SYSTEM DOWNTIME!!!

[wurst]

hi, im gona put te boxes into another shelf, they will be off today in the evening from 19pm cest to ~21h...

[wurst]

now is up again!
good bye copper, good morning sunshine! (:

[bky]

good morning sunshine! (:

Watch this vid @ Youtube (new Window)

[wurst]

sunshine is now kinda inside those sunshine-colored wires. (:

[wurst]

ok, next downtime will come soon...

heres the plan:
whenever i had something to do with storage, it came out that performance was poor in the end.
but separate storage is so nice to have.
to devide the hypervisors and rest-pc from the disks, it makes everything much more flexible and fail-over resistant
ofc ONLY when storage runs and performs, and theres the problem:
they cost like nuclear weapons.
and they perform like nuclear waste.
dont believe theres spare parts that you can easily buy.
so its time for a test i would say.

im planning to run NFS, thats imple and I can access files.
Since I run VMware, its ok for me to have non block devices as Storage.
Important is whats behind, I want to run it soft.
Theres MD devices, XFS, ZFS, just to mention some.

Now i got an old Dell r730xd in my hands, think that should be a great base.
its equipped with small CPUs, delivering only 4GT/S CPU Interconnect.
Anyhow, that was cheap in the bay, additionally the 730xd had 2 16x full featured PCIe V3 Slots.
Two of them shall be equipped with cheap delock 89835 cards, the board supports PCIe Bifurcation.
This means 8 NVMe Drives, all powered by separate Lanes, 4 each connected to separate Processors.
Thats the thing what I want to test:
How does a bunch of NVMe Drives behave in that constellation.
Sequential RW should be around 20GB/Sec, i expect to see much more then a million IOPS @ 4K random.

For the Rest...
The System 64GB DDR4 for Cache, thought that would be nice.
A PERC 730 will connect the 24-bay front where all the old Disks from the other servers will rest in peace.

Im having just gigabit switches, but they are redundant and have plenty of ports.
Its 2x ex4200 VC with PCIe Backplane.
The Hpervisors are old max pimped HP 385p G7/8, they will need 3x NC364T, this bings 16 Links per box, cheap card... (:
Juniper can have up to 8 physical Link per aggregated interface, so they will be devided in 8x (storage) 6x (LAN) 2x (WAN)
That would bring just 1GB/Sec to the Hypervisor, even if its 8 Cables, anyhow, im curious for the IOPS capability of such a cheap setup.
FibreChannel 8GBit/s is comparable by speed, it stops @ 120K IOPS by design.
However, i like the idea of investing in ethernet switch redundancy, using L3 for everything from Disk to WAN traffic.

Hm.
So there will be some downtime soon, dunno when exactly.
Ill have to stick some NICs in teh Servers so they look kinda hairy after.