSCSI performance was about 10MB/s and the only change I've done meanwhile has been restarting iscsi daemon :-( -- Angel L. Mateo Martínez Sección de Telemática Área de Tecnologías de la Información _o) y las Comunicaciones Aplicadas (ATICA) / \\ http://www.um.es/atica_(___V Tfo: 868887590 Fax: 86337 --~--~-~--~~~---~--~~ You received this message because you are subscribed to the Google Groups "open-iscsi" group. To post to this group, send email to open-iscsi@googlegroups.com To unsubscribe from this group, send email to open-iscsi+unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/open-iscsi -~--~~~~--~~--~--~---
Re: iscsi on debian lenny and netxtreme II cards
El vie, 04-09-2009 a las 10:46 -0500, Mike Christie escribió: > You cannot use this transport with the upstream open-iscsi tools. The > 2.6.31 kernel will have a bnx2i driver for that card, but you will still > need a special broadcom daemon to use the offload capabilites of the > card. Broadcom and other offload guys like Chelsio are working on > merging a common daemon/lib into the open-iscsi package right now. > But, open-iscsi reports me that transport: felis305:~# iscsiadm -m iface default tcp,default,default,unknown iser iser,default,default,unknown bnx2i bnx2i,default,default,unknown -- Angel L. Mateo Martínez Sección de Telemática Área de Tecnologías de la Información _o) y las Comunicaciones Aplicadas (ATICA) / \\ http://www.um.es/atica_(___V Tfo: 868887590 Fax: 86337 --~--~-~--~~~---~--~~ You received this message because you are subscribed to the Google Groups "open-iscsi" group. To post to this group, send email to open-iscsi@googlegroups.com To unsubscribe from this group, send email to open-iscsi+unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/open-iscsi -~--~~~~--~~--~--~---
Re: iscsi on debian lenny and netxtreme II cards
On 09/04/2009 06:54 AM, Angel L. Mateo wrote: > Hello everyone, > > I am introducing myself to iSCSI world, so maybe these are newbie > questions. I'm sorry. > > My first test has been to create a LUN (in a celerra file server) and > export it with iscsi. I have mounted this lun in a debian server > (lenny), created a file system, and mounted it. The problem I have is > that performance is poor than in a NFS FS (with NFS I get ~50MB/s and > with iSCSI I get ~10MB/s). I am mounting NFS and iSCSI filesystem over a > bonding (active/pasive) of two gigabit cards. I am using tcp transport > for iSCSI. Can you try different IO schedulers echo noop > /sys/block/sdXYZ/queue/scheduler And for your IO test what are you using and what IO sizes are used? For a good throughput you want lots of IO (around the queue depth of the target which is probably around 128 or 256) and larger IOs (around 64 or 128K). > > Another question related with this, is that one of my cards is a > Broadcom NetXtremm II BMC5708S wich supports iSCSI. Although some man > pages talk about a bnx2i transport, I haven't found any kernel module > for this. Could I use this transport? How? Should it improves > performance? > You cannot use this transport with the upstream open-iscsi tools. The 2.6.31 kernel will have a bnx2i driver for that card, but you will still need a special broadcom daemon to use the offload capabilites of the card. Broadcom and other offload guys like Chelsio are working on merging a common daemon/lib into the open-iscsi package right now. I will let the broadcom guys talk about perf more because I do not have any numbers handy. From my testing I have seen it performs well with smaller IOs (less than 32K) and iscsi_tcp does not. With larger IOs and lots of them the throughput seems to the be same (at least with a 1 gig network), but the braodcom cpu usage is much less. iscsi_tcp has a xmit thread which can take up almost 100% of the cpu at times, but broadcom does not have that thread since it does the same operations in hardware. --~--~-~--~~~---~--~~ You received this message because you are subscribed to the Google Groups "open-iscsi" group. To post to this group, send email to open-iscsi@googlegroups.com To unsubscribe from this group, send email to open-iscsi+unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/open-iscsi -~--~~~~--~~--~--~---