# RDO packstack very slow write on cinder volumes

I have an RDO packstack Newton setup with 1 controller node and 2 compute nodes, all on VMs run by KVM.

The cinder volume service is on the controller node as well. It's LVM accessed with ISCSI.

The write speed on a volume attached to an instance is extremely low, ~5MB/s

#  dd if=/dev/zero of=baloon bs=1048576 count=200 oflag=dsync
200+0 records in
200+0 records out
209715200 bytes (210 MB) copied, 39.5572 s, 5.3 MB/s


The network speed between compute and storage(controller) nodes is 100MB/s

Any troubleshooting and tuning tips are appreciated.

Thanks.

edit retag close merge delete

Sort by ยป oldest newest most voted

Hey,

The first bottle neck I'd check would be your connection 100MB/s, 5MB/s isn't great but it won't improve by much unless you go 1G or 10G.

I'm assuming you're probably not running on SSD / fast HDD or top of the line RAM/CPU. Which is fine as it should still work but not very fast, on top of your sluggish network all of a sudden 5.3 MB/s doesn't sound that bad.

Just to put figures in perspective, found the below on some website:

*M = mega m = milli B = byte b = bit

When referring to disk usage, we measure throughput in megabytes per second, or MB/s. Notice the capital M for mega and the capital B for bytes. When referring to network performance, we measure throughput in megabits per second, or Mb/s. Notice the lowercase b. A bit is eight times smaller than a byte. You can figure out your 100Mb/s network's maximum theoretical throughput in MB simply by dividing by 8. 100 / 8 = 12.5.

TCP/IP has ~ a 10% overhead, as does Ethernet, so realistically you'll only see about 80% of that at the high end. A little more basic math shows that 12.5 * .8 = 10. You should expect to be able to write at about 10MB/s over your 100Mb/s network. This lines up perfectly with what you are seeing.*

more