[lustre-discuss] No read throughput shown for the sequential read write Filebench workload

Andreas Dilger adilger at whamcloud.com
Sat Oct 16 00:49:47 PDT 2021


I would guess that all of the reads are handled from the client page cache?

Cheers, Andreas

On Oct 13, 2021, at 06:37, Md Hasanur Rashid via lustre-discuss <lustre-discuss at lists.lustre.org> wrote:


Hello Everyone,


I am running a Filebench workload which is provided below:

define fileset name="testF",entries=100,filesize=16m,prealloc,path="/mnt/hasanfs/tmp1"

define process name="readerP",instances=2 {
  thread name="readerT",instances=4 {
  flowop openfile name="openOP",filesetname="testF"
  flowop writewholefile name="writeOP",iters=4,filesetname="testF"
  flowop readwholefile name="readOP",iters=1,filesetname="testF"
  flowop closefile name="closeOP"
  }
}

create files
system "sync"
system "echo 3 > /proc/sys/vm/drop_caches"

run 60


I am running the workload in a Lustre cluster. When I check the log from the server-side, it shows the following (I have provided stats of one OSS):

obdfilter.hasanfs-OST0000.stats=
snapshot_time             1633361244.519122 secs.usecs
read_bytes                1 samples [bytes] 4096 4096 4096
write_bytes               8479 samples [bytes] 1048576 1048576 8890875904
destroy                   13 samples [reqs]
statfs                    76 samples [reqs]
preprw                    8480 samples [reqs]
commitrw                  8480 samples [reqs]
ping                      57 samples [reqs]

obdfilter.hasanfs-OST0000.brw_stats=
snapshot_time:         1633361244.519588 (secs.usecs)

                           read      |     write
pages per bulk r/w     rpcs  % cum % |  rpcs        % cum %
256:                     0   0   0   | 8479 100 100

                           read      |     write
discontiguous pages    rpcs  % cum % |  rpcs        % cum %
0:                       0   0   0   | 8479 100 100

                           read      |     write
discontiguous blocks   rpcs  % cum % |  rpcs        % cum %
0:                       0   0   0   | 8479 100 100

                           read      |     write
disk fragmented I/Os   ios   % cum % |  ios         % cum %
1:                       0   0   0   | 7499  88  88
2:                       0   0   0   |  980  11 100

                           read      |     write
disk I/Os in flight    ios   % cum % |  ios         % cum %
1:                       0   0   0   | 7921  83  83
2:                       0   0   0   | 1407  14  98
3:                       0   0   0   |  116   1  99
4:                       0   0   0   |   15   0 100

                           read      |     write
I/O time (1/1000s)     ios   % cum % |  ios         % cum %
2:                       0   0   0   | 1963  23  23
4:                       0   0   0   | 5627  66  89
8:                       0   0   0   |  837   9  99
16:                      0   0   0   |   25   0  99
32:                      0   0   0   |   27   0 100

                           read      |     write
disk I/O size          ios   % cum % |  ios         % cum %
4K:                      0   0   0   |   62   0   0
8K:                      0   0   0   |  127   1   1
16K:                     0   0   0   |  113   1   3
32K:                     0   0   0   |    0   0   3
64K:                     0   0   0   |    0   0   3
128K:                    0   0   0   |   65   0   3
256K:                    0   0   0   |    0   0   3
512K:                    0   0   0   | 1127  11  15
1M:                      0   0   0   | 7965  84 100


Can anyone please explain to me why I am not seeing any read operations in the stats?

Thanks,
Md. Hasanur Rashid

_______________________________________________
lustre-discuss mailing list
lustre-discuss at lists.lustre.org
http://lists.lustre.org/listinfo.cgi/lustre-discuss-lustre.org
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.lustre.org/pipermail/lustre-discuss-lustre.org/attachments/20211016/87633a2a/attachment-0001.html>


More information about the lustre-discuss mailing list