Re: Antw: Re: Q: Slow extraordinarily slow performance of "dmraid -s -c -c -c"
It might be due to retrying what something (initiator or target) thinks is a bad IO. On 11/05/2012 06:47 AM, Ulrich Windl wrote: > Hi! > > I have an update on the "ioctl() takes 16 seconds" issue: > > It seems (don't ask me for an explanation) if the buffer size for the SCSI > response is too small (in my case less than 18 octets), the wsystemcall to > read the SCSI serial number takes 15-16 seconds, while with a buffer of 18 > octets or more, it takes 3ms (over an iSCSI-FC-Gateway). > > Maybe Mike finds this output of a test program useful: > === > buflen = 17 > testprog /dev/sdu 17 > SCSI cmd sent: 12 1 80 0 11 0 > ioctl duration: 15976 msecs > SCSI response: 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 > resplen = 17, pgcode = 0x 0, pglen = 0 > > /dev/sdu serial no: > : > real0m15.979s > user0m0.000s > sys 0m0.000s > === > buflen = 18 > testprog /dev/sdu 18 > SCSI cmd sent: 12 1 80 0 12 0 > ioctl duration: 0 msecs > SCSI response: 0 80 0 e 50 42 35 41 38 44 34 41 41 55 36 38 33 39 > resplen = 18, pgcode = 0x80, pglen = 14 > > /dev/sdu serial no: PB5A8D4AAU6839 > : > real0m0.003s > user0m0.000s > sys 0m0.000s > === > > Regards, > Ulrich > Ulrich Windl schrieb am 24.09.2012 um 08:57 in Nachricht <50600460.815 : 161 : > 60728>: > Michael Christie schrieb am 24.09.2012 um 04:46 in >> Nachricht <1f8fae71-ee7c-4785-a648-24758b18e...@cs.wisc.edu>: >>> From what I can tell, it is a dm raid issue or a target issue. >>> >>> The len dm raid is using might not be right, but I am not sure. I do not >>> know why the target does not like it. >>> >>> >>> On Sep 21, 2012, at 5:59 AM, Ulrich Windl >>> >>> wrote > # time sg_inq -e -p 80- /dev/sdc > unrecognized multiplier > Bad argument to '--page=', expecting 0 to 255 inclusive > > Modified: > # time sg_inq -e -p 80 /dev/sdc >>> >>> I goofed when I wrote the example. You need to pass it in hex so it should >>> be >>> >>> sg_inq -e -p 0x80 /dev/sdc >>> >>> I think that will end up working ok though. sg_inq seems to use a different >>> len than what dm raid does. >> >> Hi Mike! >> >> Thanks for that. Still, that command responds as quick as possible, so maybe >> it's actually more dmraid than iSCSI. >> >> # time sg_inq -e -p 0x80 /dev/sdc >> VPD INQUIRY: Unit serial number page >> Unit serial number: PB5A8D3AATZBSH >> >> real0m0.025s >> user0m0.000s >> sys 0m0.000s >> >> Regards, >> 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-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?hl=en.
Re: Antw: Re: Q: Slow extraordinarily slow performance of "dmraid -s -c -c -c"
Hi! I have an update on the "ioctl() takes 16 seconds" issue: It seems (don't ask me for an explanation) if the buffer size for the SCSI response is too small (in my case less than 18 octets), the wsystemcall to read the SCSI serial number takes 15-16 seconds, while with a buffer of 18 octets or more, it takes 3ms (over an iSCSI-FC-Gateway). Maybe Mike finds this output of a test program useful: === buflen = 17 testprog /dev/sdu 17 SCSI cmd sent: 12 1 80 0 11 0 ioctl duration: 15976 msecs SCSI response: 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 resplen = 17, pgcode = 0x 0, pglen = 0 /dev/sdu serial no: : real0m15.979s user0m0.000s sys 0m0.000s === buflen = 18 testprog /dev/sdu 18 SCSI cmd sent: 12 1 80 0 12 0 ioctl duration: 0 msecs SCSI response: 0 80 0 e 50 42 35 41 38 44 34 41 41 55 36 38 33 39 resplen = 18, pgcode = 0x80, pglen = 14 /dev/sdu serial no: PB5A8D4AAU6839 : real0m0.003s user0m0.000s sys 0m0.000s === Regards, Ulrich >>> Ulrich Windl schrieb am 24.09.2012 um 08:57 in Nachricht <50600460.815 : >>> 161 : 60728>: > >>> Michael Christie schrieb am 24.09.2012 um 04:46 in > Nachricht <1f8fae71-ee7c-4785-a648-24758b18e...@cs.wisc.edu>: > > From what I can tell, it is a dm raid issue or a target issue. > > > > The len dm raid is using might not be right, but I am not sure. I do not > > know why the target does not like it. > > > > > > On Sep 21, 2012, at 5:59 AM, Ulrich Windl > > > > wrote > > >> # time sg_inq -e -p 80- /dev/sdc > > >> unrecognized multiplier > > >> Bad argument to '--page=', expecting 0 to 255 inclusive > > >> > > >> Modified: > > >> # time sg_inq -e -p 80 /dev/sdc > > > > I goofed when I wrote the example. You need to pass it in hex so it should > > be > > > > sg_inq -e -p 0x80 /dev/sdc > > > > I think that will end up working ok though. sg_inq seems to use a different > > len than what dm raid does. > > Hi Mike! > > Thanks for that. Still, that command responds as quick as possible, so maybe > it's actually more dmraid than iSCSI. > > # time sg_inq -e -p 0x80 /dev/sdc > VPD INQUIRY: Unit serial number page > Unit serial number: PB5A8D3AATZBSH > > real0m0.025s > user0m0.000s > sys 0m0.000s > > Regards, > 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-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?hl=en.
Re: Antw: Re: Q: Slow extraordinarily slow performance of "dmraid -s -c -c -c"
>>> Michael Christie schrieb am 24.09.2012 um 04:46 in Nachricht <1f8fae71-ee7c-4785-a648-24758b18e...@cs.wisc.edu>: > From what I can tell, it is a dm raid issue or a target issue. > > The len dm raid is using might not be right, but I am not sure. I do not > know why the target does not like it. > > > On Sep 21, 2012, at 5:59 AM, Ulrich Windl > wrote > >> # time sg_inq -e -p 80- /dev/sdc > >> unrecognized multiplier > >> Bad argument to '--page=', expecting 0 to 255 inclusive > >> > >> Modified: > >> # time sg_inq -e -p 80 /dev/sdc > > I goofed when I wrote the example. You need to pass it in hex so it should > be > > sg_inq -e -p 0x80 /dev/sdc > > I think that will end up working ok though. sg_inq seems to use a different > len than what dm raid does. Hi Mike! Thanks for that. Still, that command responds as quick as possible, so maybe it's actually more dmraid than iSCSI. # time sg_inq -e -p 0x80 /dev/sdc VPD INQUIRY: Unit serial number page Unit serial number: PB5A8D3AATZBSH real0m0.025s user0m0.000s sys 0m0.000s Regards, 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-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?hl=en.
Re: Antw: Re: Q: Slow extraordinarily slow performance of "dmraid -s -c -c -c"
>From what I can tell, it is a dm raid issue or a target issue. The len dm raid is using might not be right, but I am not sure. I do not know why the target does not like it. On Sep 21, 2012, at 5:59 AM, Ulrich Windl wrote >> # time sg_inq -e -p 80- /dev/sdc >> unrecognized multiplier >> Bad argument to '--page=', expecting 0 to 255 inclusive >> >> Modified: >> # time sg_inq -e -p 80 /dev/sdc I goofed when I wrote the example. You need to pass it in hex so it should be sg_inq -e -p 0x80 /dev/sdc I think that will end up working ok though. sg_inq seems to use a different len than what dm raid does. -- 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?hl=en.
Re: Antw: Re: Q: Slow extraordinarily slow performance of "dmraid -s -c -c -c"
>> >> sg_inq -E /dev/sdc > > # time sg_inq -E /dev/sdc > VPD INQUIRY: extended INQUIRY data page > SPT=0 GRD_CHK=0 APP_CHK=0 REF_CHK=0 > UASK_SUP=0 GROUP_SUP=0 PRIOR_SUP=0 HEADSUP=0 ORDSUP=0 SIMPSUP=0 > WU_SUP=0 CRD_SUP=0 NV_SUP=1 V_SUP=1 > P_I_I_SUP=0 LUICLR=0 CBCS=0 R_SUP=0 > Multi I_T nexus microcode download=0 > Extended self-test completion minutes=0 > > real0m0.004s > user0m0.000s > sys 0m0.000s > > >> or >> sg_inq -e -p 80- /dev/sdc >> does it take a long time and what is the output? > > # time sg_inq -e -p 80- /dev/sdc > unrecognized multiplier > Bad argument to '--page=', expecting 0 to 255 inclusive > > Modified: > # time sg_inq -e -p 80 /dev/sdc > Only hex output supported. sg_vpd decodes more pages. > VPD INQUIRY, page code=0x50: >inquiry: field in cdb illegal (page not supported) > > real0m0.006s > user0m0.000s > sys 0m0.000s > > So here the response is "immediate". Maybe a bug in dmraid... To make sure do a: "tcpdump -w iscsi.out -i your-eth-dev" and send the iscsi.out file. We can see if the target is sending a response quickly or if the iscsi layer is dropping something. -- 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?hl=en.