Performance refernce
fmendoline
1 Post
February 1, 2018, 4:56 pmQuote from fmendoline on February 1, 2018, 4:56 pmHello,
Petasan is very new to me and I am looking for reference values. I would be happy if experienced Petasan users could give me an estimate of the expected performance.
Setup:
3 x cluster nodes with the following features each:
2 x Intel E5-2650 (20M 6.00 Cache, 2.00 GHz, 8.00 GT/s Intel
128 GB Ram
4 x 8TB Seagate Enterprise Capacity 3.5 ", 4Kn HDD as OSD
1 x 500GB M2 SSD as Journal
2 Replicas
With PetaSan 1.5:
Cluster Benchmark, 4K IOPS rados benchmark, 1 thread, 1min, 1 client = 506 Write / 2053 read
Cluster Benchmark, 4K IOPS rados benchmark, 8 thread, 1min, 1 client = 2338Write / 15633 read
Thanks for your info / help!
Hello,
Petasan is very new to me and I am looking for reference values. I would be happy if experienced Petasan users could give me an estimate of the expected performance.
Setup:
3 x cluster nodes with the following features each:
2 x Intel E5-2650 (20M 6.00 Cache, 2.00 GHz, 8.00 GT/s Intel
128 GB Ram
4 x 8TB Seagate Enterprise Capacity 3.5 ", 4Kn HDD as OSD
1 x 500GB M2 SSD as Journal
2 Replicas
With PetaSan 1.5:
Cluster Benchmark, 4K IOPS rados benchmark, 1 thread, 1min, 1 client = 506 Write / 2053 read
Cluster Benchmark, 4K IOPS rados benchmark, 8 thread, 1min, 1 client = 2338Write / 15633 read
Thanks for your info / help!
admin
2,930 Posts
February 2, 2018, 12:05 pmQuote from admin on February 2, 2018, 12:05 pmHi,
Performance varies a lot with hardware and network. The values look good, given you are using spinning disks which have iops limits. It will be good to benchmark your throughput MB/s as well. If iops is more important to you than throughput then consider using an all ssd/nvme solution and not hdds.
Another thing is most probably your 4 x hdds per node is a low number and you can double or triple performance by adding more 4 x hdds + ssd journal. The resource load report that is generated will help you know which resources are the most busy and are therefore your main bottlenecks.
Hi,
Performance varies a lot with hardware and network. The values look good, given you are using spinning disks which have iops limits. It will be good to benchmark your throughput MB/s as well. If iops is more important to you than throughput then consider using an all ssd/nvme solution and not hdds.
Another thing is most probably your 4 x hdds per node is a low number and you can double or triple performance by adding more 4 x hdds + ssd journal. The resource load report that is generated will help you know which resources are the most busy and are therefore your main bottlenecks.
Last edited on February 2, 2018, 12:07 pm by admin · #2
Performance refernce
fmendoline
1 Post
Quote from fmendoline on February 1, 2018, 4:56 pmHello,
Petasan is very new to me and I am looking for reference values. I would be happy if experienced Petasan users could give me an estimate of the expected performance.
Setup:
3 x cluster nodes with the following features each:
2 x Intel E5-2650 (20M 6.00 Cache, 2.00 GHz, 8.00 GT/s Intel
128 GB Ram
4 x 8TB Seagate Enterprise Capacity 3.5 ", 4Kn HDD as OSD
1 x 500GB M2 SSD as Journal
2 ReplicasWith PetaSan 1.5:
Cluster Benchmark, 4K IOPS rados benchmark, 1 thread, 1min, 1 client = 506 Write / 2053 read
Cluster Benchmark, 4K IOPS rados benchmark, 8 thread, 1min, 1 client = 2338Write / 15633 readThanks for your info / help!
Hello,
Petasan is very new to me and I am looking for reference values. I would be happy if experienced Petasan users could give me an estimate of the expected performance.
Setup:
3 x cluster nodes with the following features each:
2 x Intel E5-2650 (20M 6.00 Cache, 2.00 GHz, 8.00 GT/s Intel
128 GB Ram
4 x 8TB Seagate Enterprise Capacity 3.5 ", 4Kn HDD as OSD
1 x 500GB M2 SSD as Journal
2 Replicas
With PetaSan 1.5:
Cluster Benchmark, 4K IOPS rados benchmark, 1 thread, 1min, 1 client = 506 Write / 2053 read
Cluster Benchmark, 4K IOPS rados benchmark, 8 thread, 1min, 1 client = 2338Write / 15633 read
Thanks for your info / help!
admin
2,930 Posts
Quote from admin on February 2, 2018, 12:05 pmHi,
Performance varies a lot with hardware and network. The values look good, given you are using spinning disks which have iops limits. It will be good to benchmark your throughput MB/s as well. If iops is more important to you than throughput then consider using an all ssd/nvme solution and not hdds.
Another thing is most probably your 4 x hdds per node is a low number and you can double or triple performance by adding more 4 x hdds + ssd journal. The resource load report that is generated will help you know which resources are the most busy and are therefore your main bottlenecks.
Hi,
Performance varies a lot with hardware and network. The values look good, given you are using spinning disks which have iops limits. It will be good to benchmark your throughput MB/s as well. If iops is more important to you than throughput then consider using an all ssd/nvme solution and not hdds.
Another thing is most probably your 4 x hdds per node is a low number and you can double or triple performance by adding more 4 x hdds + ssd journal. The resource load report that is generated will help you know which resources are the most busy and are therefore your main bottlenecks.