My apologies, I did not bother the first time with collection of any stats. I took the times it ran from the display at the end of the run, and the size of the qtree I was moving.

LOG: DUMP: 60892781 KB LOG: DUMP: DUMP IS DONE The transfer is complete. Elapsed time: 2 hours, 44 minutes, 33 seconds.

Then, using the above numbers I got about 6 MB/s (50 Mbits/sec)

---------------

On recommendation from some of you asking to get sysstat stats, I re-ran the test, however, this time I only picked a 30 GB qtree (Oracle dbf files). Here is the output from ndmpcopy again.

LOG: DUMP: 30502534 KB LOG: DUMP: DUMP IS DONE The transfer is complete. Elapsed time: 1 hours, 30 minutes, 11 seconds.

Using my previous calc, I get about 5,648 KB/s (about 5.5 MB/s)

The environment is: o 2 x F840 clustered o DOT 6.2.2 o 2 x GigE fibre (one primary, the other for partner fail-over)

Here is a portion of output from 'sysstat -x 1'. Note the CPU is not busy.

Q: permit me to ask again; does the ndmpcopy process (on the same head) use the loopback to do that or does it use the normal NIC ?

Thanks, George

CPU NFS CIFS HTTP Total Net kB/s Disk kB/s Tape kB/s Cache Cache CP CP Disk DAFS in out read write read write age hit time ty util 31% 1146 0 0 1146 319 314 4612 0 0 0 6 91% 0% - 9% 0 35% 1385 0 0 1385 774 10172 13240 0 0 0 6 95% 0% - 7% 0 30% 1286 0 0 1286 604 346 4376 24 0 0 6 91% 0% - 10% 0 31% 1250 0 0 1250 310 3019 7044 0 0 0 6 92% 0% - 9% 0 37% 1506 0 0 1506 573 13202 15916 0 0 0 6 95% 0% - 12% 0 55% 926 0 0 926 409 12351 20281 17574 0 0 6 99% 61% F 60% 0 51% 824 0 0 824 308 1874 14824 34444 0 0 6 97% 100% : 100% 0 39% 1771 0 0 1771 987 2325 6836 20192 0 0 6 94% 79% : 54% 0 34% 1927 0 0 1927 532 5313 8440 24 0 0 6 93% 0% - 7% 0 35% 1422 0 0 1422 399 2428 4460 0 0 0 6 92% 0% - 7% 0 40% 1508 0 0 1508 628 2306 4536 0 0 0 6 92% 0% - 8% 0 35% 1431 0 0 1431 434 5428 6948 24 0 0 6 93% 0% - 9% 0 32% 1241 0 0 1241 377 1804 6548 0 0 0 6 92% 0% - 9% 0 47% 668 0 0 668 557 294 6022 26396 0 0 6 98% 63% F 53% 0 42% 422 0 0 422 305 1706 7276 40628 0 0 6 96% 100% : 84% 0 30% 1262 0 0 1262 482 1161 5272 124 0 0 6 92% 22% : 10% 0 32% 1413 0 0 1413 669 781 4644 0 0 0 6 91% 0% - 9% 0 33% 1278 0 0 1278 541 6848 10736 24 0 0 6 93% 0% - 9% 0 36% 1144 0 0 1144 486 18780 20128 0 0 0 6 95% 0% - 19% 0 17% 33 0 0 33 233 15 4028 0 0 0 6 89% 0% - 6% 0 CPU NFS CIFS HTTP Total Net kB/s Disk kB/s Tape kB/s Cache Cache CP CP Disk DAFS in out read write read write age hit time ty util 22% 75 0 0 75 184 13 4784 24 0 0 6 89% 0% - 10% 0 27% 215 0 0 215 640 5680 9637 9350 0 0 6 99% 35% F 28% 0 52% 1280 0 0 1280 598 8234 16459 38990 0 0 6 96% 100% : 100% 0 34% 1499 0 0 1499 611 8672 10712 22044 0 0 6 97% 87% : 53% 0 9% 779 0 0 779 330 2993 3672 8 0 0 6 95% 0% - 4% 0 16% 152 0 0 152 295 3782 6008 0 0 0 6 93% 0% - 6% 0 8% 72 0 0 72 144 1688 3652 24 0 0 6 92% 0% - 4% 0 7% 26 0 0 26 102 39 1672 0 0 0 6 89% 0% - 4% 0 13% 105 0 0 105 304 1560 4136 0 0 0 6 91% 0% - 5% 0 13% 178 0 0 178 450 3450 5588 24 0 0 6 93% 0% - 6% 0 29% 835 0 0 835 234 19530 20620 0 0 0 6 98% 0% - 26% 0 27% 668 0 0 668 410 12298 13272 0 0 0 6 96% 0% - 26% 0 20% 440 0 0 440 204 7046 9592 24 0 0 6 95% 0% - 22% 0 26% 407 0 0 407 105 7474 10317 0 0 0 6 98% 8% T 22% 0 45% 248 0 0 248 253 6014 16320 33970 0 0 6 99% 100% : 96% 0 28% 31 0 0 31 165 140 5852 25700 0 0 6 89% 79% : 54% 0 12% 109 0 0 109 124 3036 4260 0 0 0 6 92% 0% - 5% 0 16% 115 0 0 115 203 2048 4872 0 0 0 6 91% 0% - 6% 0 13% 30 0 0 30 185 9 3008 24 0 0 6 89% 0% - 5% 0 4% 51 0 0 51 71 845 2540 0 0 0 6 93% 0% - 3% 0

On Thu, 30 Oct 2003 16:10:02 +0100, Phil Hill phill@ultra.de wrote:

Hi Gerorge,

is it possible that you see 50 Mbytes/sec which is ca. 180 GB/h.

The "sysstat 1" will give you kbytes.

Please post a bit of your "sysstat 1" while the ndmpcopy is running.

Phil

George Kahler wrote:

...

Hi, when I ndmpcopy from say, volA to volB on the same head does ndmpcopy use the loopback to do this ?

If so, at what speed can the loopback run at ? I can not seem to go faster then about 20 GB/hr (about 50 Mbits/sec).

I do have GigE NICs in the filer for regular nfs connections to clients.

Thanks, George

---

George Kahler e-mail: george@yorku.ca Sr. Systems Administrator humans: (416) 736-2100 x.22699 Computing and Network Services machines: (416) 736-5830 Ontario, Canada, M3J-1P3

-- Philipp Hill Technischer Dienst Ultra Consulting Network GmbH Tel: 02 12 - 23 11 029 Schimmelbuschweg 14 Fax: 02 12 - 31 47 73 42719 Solingen Mobil: 01 73 - 27 54 934 Web: www.ultra.de

Hi all

this mail to let you know about my experience :

i used ndmpcopy (unix/linux utility) from a linux machine to start the ndmpcopy operation on a single head head : F740 DOT 613r2 data : around 200 Giga

it takes roughlty 13 hours to make a level 0 ndmp copy in sysstat we were able to see : net in 5000KB/s net out 5000KB/s and CPU beetween 90 and 100% at most active moments

so here are my conclusions : - ndmpcopy on a single head consume lot of CPU - ndmpcopy use the loopback device and is limited to 36G/hr - as the flow is in + out on a single head , there is an overall copy of 15G/hr - using two head can improve the time to copy as CPU would be less solicited (as disks) and NIC better used (full 36G/hr at least)

* my wish would be to test an ndmpcopy throught two Giga interface plugged in the same Filer, i am pretty sure we can get better time in this configuration i am still waiting for this ocasion asa I can test this , i will let you know that

Lance R Bailey wrote:

...

Q: permit me to ask again; does the ndmpcopy process (on the same head) use the loopback to do that or does it use the normal NIC ?

my experience is that it uses the nic. i'd *love* to be told how to get it to stay on the bus and use the loopback...