This post is more than 5 years old
5 Practitioner
•
274.2K Posts
1
4460
Cloud pool jobs throughput to ECS very slow
Hi,
I have configured ISILON cloud pools to archive data to ECS cluster. For archiving about 3 GB (file size 16 MB each) it takes 35 minutes. I am getting about 1.5MBps write bandwidth from Isilon to ECS. I have 10 GB nics and running latest versions of isilon and ECS. Is there anyway to increase the throughput?
Anonymous
5 Practitioner
5 Practitioner
•
274.2K Posts
2
August 18th, 2017 04:00
Hi,
I was able to fix he issue. We were using IP addresses to connect to ECS cluster from ISILON. We changed this to FQDN of the load balancer and added a wildcard entry which fixed the issues. Now I am getting speeds upto 200 MBps to ECS cluster..
sjones51
252 Posts
3
June 12th, 2017 08:00
Hello arjunnagaraju,
I can't speak to how to specifically improve throughput without a deep dive into the network. There are general things that I am sure you have already considered, like making sure the entire network path supports 10GB and you don't have a bottleneck somewhere. In regards to the cluster itself, there are a few papers out there that you could start with.
Isilon OneFS Cluster Performance Metrics Tips and Tricks
https://support.emc.com/docu78704
Isilon Performance Troubleshooting Guide
https://support.emc.com/docu80766
There are also more Isilon/ECS papers and videos here:CloudPools with ECS: New whitepaper and videos
SKT2
2 Intern
2 Intern
•
1.3K Posts
0
August 21st, 2017 01:00
what is the wild card entry you are referring here? What is the significance of that?
Brian_Coulombe_
107 Posts
1
November 9th, 2017 05:00
Pulling up this old thread for a second.
IIRC, from EMC the ECS bandwidth from Isilon to ECS is capped per node at 50mb/s. So I think it doesn't matter if you go through a load balancer or not. If you have 9 nodes connected to ECS, you will be capped at 450mb/s (proven on my 9 node cluster).
Yes you should use VLAN connectivity (ie: all interfaces on every node) and not use direct IPs (because that alone will choke your bandwidth) but I would like EMC to make changes to Cloudpools the way SyncIQ works. We should be able to limit the bandwidth for the cloudpools job during working hours and open it up on off hours.
That'll get data faster to and from ECS (especially during a full recall) and not impact users. FWIW, just take the SyncIQ bandwidth policies and apply it to cloudpools.