Re: openiscsi 10gbe network
On Tue, Nov 24, 2009 at 08:07:12AM -0800, Chris K. wrote: Hello, I'm writing in regards to the performance with open-iscsi on a 10gbe network. On your website you posted performance results indicating you reached read and write speeds of 450 MegaBytes per second. In our environment we use Myricom dual channel 10gbe network cards on a gentoo linux system connected via fiber to a 10gbe interfaced SAN with a raid 0 volume mounted with 4 15000rpm SAS drives. Unfortunately, the maximum speed we are acheiving is 94 MB/s. We do know that the network interfaces can stream data at 822MB/s (results obtained with netperf). we know that local read performance on the disks is 480MB/s. When using netcat or direct tcp/ip connection we get speeds in this range, however when we connect a volume via the iscsi protocol using the open-iscsi initiator we drop to 94MB/s(best result. Obtained with bonnie++ and dd). What block size are you using with dd? Try: dd if=/dev/foo of=/dev/null bs=1024k count=32768 How's the CPU usage on both the target and the initiator when you run that? Is there iowait? Did you try with nullio LUN from the target? -- Pasi -- You received this message because you are subscribed to the Google Groups open-iscsi group. To post to this group, send email to open-is...@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?hl=en.
Re: iSCSI latency issue
Shachar f, on 11/25/2009 07:57 PM wrote: I'm running open-iscsi with scst on Broadcom 10Gig network and facing write latency issues. When using netperf over an idle network the latency for a single block round trip transfer is 30 usec and with open-iscsi it is 90-100 usec. I see that Nagle (TCP_NODELAY) is disabled when openning socket on the initiator side and I'm not sure about the target side. Vlad, Can you elaborate on this? TCP_NODELAY is always enabled in iSCSI-SCST. You can at any time have latency statistics on the target side by enabling CONFIG_SCST_MEASURE_LATENCY (see README). Better also enable CONFIG_PREEMPT_NONE to not count CPU scheduler latency. Are others in the mailing list aware to possible environment changes that effext latency? more info - I'm running this test with Centos5.3 machines with almost latest open-iscsi. Thanks, Shachar -- You received this message because you are subscribed to the Google Groups open-iscsi group. To post to this group, send email to open-is...@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?hl=en.
Re: openiscsi 10gbe network
The dd command I am running is time dd if=/dev/zero bs=1024k of=/mnt/ iscsi/10gfile.txt count=10240 My fs is xfs (mkfs.xfs -d agcount=8 -l internal,size=128m -n size=8k - i size=2048 /dev/sdb1 -f) those are the parameters used to format the drive. Here are the top values: Cpu(s): 0.0%us, 6.1%sy, 0.0%ni, 25.0%id, 67.2%wa, 0.1%hi, 1.7%si, 0.0%st I have not tried nullio LUN from target. I'm not sure how to go about it actually... Thanks for your help ! On Nov 25, 5:04 am, Pasi Kärkkäinen pa...@iki.fi wrote: On Tue, Nov 24, 2009 at 08:07:12AM -0800, Chris K. wrote: Hello, I'm writing in regards to the performance with open-iscsi on a 10gbe network. On your website you posted performance results indicating you reached read and write speeds of 450 MegaBytes per second. In our environment we use Myricom dual channel 10gbe network cards on a gentoo linux system connected via fiber to a 10gbe interfaced SAN with a raid 0 volume mounted with 4 15000rpm SAS drives. Unfortunately, the maximum speed we are acheiving is 94 MB/s. We do know that the network interfaces can stream data at 822MB/s (results obtained with netperf). we know that local read performance on the disks is 480MB/s. When using netcat or direct tcp/ip connection we get speeds in this range, however when we connect a volume via the iscsi protocol using the open-iscsi initiator we drop to 94MB/s(best result. Obtained with bonnie++ and dd). What block size are you using with dd? Try: dd if=/dev/foo of=/dev/null bs=1024k count=32768 How's the CPU usage on both the target and the initiator when you run that? Is there iowait? Did you try with nullio LUN from the target? -- Pasi -- You received this message because you are subscribed to the Google Groups open-iscsi group. To post to this group, send email to open-is...@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?hl=en.
iSCSI latency issue
I'm running open-iscsi with scst on Broadcom 10Gig network and facing write latency issues. When using netperf over an idle network the latency for a single block round trip transfer is 30 usec and with open-iscsi it is 90-100 usec. I see that Nagle (TCP_NODELAY) is disabled when openning socket on the initiator side and I'm not sure about the target side. Vlad, Can you elaborate on this? Are others in the mailing list aware to possible environment changes that effext latency? more info - I'm running this test with Centos5.3 machines with almost latest open-iscsi. Thanks, Shachar -- You received this message because you are subscribed to the Google Groups open-iscsi group. To post to this group, send email to open-is...@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?hl=en.
Re: openiscsi 10gbe network
Thank you for your response. The SAN is a 10gbe Nimbus with I believe to be iscsitarget(http://iscsitarget.sourceforge.net/) as it's target server. The switch is a Cisco Nexus5010 set to jumbo frame and flow control. We have through tcp/ip performance tests in conjunction with Cisco proved that this works. Furthermore using netcat and dd conjointly we have achieved speeds around 200MB/s. This is far from the 822MB/s shown in our testing with netperf and Cisco's performance tests, but it is way above what we are getting with iscsi at 94MB/s which technically is a GiG network not a 10gbe network. I am not familiar with no-op-io-scheduler where exactly is this set and what are it's implications ? Thank you once again for your help. On Wed, Nov 25, 2009 at 4:11 AM, Boaz Harrosh bharr...@panasas.com wrote: On 11/24/2009 06:07 PM, Chris K. wrote: Hello, I'm writing in regards to the performance with open-iscsi on a 10gbe network. On your website you posted performance results indicating you reached read and write speeds of 450 MegaBytes per second. In our environment we use Myricom dual channel 10gbe network cards on a gentoo linux system connected via fiber to a 10gbe interfaced SAN with a raid 0 volume mounted with 4 15000rpm SAS drives. That is the iscsi-target machine, right? What is the SW environment of the initiator box? Unfortunately, the maximum speed we are acheiving is 94 MB/s. We do know that the network interfaces can stream data at 822MB/s (results obtained with netperf). we know that local read performance on the disks is 480MB/s. When using netcat or direct tcp/ip connection we get speeds in this range, however when we connect a volume via the iscsi protocol using the open-iscsi initiator we drop to 94MB/s(best result. Obtained with bonnie++ and dd). What iscsi target are you using? Mike, is it still best to use no-op-io-scheduler on initiator? Boaz We were wondering if you would have any recommendations in terms of configuring the initiator or perhaps the linux system to achieve higher throughput. We have also set the the interfaces on both ends to jumbo frames (mtu 9000). We have also modified sysctl parameters to look as follows : net.core.rmem_max = 16777216 net.core.wmem_max = 16777216 net.ipv4.tcp_rmem = 4096 87380 16777216 net.ipv4.tcp_wmem = 4096 65536 16777216 net.core.netdev_max_backlog = 25 Any help would greatly be appreciated, Thank you for your time and your work. -- You received this message because you are subscribed to the Google Groups open-iscsi group. To post to this group, send email to open-is...@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?hl=en.
Re: openiscsi 10gbe network
Here is the dd command : time dd if=/dev/zero bs=1024k of=/mnt/iscsi/ 10gfile.txt count=10240 Here are the cpu values : Cpu(s): 0.0%us, 8.7%sy, 0.0%ni, 25.0%id, 64.0%wa, 0.4%hi, 1.9%si, 0.0%st - Client Cpu(s): 0.6%us, 2.8%sy, 0.0%ni, 86.4%id, 9.7%wa, 0.0%hi, 0.4%si, 0.0%st - SAN I have not tried the nullio LUN from the target... I'm not sure how to go about this ...? Thank you for your help. On Nov 25, 5:04 am, Pasi Kärkkäinen pa...@iki.fi wrote: On Tue, Nov 24, 2009 at 08:07:12AM -0800, Chris K. wrote: Hello, I'm writing in regards to the performance with open-iscsi on a 10gbe network. On your website you posted performance results indicating you reached read and write speeds of 450 MegaBytes per second. In our environment we use Myricom dual channel 10gbe network cards on a gentoo linux system connected via fiber to a 10gbe interfaced SAN with a raid 0 volume mounted with 4 15000rpm SAS drives. Unfortunately, the maximum speed we are acheiving is 94 MB/s. We do know that the network interfaces can stream data at 822MB/s (results obtained with netperf). we know that local read performance on the disks is 480MB/s. When using netcat or direct tcp/ip connection we get speeds in this range, however when we connect a volume via the iscsi protocol using the open-iscsi initiator we drop to 94MB/s(best result. Obtained with bonnie++ and dd). What block size are you using with dd? Try: dd if=/dev/foo of=/dev/null bs=1024k count=32768 How's the CPU usage on both the target and the initiator when you run that? Is there iowait? Did you try with nullio LUN from the target? -- Pasi -- You received this message because you are subscribed to the Google Groups open-iscsi group. To post to this group, send email to open-is...@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?hl=en.
Re: !!!!Help: Problem when I login the iscsi hard disk
Ricky wrote: sda: got wrong page You mean this right? The linux scsi layer was trying to figure out the cache type. It got an unexpected answer and so ... sda: assuming drive cache: write through it used the default of write through cache. sd 6:0:0:0: Attached scsi disk sda sd 6:0:0:0: Attached scsi generic sg0 type 0 You have new mail in /var/spool/mail/root -- You received this message because you are subscribed to the Google Groups open-iscsi group. To post to this group, send email to open-is...@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?hl=en. -- You received this message because you are subscribed to the Google Groups open-iscsi group. To post to this group, send email to open-is...@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?hl=en.
Re: [Patch 1/2] iscsiadm: login_portal() misses outputting logs for iscsid_req_by_rec()
Yangkook Kim wrote: Thanks for your patch. I tested your patch and it worked fine. So, next you will upload this patch to the git tree and the patch will become the part of source code in the next release of open-iscsi. Is my understanding correct? Yeah. I merged it and uploaded it to the git tree. The commit id is fb4f2d3072bee96606d01e3535c100dc99b8d331. It can take a couple of hours to show up (the disks have to get synced up or something), so you should see it shortly. When I make the next release it will be included. I am asking this question because I just want to know the normal development process of this and other linux project. No problem. 2009/11/24, Mike Christie micha...@cs.wisc.edu: Mike Christie wrote: Yangkook Kim wrote: Hi, Mike. Thank you for your patch. I do not want to add a login log message to the iscsid_req_* functions because they are generic and could be used for any operation. Yes, that's perfectly right idea. That should be bettet than my patch. I tried your patch, but that still does not output login-success message when calling iscsid_req_by_rec. It seems that log_login_msg() would not be called in either login_portal() or iscsid_logout_reqs_wait() when iscsid_req_by_rec returns success. I probably missed something. I will look at it tomorrow again. Nope. You are right. Nice catch. I messed up. I was only concentrating on the error paths. I will fix up my patch and resend. Thanks. Here is a corrected patch. -- You received this message because you are subscribed to the Google Groups open-iscsi group. To post to this group, send email to open-is...@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?hl=. -- You received this message because you are subscribed to the Google Groups open-iscsi group. To post to this group, send email to open-is...@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?hl=en. -- You received this message because you are subscribed to the Google Groups open-iscsi group. To post to this group, send email to open-is...@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?hl=en.
Re: openiscsi 10gbe network
Boaz Harrosh wrote: On 11/24/2009 06:07 PM, Chris K. wrote: Hello, I'm writing in regards to the performance with open-iscsi on a 10gbe network. On your website you posted performance results indicating you reached read and write speeds of 450 MegaBytes per second. In our environment we use Myricom dual channel 10gbe network cards on a gentoo linux system connected via fiber to a 10gbe interfaced SAN with a raid 0 volume mounted with 4 15000rpm SAS drives. That is the iscsi-target machine, right? What is the SW environment of the initiator box? Unfortunately, the maximum speed we are acheiving is 94 MB/s. We do know that the network interfaces can stream data at 822MB/s (results obtained with netperf). we know that local read performance on the disks is 480MB/s. When using netcat or direct tcp/ip connection we get speeds in this range, however when we connect a volume via the iscsi protocol using the open-iscsi initiator we drop to 94MB/s(best result. Obtained with bonnie++ and dd). What iscsi target are you using? Mike, is it still best to use no-op-io-scheduler on initiator? Sometimes. Chris, try doing echo noop /sys/block/sdXYZ/queue/scheduler Then rerun your tests. For your tests you might want something that can do more IO. If you can could try disktest or fio or even do multiple dds at the same time. Also what is the output of iscsiadm -m session -P 3 -- You received this message because you are subscribed to the Google Groups open-iscsi group. To post to this group, send email to open-is...@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?hl=en.
Re: !!!!Help: Problem when I login the iscsi hard disk
This information will also come out when I fdisk /dev/sda. And I can not mkfs this disk. I think the cache type should be write back. But I do not how to handle this situation. 2009/11/26 Mike Christie micha...@cs.wisc.edu Ricky wrote: sda: got wrong page You mean this right? The linux scsi layer was trying to figure out the cache type. It got an unexpected answer and so ... sda: assuming drive cache: write through it used the default of write through cache. sd 6:0:0:0: Attached scsi disk sda sd 6:0:0:0: Attached scsi generic sg0 type 0 You have new mail in /var/spool/mail/root -- You received this message because you are subscribed to the Google Groups open-iscsi group. To post to this group, send email to open-is...@googlegroups.com. To unsubscribe from this group, send email to open-iscsi+unsubscr...@googlegroups.comopen-iscsi%2bunsubscr...@googlegroups.com . For more options, visit this group at http://groups.google.com/group/open-iscsi?hl=en. -- You received this message because you are subscribed to the Google Groups open-iscsi group. To post to this group, send email to open-is...@googlegroups.com. To unsubscribe from this group, send email to open-iscsi+unsubscr...@googlegroups.comopen-iscsi%2bunsubscr...@googlegroups.com . For more options, visit this group at http://groups.google.com/group/open-iscsi?hl=en. -- You received this message because you are subscribed to the Google Groups open-iscsi group. To post to this group, send email to open-is...@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?hl=en.
Re: [Scst-devel] iSCSI latency issue
On Wed, Nov 25, 2009 at 5:57 PM, Shachar f shacharf...@gmail.com wrote: I'm running open-iscsi with scst on Broadcom 10Gig network and facing write latency issues. When using netperf over an idle network the latency for a single block round trip transfer is 30 usec and with open-iscsi it is 90-100 usec. I see that Nagle (TCP_NODELAY) is disabled when openning socket on the initiator side and I'm not sure about the target side. Vlad, Can you elaborate on this? Are others in the mailing list aware to possible environment changes that effext latency? more info - I'm running this test with Centos5.3 machines with almost latest open-iscsi. Please make sure that interrupt coalescing has been disabled -- see also ethtool -c. Bart. -- You received this message because you are subscribed to the Google Groups open-iscsi group. To post to this group, send email to open-is...@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?hl=en.
Re: openiscsi 10gbe network
On 25 Nov 2009 at 14:15, Chris K. wrote: Here are the cpu values : Cpu(s): 0.0%us, 8.7%sy, 0.0%ni, 25.0%id, 64.0%wa, 0.4%hi, A note: I don't know how well open-iscsi uses multiple threads, but looking at individual CPUs may be interesting, as the above is only an average for multiple CPUs. Press '1' in top to switch to individual CPU display. Hope you don't have too many cores ;-) Here's some example for the different displays: Cpu(s): 23.0%us, 1.2%sy, 0.0%ni, 73.8%id, 1.9%wa, 0.0%hi, 0.2%si, 0.0%st Cpu0 : 4.2%us, 0.5%sy, 0.1%ni, 89.2%id, 5.6%wa, 0.1%hi, 0.3%si, 0.0%st Cpu1 : 4.8%us, 0.5%sy, 0.1%ni, 94.0%id, 0.6%wa, 0.0%hi, 0.0%si, 0.0%st Cpu2 : 7.9%us, 0.7%sy, 0.0%ni, 90.7%id, 0.7%wa, 0.0%hi, 0.0%si, 0.0%st Cpu3 : 8.6%us, 0.7%sy, 0.0%ni, 90.2%id, 0.4%wa, 0.0%hi, 0.0%si, 0.0%st Have fun! Ulrich -- You received this message because you are subscribed to the Google Groups open-iscsi group. To post to this group, send email to open-is...@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?hl=en.