OI + Nappit read speed

[Freak1]

Hi guys.

I am having some trouble with the speed on my OI/Nappit All-In-One.

When i copy files to the box i get 95 mb/sec but when i copy from the box i only get 50 mb/sec.

I googled it and after that removed TeraCopy that got me from 30 mb/sec to 50 mb/sec.

I tried using both e1000g0 and vmxnet3s0

I also found that i could change settings on the network in OI to RX only what i get is"e1000g0 PERM=RW VALUE=no DEFAULT=bi" "vmxnet3s0 PERM=-- VALUE=-- DEFAULT=no"
Not sure on how to change it i tried dladm set-linkprop -p flowctrl=rx e1000g0 but nothing changed. What am I doing wrong?

 

[Freak1]

noone?

 

[_Gea]

noone?

In case of (speed) problems:
from http://www.napp-it.org/manuals/nas.html

-look at napp-it system - log and system statistics - (disk) (wait value of disks)
a lot of waits indicates a disk problem, look also at cpu
- try a local dd benchmark to check disk subsystem

on server side most problems are due to:
- badly supported mainboards (Use Intel server class mainboards, prefered SuperMicro)
- badly supported disc controller (Use LSI HBA ex LSI 9211 and compatibles)
- hardware problems (mostly power, cabling, disc or controller related)
look at startup-messages during boot (switch to console with tab-key on libe editions)
- RAM problems, use ECC to get informed
- Nic problems (best: use Intel Nics only)
- Bios problem (update bios)
- active state powermanagement enabled (disable in bios)
- update OS to newest (maybee you problem is solved then)

on client side
- most problems are due to some Nics (mostly Realtek)
update driver to newest (may help even when no problems with oher Windows servers, only with your ZFS NAS)
- compare another client, best with Intel Nics (mostly trouble free)
- check network, try to connect directly via crossover cable

other reasons/ solutions
- avoid copy tools like Teracopy etc. They may slow down speed with ZFS. Disable and compare to a basic copy
- Try a NAS test tool for sequential performance like http://www.808.dk/?code-csharp-nas-performance
- Prefer fast pool layouts like striped mirrors over slow ones with higher capacity like Raid-Z (1-3)
- check/replace your network cables (helps quite often) and network switches/ try to connect directly via crossover cable

- use more RAM
- avoid 4k disks
- disable dedup and compress

slow sync writes
- With slow writes especially on ESXi/NFS, disable sync-property and recheck performance.
If values are much better, add a ZIL SSD and re-enable sync.

 

[Freak1]

Thanks for the answer.

I did a benchmark i got 652 MB/s seq-read and 447 MB/s seq-write.

All cables are cat 6, it has 4 vcpu and 16GB ram in ESXi 5, I use ECC ram, Intel NIC on both ends.

Not really sure where to start

 

[Silenus]

When you are copying from the the box...where are you copying too?

Within another VM on the same box? An external destination? Are you certain your destination can actaully write fast enough and is not a bottleneck? What is your disk config?

For example, I have an OI system with a 9 disk RAIDz2. Is not an 'ideal' number of disks nor did I bother to properly align it. Still I can read from, OR write to it at max gigabit speed on large seqential transfers. But if I read from one share on it and write back to another share that is on the the same set of disks, it slows down and I can barely get 40-50 MB/s. If you are reading from and writing back to the same pool simultaneously in your all in one, that could explain the slowish speeds.

 

[obrith]

If there is any CPU contention at all you might see far better performance from 2vCPUs than 4. Current virtualization is very bad at CPU scheduling of that many threads, unless you do not over-commit the CPU cores or set affinity.

 

[Freak1]

Silenus I have tried copying to 3 different PCs 2 Windows 2008r2 servers and a windows 7 labtop with SSD. When copying back and forth between the 3 i get above 90 mb/sec both ways so I'm sure its the OI. None of the testet are VMs on the same machine.

Obrith I thought on that too, but when i go in wmware settings it says guests can get unstable if i change it. Is it worth trying anyway?