Sign Up
Log In
Log In
or
Sign Up
Places
All Projects
Status Monitor
Collapse sidebar
openSUSE:Leap:15.5:Update
s390-tools.23971
s390-tools-sles15sp3-ziomon-correct-throughput-...
Overview
Repositories
Revisions
Requests
Users
Attributes
Meta
File s390-tools-sles15sp3-ziomon-correct-throughput-calculation-in-ziorep_prin.patch of Package s390-tools.23971
Subject: [PATCH] [BZ 195999] ziomon: correct throughput calculation in ziorep_printers.cpp From: Fedor Loshakov <loshakov@linux.ibm.com> Description: ziomon: correct throughput calculation in ziorep_printers.cpp Symptom: During evaluation of the collected ziomon data from the system, there were low throughput values for adapter noticed in ziorep_utilization virtual adapter report. There were also low values for throughput of devices noticed in ziorep_traffic report. And were no aggregation of throughput for different aggregation parameters provided. Although, iostat tool showed decent throughput values for each scsi disk, which belongs to investigated zfcp adapter. Problem: During throughput calculation, there were sum of sizes of all read and write requests divided by average dispatch to complete time (d2c) of all read and write requests respectively. This is not correct, because the resulted value represents physical measure other than throughput. In order to get correct throughput physical measure sizes of all read and write requests should be devided on time interval, during which those requests were started and finished. This problem can be observed in ziorep_utilization virtual adapter report for zfcp adapter and also in ziorep_traffic report for each separate device for different aggregation parameters. Solution: Use time interval, during which read and write requests were started and finished, for calculation of throughput of zfcp adapter instead of d2c time of read and write requests. Apply this approach for ziorep_utilization virtual adapter report and for ziorep_traffic report. Reproduction: Create setup, which includes at least two zfcp adapters, which are both connected to the same SCSI disks, so that there is a number of multipath devices created. Launch FIO job as a I/O workload against those multipath devices. During this FIO workload lauch ziomon tool to collect ziomon data. After finishing of FIO job and ziomon tool work, evaluate ziomon results with using of ziorep_utilization and ziorep_traffic tools. Compare throughput values from ziorep_* tools with ones from FIO log or from iostat metric. Upstream-ID: f8331a4b8e88e77c8f48f7a2c25fc7aee52b333c Problem-ID: 195999 Upstream-Description: ziomon: correct throughput calculation in ziorep_printers.cpp During evaluation of the collected ziomon data from the system with following multipath configuration: $ multipath -ll 3600507630bffc3200000000000005294 dm-0 IBM,2107900 size=20G features='1 queue_if_no_path' hwhandler='1 alua' wp=rw `-+- policy='service-time 0' prio=50 status=active |- 1:0:0:1083457618 sdd 8:48 active ready running `- 0:0:0:1083457618 sda 8:0 active ready running 3600507630bffc3200000000000005394 dm-1 IBM,2107900 size=20G features='1 queue_if_no_path' hwhandler='1 alua' wp=rw `-+- policy='service-time 0' prio=50 status=active |- 1:0:0:1083457619 sdf 8:80 active ready running `- 0:0:0:1083457619 sdb 8:16 active ready running 3600507630bffc3200000000000005494 dm-2 IBM,2107900 size=20G features='1 queue_if_no_path' hwhandler='1 alua' wp=rw `-+- policy='service-time 0' prio=50 status=active |- 1:0:0:1083457620 sdg 8:96 active ready running `- 0:0:0:1083457620 sdc 8:32 active ready running 3600507630bffc3200000000000005594 dm-3 IBM,2107900 size=20G features='1 queue_if_no_path' hwhandler='1 alua' wp=rw `-+- policy='service-time 0' prio=50 status=active |- 1:0:0:1083457621 sdh 8:112 active ready running `- 0:0:0:1083457621 sde 8:64 active ready running there were low throughput values for adapter noticed in ziorep_utilization virtual adapter report: $ ziorep_utilization adp_line_speed.log ... CHP Bus-ID |qdio util.%|queu|fail|-thp in MB/s-|I/O reqs-| ID max avg full erc rd wrt rd wrt 2021-08-18 12:56:44 61/0.0.1946 32.8 3.1 0 0 0.0 84.5 21 204K 12:57:04 ... there were also low values for throughput of devices noticed in ziorep_traffic report. And were no aggregation of throughput for different aggregation parameters provided: $ ziorep_traffic adp_line_speed.log -i 0 WWPN LUN |I/O rt MB/s|thrp in MB/s-|... min max avg stdev ... 2021-08-18 12:58:24 0x500507630b09c320:0x4052409400000000 0.0 219.2 85.7 3.352K ... 0x500507630b09c320:0x4053409400000000 0.0 348.6 84.6 3.330K ... 0x500507630b09c320:0x4054409400000000 0.0 268.6 84.0 3.317K ... 0x500507630b09c320:0x4055409400000000 0.0 354.2 83.7 3.312K ... $ ziorep_traffic adp_line_speed.log -i 0 -Cu Bus-ID |I/O rt MB/s|thrp in MB/s-|... min max avg stdev ... 2021-08-18 12:58:24 0.0.1946 0.0 354.2 84.5 3.328K ... although, iostat tool showed decent throughput values for each scsi disk, which belongs to investigated zfcp adapter. In this example each scsi disk has throughput more than 300MB/s, which is much larger, than 84.5MB/s throughput for the adapter: $ iostat -x 1 ... Device r/s w/s rkB/s wkB/s ... dasda 0.00 0.00 0.00 0.00 ... dasdb 0.00 0.00 0.00 0.00 ... sda 0.00 2556.00 0.00 327168.00 ... sdc 0.00 2554.00 0.00 326912.00 ... sdb 0.00 2559.00 0.00 327552.00 ... sdf 0.00 2865.00 0.00 366720.00 ... sdd 0.00 2843.00 0.00 363904.00 ... sdg 0.00 2845.00 0.00 364160.00 ... sde 0.00 2471.00 0.00 315416.00 ... sdh 0.00 2768.00 0.00 353768.00 ... dm-0 0.00 5399.00 0.00 691072.00 ... dm-1 0.00 5424.00 0.00 694272.00 ... dm-2 0.00 5400.00 0.00 691200.00 ... dm-3 0.00 5240.00 0.00 669312.00 ... Use Frameset interval size for calculation of throughput of zfcp adapter instead of d2c (dispatch to complete) time of request, which is used for I/O rate calculation. Also use Frameset interval instead of total_latency.sum (which in fact is a sum of d2c times for each repuest in specifed interval). Use calc_avg() function for throughput calculation. With fix applied, virtual adapter report now contains correct value for adpater throughput for specified interval: $ ziorep_utilization adp_line_speed.log ... CHP Bus-ID |qdio util.%|queu|fail|-thp in MB/s-|I/O reqs-| ID max avg full erc rd wrt rd wrt 2021-08-18 12:56:44 61/0.0.1946 32.8 3.1 0 0 0.0 1.336K 21 204K ... With fix applied, traffic report now contains correct values of throughput for each device and aggregation now works correctly: $ ziorep_traffic adp_line_speed.log -i 0 WWPN LUN |I/O rt MB/s|thrp in MB/s-|... min max avg stdev ... 2021-08-18 12:58:24 0x500507630b09c320:0x4052409400000000 0.0 219.2 339.6 3.352K ... 0x500507630b09c320:0x4053409400000000 0.0 348.6 335.4 3.330K ... 0x500507630b09c320:0x4054409400000000 0.0 268.6 333.1 3.317K ... 0x500507630b09c320:0x4055409400000000 0.0 354.2 331.9 3.312K ... $ ziorep_traffic adp_line_speed.log -i 0 -Cu Bus-ID |I/O rt MB/s|thrp in MB/s-|... min max avg stdev ... 2021-08-18 12:58:24 0.0.1946 0.0 354.2 1.340K 3.328K ... Signed-off-by: Fedor Loshakov <loshakov@linux.ibm.com> Reviewed-by: Benjamin Block <bblock@linux.ibm.com> Signed-off-by: Jan Hoeppner <hoeppner@linux.ibm.com> Signed-off-by: Fedor Loshakov <loshakov@linux.ibm.com> --- ziomon/ziorep_printers.cpp | 44 ++++++++++++++++++++++++++++++-------------- ziomon/ziorep_printers.hpp | 15 ++++++++++----- 2 files changed, 40 insertions(+), 19 deletions(-) --- a/ziomon/ziorep_printers.cpp +++ b/ziomon/ziorep_printers.cpp @@ -478,21 +478,22 @@ void VirtAdapterPrinter::print_failures( } void VirtAdapterPrinter::print_throughput(FILE *fp, - const struct blkiomon_stat *stat) + const struct blkiomon_stat *stat, + const __u64 interval) { double tmp; - if (!stat || stat->d2c_r.num <= 0 || stat->size_r.num <= 0) + if (!stat || stat->size_r.num <= 0) tmp = 0; else - tmp = stat->size_r.sum/(double)stat->d2c_r.sum; + tmp = calc_avg(stat->size_r.sum,interval); print_delimiter(fp); print_abbrev_num(fp, tmp); - if (!stat || stat->d2c_w.num <= 0 || stat->size_w.num <= 0) + if (!stat || stat->size_w.num <= 0) tmp = 0; else - tmp = stat->size_w.sum/(double)stat->d2c_w.sum; + tmp = calc_avg(stat->size_w.sum,interval); print_delimiter(fp); print_abbrev_num(fp, tmp); } @@ -536,6 +537,12 @@ int VirtAdapterPrinter::print_frame(FILE const struct blkiomon_stat *blk_stat; const struct zfcpdd_dstat *zfcp_stat; + /** + * Receive Frameset interval and convert it + * from seconds to microseconds + */ + __u64 interval = frameset.get_duration() * 1000000; + list<__u32> devnos; devnos = ((StagedDeviceFilter*)&dev_filt)->get_filter_devnos(); @@ -557,7 +564,7 @@ int VirtAdapterPrinter::print_frame(FILE print_queue_fill(fp, zfcp_stat, util); print_queue_full(fp, util); print_failures(fp, ioerr); - print_throughput(fp, blk_stat); + print_throughput(fp, blk_stat, interval); print_num_requests(fp, blk_stat); if (lrc) { fprintf(stderr, "%s: Did not find matching data in" @@ -810,6 +817,12 @@ int TrafficPrinter::print_frame(FILE *fp const struct zfcpdd_dstat *zfcp_stat = NULL; const AggregationCollapser *agg_col; + /** + * Receive Frameset interval and convert it + * from seconds to microseconds + */ + __u64 interval = frameset.get_duration() * 1000000; + switch (m_agg_crit) { case none: get_device_list(lst_32, dev_filt); @@ -847,7 +860,7 @@ int TrafficPrinter::print_frame(FILE *fp blk_stat = frameset.get_blkiomon_stat_by_wwpn(*i); zfcp_stat = frameset.get_zfcpdd_stat_by_wwpn(*i); print_device_wwpn(fp, *i); - print_data_row(fp, blk_stat, zfcp_stat); + print_data_row(fp, blk_stat, zfcp_stat, interval); } } else if (m_agg_crit == all) { @@ -856,7 +869,7 @@ int TrafficPrinter::print_frame(FILE *fp blk_stat = frameset.get_first_blkiomon_stat(); zfcp_stat = frameset.get_first_zfcpdd_stat(); print_device_all(fp); - print_data_row(fp, blk_stat, zfcp_stat); + print_data_row(fp, blk_stat, zfcp_stat, interval); } else { for (list<__u32>::const_iterator i = lst_32.begin(); @@ -894,7 +907,7 @@ int TrafficPrinter::print_frame(FILE *fp } if (rc ) return -1; - print_data_row(fp, blk_stat, zfcp_stat); + print_data_row(fp, blk_stat, zfcp_stat, interval); } } @@ -930,7 +943,8 @@ void SummaryTrafficPrinter::print_toplin } -void SummaryTrafficPrinter::print_throughput(FILE *fp, const struct blkiomon_stat *stat) +void SummaryTrafficPrinter::print_throughput(FILE *fp, const struct blkiomon_stat *stat, + const __u64 interval) { struct minmax thrp_data, total_size, total_latency; double tmp; @@ -968,7 +982,7 @@ void SummaryTrafficPrinter::print_throug tmp = 0; if (stat && total_size.sum > 0) - tmp = calc_avg(total_size.sum, total_latency.sum); + tmp = calc_avg(total_size.sum, interval); print_delimiter(fp); print_abbrev_num(fp, tmp); @@ -1101,14 +1115,15 @@ void SummaryTrafficPrinter::print_fabric void SummaryTrafficPrinter::print_data_row(FILE *fp, const struct blkiomon_stat *blk_stat, - const struct zfcpdd_dstat *zfcp_stat) + const struct zfcpdd_dstat *zfcp_stat, + const __u64 interval) { if (!blk_stat) blk_stat = get_empty_blkiomon_stat(); if (!zfcp_stat) zfcp_stat = get_empty_zfcpdd_dstat(); - print_throughput(fp, blk_stat); + print_throughput(fp, blk_stat, interval); print_request_stats(fp, blk_stat); print_io_subsystem_latency(fp, blk_stat); print_channel_latency(fp, zfcp_stat); @@ -1177,7 +1192,8 @@ void DetailedTrafficPrinter::print_topli void DetailedTrafficPrinter::print_data_row(FILE *fp, const struct blkiomon_stat *blk_stat, - const struct zfcpdd_dstat *zfcp_stat) + const struct zfcpdd_dstat *zfcp_stat, + const __u64 interval __attribute__ ((unused))) { if (!blk_stat) blk_stat = get_empty_blkiomon_stat(); --- a/ziomon/ziorep_printers.hpp +++ b/ziomon/ziorep_printers.hpp @@ -128,7 +128,8 @@ private: const struct adapter_utilization *res); void print_queue_full(FILE *fp, const struct adapter_utilization *res); void print_failures(FILE *fp, const struct ioerr_cnt *cnt); - void print_throughput(FILE *fp, const struct blkiomon_stat *stat); + void print_throughput(FILE *fp, const struct blkiomon_stat *stat, + const __u64 interval); void print_num_requests(FILE *fp, const struct blkiomon_stat *stat); }; @@ -147,7 +148,8 @@ protected: * Print the actual row, excluding the first column */ virtual void print_data_row(FILE *fp, const struct blkiomon_stat *blk_stat, - const struct zfcpdd_dstat *zfcp_stat) = 0; + const struct zfcpdd_dstat *zfcp_stat, + const __u64 interval) = 0; void print_topline_prefix1(FILE *fp); void print_topline_prefix2(FILE *fp); @@ -185,8 +187,10 @@ public: private: virtual void print_data_row(FILE *fp, const struct blkiomon_stat *blk_stat, - const struct zfcpdd_dstat *zfcp_stat); - void print_throughput(FILE *fp, const struct blkiomon_stat *stat); + const struct zfcpdd_dstat *zfcp_stat, + const __u64 interval); + void print_throughput(FILE *fp, const struct blkiomon_stat *stat, + const __u64 interval); void print_request_stats(FILE *fp, const struct blkiomon_stat *stat); void print_io_subsystem_latency(FILE *fp, const struct blkiomon_stat *stat); void print_channel_latency(FILE *fp, const struct zfcpdd_dstat *stat); @@ -207,7 +211,8 @@ public: private: virtual void print_data_row(FILE *fp, const struct blkiomon_stat *blk_stat, - const struct zfcpdd_dstat *zfcp_stat); + const struct zfcpdd_dstat *zfcp_stat, + const __u64 interval __attribute__ ((unused))); void print_histogram_io_reqs(FILE *fp, const struct blkiomon_stat *stat); void print_histogram_io_subs_lat(FILE *fp,
Locations
Projects
Search
Status Monitor
Help
OpenBuildService.org
Documentation
API Documentation
Code of Conduct
Contact
Support
@OBShq
Terms
openSUSE Build Service is sponsored by
The Open Build Service is an
openSUSE project
.
Sign Up
Log In
Places
Places
All Projects
Status Monitor
