Výkon iSCSI přes ethernet, optimalizace

[Fox!MURDER]

...

takhle by to bylo, pokud by nad tim nikdo vubec nepremejslel a udelal to nejjednodussim zpusobem ...

jinak ale staci obetovat trochu pameti na buffery a neni problem ...

viz. zapis na

Cipher name:    aes
Cipher mode:    xts-essiv:sha256
Hash spec:      sha512

Device:         rrqm/s   wrqm/s     r/s     w/s    rMB/s    wMB/s avgrq-sz avgqu-sz   await r_await w_await  svctm  %util
sdc               0.00     0.00    0.00  475.00     0.00   211.21   910.64     0.26    0.55    0.00    0.55   0.47  22.50
dm-0              0.00     0.00    0.00  475.00     0.00   211.21   910.64     0.22    0.46    0.00    0.46   0.39  18.70
sdc               0.00     0.00    0.00  557.00     0.00   247.74   910.91     0.30    0.54    0.00    0.54   0.45  25.10
dm-0              0.00     0.00    0.00  557.00     0.00   247.74   910.91     0.26    0.46    0.00    0.46   0.39  21.60
sdc               0.00     0.00    0.00  495.00     0.00   220.21   911.11     0.28    0.56    0.00    0.56   0.47  23.30
dm-0              0.00     0.00    0.00  495.00     0.00   220.21   911.11     0.23    0.47    0.00    0.47   0.40  19.80

----total-cpu-usage---- ---system-- ---io/sdc-----io/dm-0-- --dsk/sdc-----dsk/dm-0-
usr sys idl wai hiq siq| int   csw | read  writ: read  writ| read  writ: read  writ
  0  83   0  15   0   3|4592  3045 |   0   507 :   0   507 |   0   225M:   0   225M
  0  96   0   2   0   2|4728  1249 |   0   619 :   0   620 |   0   275M:   0   275M
  0  78   0  21   0   1|5363   881 |   0   486 :   0   488 |   0   217M:   0   217M
  0  78   0  21   0   1|5379  1033 |   0   479 :   0   477 |   0   212M:   0   213M

context switchu tam moc neni a preruseni taky ne (vzhledem k tomu, ze 4000 je klidovej stav pri 4 CPU a HZ=1000 kernelu)
HW: 1x Xeon E5504, Areca 1680, 5x2TB WD2002FYPS RAID6, 6GB RAM

(samozrejme se jedna o velmi jednoduchej test-case - sekvencni zapis na fs)
(ty vypisy jsou chycene v rozdilnych casech - jsou spis informativni, nez aby mely velkou vypovidaci hodnotu)
(zkusim jeste najit nezatizenej sw raid5/6 ...)

A nemuzu ted narychlo najit benchmark kernelove implementace AES256-XTS, ale pro obrazek rychlost OpenSSL implementace AES-256-CBC pri vyuziti vsech 4 jader (kernel ma AES implementovany v assembleru a tim padem o neco rychlejsi).

The 'numbers' are in 1000s of bytes per second processed.
type             16 bytes     64 bytes    256 bytes   1024 bytes   8192 bytes
aes-256 cbc     171594.18k   179896.87k   182058.07k   182586.71k   182591.49k

btw. marty - nahod komplet konfigurak ietd ...