Hi everyone,
?
I find the disk in UKB RAP is very slow. The performance of sequency write only gives 50-80MB/s although it's claimed to be the SSD, the speed is only the HDD.
?
I tested both ssd1 and ssd3 instance type, all very similar.
?
This would make data to be downloaded very slow (dx download). Althought the speed in dx to memory is also slow (at most 100MB/s).
Is there any way to speed this up? Both for the disk speed and also the dx download speed.
Thanks.
Not knowing much about this, but I would first consult Rate card (maybe you did). There is a list of all available instance types with mapping to EC2 Instance Type (AWS). If you select a specific instance type, you can search the AWS pages and get more info about the particular instance type and its performance. E.g. I took i3en.3xlarge (mem3_ssd3_x12) and consulted this page: https://aws.amazon.com/ec2/instance-types/i3en/
With that, I was able to find some differences in the instance types when taking Instance Storage (GB) vs. Network Bandwidth (Gbps) vs. EBS Bandwidth (Gbps) into consideration. For example, i3en.3xlarge had better Network Bandwidth as well as EBS Bandwidth.
Thanks for the information. I tested different types of instances, the speed is different indeed. But I feel strange that the disk speed for mem3_ssd3_x2 is only 178 - 200MB/s, it has the 1.2TB of SSD (normal nodes would be > 700MB/s). ?
The basic type mem1_ssd1_v2_x2, has 50GB of SSD, the disk speed is extremely slow. (30-40MB/s). In comparision, my free test node from AWS which has a very basic type (t2.micro, 1vCPU, 1G RAM, 30GB SSD) achieved the speed of 67MB/s.
The network testing is > 3000Mbps for all nodes (via speedtest), however, the copy from bucket to RAM disk is only about 100MB/s for some nodes.
Could you investigate this? Sometimes, the users would need robust SSD and robust download speed, rather than large CPU and RAM.
The networking speed is also affected by number of core as well for the same type of instance.
If you suspect that the worker in our platform is defected, could you report to ukbiobank-support@dnanexus.com? They could then log the case and find the engineer to test it.
Comments
4 comments
Not knowing much about this, but I would first consult Rate card (maybe you did). There is a list of all available instance types with mapping to EC2 Instance Type (AWS). If you select a specific instance type, you can search the AWS pages and get more info about the particular instance type and its performance. E.g. I took i3en.3xlarge (mem3_ssd3_x12) and consulted this page: https://aws.amazon.com/ec2/instance-types/i3en/
With that, I was able to find some differences in the instance types when taking Instance Storage (GB) vs. Network Bandwidth (Gbps) vs. EBS Bandwidth (Gbps) into consideration. For example, i3en.3xlarge had better Network Bandwidth as well as EBS Bandwidth.
Another thing (not tested on my side) might be using tools such as dxda for download (https://github.com/dnanexus/dxda).
Hi @Ondrej Klempir? ,
?
Thanks for the information. I tested different types of instances, the speed is different indeed. But I feel strange that the disk speed for mem3_ssd3_x2 is only 178 - 200MB/s, it has the 1.2TB of SSD (normal nodes would be > 700MB/s). ?
The basic type mem1_ssd1_v2_x2, has 50GB of SSD, the disk speed is extremely slow. (30-40MB/s). In comparision, my free test node from AWS which has a very basic type (t2.micro, 1vCPU, 1G RAM, 30GB SSD) achieved the speed of 67MB/s.
The network testing is > 3000Mbps for all nodes (via speedtest), however, the copy from bucket to RAM disk is only about 100MB/s for some nodes.
Could you investigate this? Sometimes, the users would need robust SSD and robust download speed, rather than large CPU and RAM.
Thanks.
Regards,
Zhili
The networking speed is also affected by number of core as well for the same type of instance.
If you suspect that the worker in our platform is defected, could you report to ukbiobank-support@dnanexus.com? They could then log the case and find the engineer to test it.
Please sign in to leave a comment.