Question about upload / download bandwidth

❓Question

Hi, I'm running some benchmarks on ovh dedicated servers in us-east-vin, my bucket is set to single region in IAD.

The benchmark was with tigrisfs. I noticed that upload / download was capped around 40 - 50 MB/s

Is this just because my account is newish?

I also noticed when looking at metrics that my traffic was going through ord and not iad

Response

What's the network capacity supported by your ovh server?

And could you share more about the benchmark?

Yep give me a second

The connections is 1 Gbps

I mounted a Tigris bucket locally using TigrisFS and ran fio benchmarks to measure raw I/O performance at different --memory-limit settings (512 MB, 1 GB, 2 GB, 4 GB). But noticed upload / download speeds to the bucket were capped at 40-50Mbps

Testing with normal tigris cli it's less bandwidth, when I use aws cli it gives me 40-50Mbps

It probably is the VM capping the bandwidth

Regarding the cli, we will take a look. Though I must say we haven’t done extensive performance benchmarks with the cli yet.

Yeah I'm experimenting with it now to verify. Could you give me a rough ballpark of what I should expect if there aren't issues on my end?

You should be easily able to see multiple GB/sec with fio on TigrisFS. With AWS cli you can achieve something similar by running multiple downloads uploads in parallel

Could you also share the fio command you are using?

# Sequential write (2 GB, 1 MB blocks)
  fio --name=seq-write --directory=<dir> --rw=write --bs=1M --size=2G --numjobs=1 --end_fsync=1 --fallocate=none

# Sequential read (2 GB, 1 MB blocks)
  fio --name=seq-read --directory=<dir> --rw=read --bs=1M --size=2G --numjobs=1

# Random read 4K (64 MB)
  fio --name=rand-read --directory=<dir> --rw=randread --bs=4K --size=64M --numjobs=1

# Random write 4K (64 MB)
  fio --name=rand-write --directory=<dir> --rw=randwrite --bs=4K --size=64M --numjobs=1 --end_fsync=1 --fallocate=none

I quickly spun up a vm on hetzner and it gave much better results.

Nice. Hetzner also has limit on egress from the VM. So likely not showing true performance. But good to hear that you are getting better results

There is also another unrelated issue with OVH that I discovered. Not sure if there is anything that be done.

For some reason OVH servers in N Virginia (Vint Hill) route through ORD instead of going directly to IAD

That is most likely their resolver IP address marked to be in a geo around Chicago. We resolve our dns name to the nearest pop based on a geo IP database.

Is this affecting you much in practice though?

It could actually matter since I'm building something with a lot of tenants and am constrained in the number of threads I can give tigrisfs.

But I don't think it's a huge issue, it's still 20ms

We are looking into it to see what we can do on our end

ty