Hello open-iscsi members,could someone give me a hint where to find more about the reason why the mentioned iscsi error appears?
We are doing benchmarks with fio and the following errors occurs after some time. The monitoring interface of the iscsi target appliance suddenly shows no more throughput for about 20secs to 2 minutes. Then throughput starts again.
The same errors also appear with a huge database import (postgresql). So it is probably not related to fio but more to heavy i/o.
/var/log/messages:Aug 16 16:47:45 ibs-mngr2 kernel: [80500.790747] connection15:0: detected conn error (1021) Aug 16 16:47:46 ibs-mngr2 iscsid: Kernel reported iSCSI connection 15:0 error (1021 - ISCSI_ERR_SCSI_EH_SESSION_RST: S
ession was dropped as a result of SCSI error recovery) state (3)Aug 16 16:47:49 ibs-mngr2 iscsid: connection15:0 is operational after recovery (1 attempts)
The following setup is used: Initiator is on a SuSE Linux Enterprise Server 11, SP3 with open-iscsi version 2.0.873-0.21.1. Latest patches installed.
Network card is a Intel 520X 10GBit (Intel Corporation 82599EB 10-Gigabit SFI/SFP+)
driver: ixgbe, version 3.11.33-k firmware version of the card 0x61c10001We also see the same behaviour with Emulex Corporation OneConnect 10Gb Cards. (Hewlett-Packard Company NC550SFP DualPort 10GbE Server Adapter)
The iscsi target is a Nimble CS260G appliance. One fiber cable is directly plugged to the linux system. (10Gb) and one copper cable (1Gb) for management only in a different subnet.
fio version and command to do the benchmark: fio version 2.1fio --name=job1 --rw=randread --size=8g -iodepth=8 --numjobs=12 > randreadoutput.log
Thanks in advance and regards Timo
smime.p7s
Description: S/MIME Kryptografische Unterschrift
