Fast .. and then slow....
by Matheson, Edward B. (LARC-D318)[LITES]
First: kickass tool .. love it!
Second: weird problem.
- I have ~320 blades connected to 32 bladeswitches .. and then trunked to a cisco 4503.
- Each bladeswitch has 4 connections to the cisco .. port-channeled and trunked.
- I found with much trouble.. that one MUST force duplex full, speed 1000 .. made a big difference
- however, I have a weird problem where i start around 230Mbps .. and end about 50Mbps on transfer to about 64 blades.
Not sure why. It stays nice and fast thru about the first 12gig .. but slows down from there until I hit the total of about 26 gig.
I am using pigz ( nice way to make this faster ) thusly:
append load_ramdisk=1 dhcp=yes port=9002 umode=snd auto=yes lang=US kbmap=US initrd=/udpcast/initrd root=01:00 compr=lzop netmodule=AUTO netmodparm= enableDiskmodule=yes diskmodule=ahci diskmodparm= disk=/dev/cciss!c0d0 udpcparam="--pipe /pigz -c --full-duplex"
( the disk is because these blades have weird controllers .. darn hp )
Anyone know why it would slow down over time? It's almost asymptotic.
PS > I am working on a few patches ..
- to incorporate pigz from the makeImage ..
- to use partedImage to read/write to the image on the drive. ( All my machines are 1 partition )
E Branson Matheson
ATOL Experiment Specialist