This post is more than 5 years old
24 Posts
0
3097
April 19th, 2018 07:00
ECS Performance regarding Latency
Hi @all,
one of my customers is working on a project where an Openstack cluster is connected via Swift on our ECS.
Now they are complaining, that the read latency, which is shown on the ECS Dashboard, is much to high at 5500ms.
I just want to understand how the number, that is shown on the Dashboard, is calculated and what the reason is for this high latency.
I hope someone can bring light to this number!
Kind regards
Matthias
No Events found!
chris_arnett
110 Posts
0
April 19th, 2018 07:00
Be aware that the "latency" numbers reported in the ECS GUI include full request time (which includes data payload). They are not true "latency" numbers as one would expect from traditional storage systems (time to first byte). So the larger the data being transferred, the larger the "latency" number in the dashboard. We are hoping to provide a time-to-first-byte latency number in a future software release.
Until then, latency can be measured by small file request time, or subjectively by examining application responsiveness. If you have reason to believe that the cluster is not as responsive as it should be, you should open an SR so that support can investigate further. Also keep in mind that as an object storage platform, ECS is not a low-latency system, and response times are subject to environmental limitations. Object storage systems scale much better horizontally by adding more client threads. This is the best way to increase data throughput.
EMC_Matze
24 Posts
0
April 19th, 2018 08:00
Hi Stu,
thanks for the explanation of this number. I expect an average latency of about 100ms to 150ms, but 5500ms seemed to be quite high, but now I understand what the reason can be!
Many thanks!
Kind regards
Matthias