#Higher latency on Metal

37 messages · Page 1 of 1 (latest)

untold plover
#

Hi

I did a test where I deployed a rust axum rest api twice, one on Metal and one on GCP. But the latency on Metal is just way higher.
The increased latency seems to be coming from the database connection.
I apologise in advance if I did something stupid :)

ornate geodeBOT
#

Project ID: N/A

untold plover
#

ebfe176f-63cb-441f-bcb3-f4a8de975077

#

Following is a screenshot of the HTTP logs of the app deployed on GCP

#

And the one on Metal

tired surge
#

did you also put the database on metal?

untold plover
#

should be yes. Metal app -> Metal database. GCP app -> GCP database

#

I am EU btw

tired surge
#

is your metal service using the metal edge as well?

untold plover
#

I have tried switching between the metal edge and the old one but doesnt seem to make that much of a difference

tired surge
#

its a DNS update so please be patient, how long has the domain been on the metal edge?

untold plover
#

Well ive been switching between them pretty quickly. Just switched back to the Metal edge. How long do you recommend waiting before testing again?

tired surge
#

you can dig the domain, when it starts returning a list of IP addresses you should be good

untold plover
#

not too familiar with networking. Could you elaborate?

tired surge
#

dig, the cli tool

untold plover
#

Is it this you're looking for?

edge.railway.app.       152     IN      A       66.33.22.3
edge.railway.app.       152     IN      A       66.33.22.1
edge.railway.app.       152     IN      A       66.33.22.4
edge.railway.app.       152     IN      A       66.33.22.2
tired surge
#

yep thats the metal edge

untold plover
#

Have a couple dipping below 10ms now. But still on the higher side

tired surge
#

can you run a traceroute to one of the ip addressees?

untold plover
#
traceroute to 66.33.22.1 (66.33.22.1), 30 hops max, 60 byte packets
 1  172.20.64.1 (172.20.64.1)  0.222 ms  0.151 ms  0.164 ms
 2  mymodem (192.168.1.1)  0.707 ms  1.157 ms  0.891 ms
 3  10.24.49.76 (10.24.49.76)  9.628 ms  9.678 ms  10.193 ms
 4  * * *
 5  ae-64-100.ibrstr6.isp.proximus.be (91.183.241.200)  17.692 ms  17.436 ms  19.219 ms
 6  brx-b4-link.ip.twelve99.net (213.248.97.112)  19.279 ms  15.113 ms  15.188 ms
 7  adm-bb2-link.ip.twelve99.net (62.115.116.230)  17.139 ms  15.470 ms  15.726 ms
 8  adm-b12-link.ip.twelve99.net (62.115.137.191)  16.426 ms  17.138 ms  17.142 ms
 9  railwaycorp-ic-390073.ip.twelve99-cust.net (62.115.196.223)  18.081 ms  18.836 ms  17.939 ms
10  * * *
11  * * *
12  * * *
13  * * *
14  * * *
15  * * *
16  * * *
17  * * *
18  * * *
19  * * *
20  * * *
21  * * *
22  * * *
23  * * *
24  * * *
25  * * *
26  * * *
27  * * *
28  * * *
29  * * *
30  * * *
tired surge
untold plover
#
edge=railway/europe-west4-drams3a
zone=europe-west4-drams3a
ip=::ffff:100.64.0.2
forwarded=91.177.28.24
hs=f09480be9e4a
req=T6a7ncZgTVaGcj5738lYAA_3554946428

How accurate should my current location be? :)

tired surge
#

as accurate as you feel comfortable announcing publicly

untold plover
#

3550 Heusden-Zolder, Limburg, Belgium

tired surge
#

fr?

untold plover
#

always welcome

tired surge
#

you know these threads are indexed publically on google?

untold plover
#

thats just zipcode btw

#

not street

tired surge
#

fair enough

untold plover
#

xd

tired surge
#

cc @raven siren - slightly slower response times on Metal.

✅ - db on metal
✅ - app on metal
✅ - region same as user
✅ - metal edge enabled
✅ - priv net for db comms

I can't think of anything else that could cause issues.

untold plover
#

Side note:
Even when querying an endpoint with no database interaction its still slightly slower.
GCP:

#

Metal:

tired surge
#

interesting, ill let phin weigh in on this, so lets wait to hear back from him

untold plover
#

Thanks for your time and help in advance!

untold plover
#

Not sure if you guys looked any further into this but I just tested again and this seems to mostly be fixed. GCP still seems to be faster but its by a difference of <5ms on average.