Tigris Base Latency

We’re using Tigris to host web content, mostly pretty small objects (<10 KB). Based on the marketing material I was hoping for low latency, but in my experience even tiny objects (200 bytes) take around 100 ms.

Most of the requests land in the 200–500 ms range:

The application runs in ams``ams and the objects are fetched with plain old GetObject``GetObject without any extra options. Before that we used a MinIO instance running on Fly with the same content and usage patterns: the latency in that setup was mostly in the 0–5 ms range.

Just to compare notes: Is this latency to be expected? Is there anything to optimise?

Solution

Further to that reply, it may also be worth loading https://debug.fly.dev/ and seeing what region code that request is served from. In my case it should be served from lhr``lhr and so have minimal latency, but it’s served from iad``iad and so has high latency. There can be some weird routing sometimes!

Someone, this is unexpected. While we don’t have our deployment in the AMS region, we do have it in nearby regions, FRA and LHR. So the expected latency is ≤ 20 ms here. Could you share the bucket name by emailing us at help@tigrisdata.com? And we will take a look at what’s going on.

Solution

I’m getting sometimes ams``ams and sometimes otp``otp.

When I first started using Fly I saw about five regions (because I didn’t scroll down) and the closest one was ams``ams, which is 2300 km away.

But today I saw otp``otp, which is only 76 km from me.

Now that’s nice.