ceph-osd: microbenchmark for 4k read operations

ceph-osd: microbenchmark for 4k read operations

[Andreas Bluemle]

[-- Attachment #1: Type: text/plain, Size: 655 bytes --]

Hi,

as discussed on todays performance meeting: find attached the
spreadsheet which I had shown during the meeting.

@Sam: your are right concerning the make_blist part: this
probably is expensive due to message signing being turned on.
For the osd daemon which I took the benchmarks, I see
    "cephx_sign_messages": "true"


Regards

Andreas

-- 
Andreas Bluemle                     mailto:Andreas.Bluemle@itxperts.de
ITXperts GmbH                       http://www.itxperts.de
Balanstrasse 73, Geb. 08            Phone: (+49) 89 89044917
D-81541 Muenchen (Germany)          Fax:   (+49) 89 89044910

Company details: http://www.itxperts.de/imprint.htm

[-- Attachment #2: summary-rr-transactions-32.csv --]
[-- Type: text/csv, Size: 999 bytes --]


summary; samples; averages; min; max

PrimaryRead; 165889; 186; 103; 655
PrimaryRead-1.1 Pipe::reader:duration_0; 165889; 64; 29; 597
PrimaryRead-1.3 Pipe::read_message:duration_0; 165889; 9; 4; 171
PrimaryRead-1.4 Pipe::read_message:receive_0; 165889; 2; 1; 133
PrimaryRead-1.5 Pipe::read_message:decode_0; 165889; 5; 2; 166
PrimaryRead-1.6 Pipe::read_message:authenticate_0; 165889; 0; 0; 102
PrimaryRead-1.8 OSD::ms_fast_dispatch:duration_0; 165889; 33; 17; 482

PrimaryRead-2.2 ShardedOpWQ::_process:duration; 165889; 66; 25; 419
PrimaryRead-2.2.2 ReplicatedPG::do_op:duration; 165889; 49; 20; 397
PrimaryRead-2.2.3 ReplicatedPG::do_op:find_ctx; 165889; 2; 2; 140
PrimaryRead-2.2.4 ReplicatedPG::do_op:execute_ctx; 165889; 38; 13; 388

PrimaryRead-4.2 Pipe::writer:duration_4096; 165888; 52; 24; 322
PrimaryRead-4.3 Pipe::writer:encode_4096; 165888; 6; 3; 172
PrimaryRead-4.4 Pipe::writer:make blist_4096; 165888; 23; 14; 193
PrimaryRead-4.5 Pipe::writer:write_message_4096; 165888; 20; 6; 268

Re: ceph-osd: microbenchmark for 4k read operations

[Mark Nelson]

On 05/20/2015 11:13 AM, Andreas Bluemle wrote:
> Hi,
>
> as discussed on todays performance meeting: find attached the
> spreadsheet which I had shown during the meeting.
>
> @Sam: your are right concerning the make_blist part: this
> probably is expensive due to message signing being turned on.
> For the osd daemon which I took the benchmarks, I see
>      "cephx_sign_messages": "true"
>
>
> Regards
>
> Andreas
>

Thank you Andreas!  Looking forward to hearing how much 
cephx_sign_messages is affecting things.

Mark