VSP 360 · Small multiplesPast 24 hours · 16 Oct 01:00 – 17 Oct 01:00 CEST · 10-minute averages
The busiest processors are on the quietest array, and the drives are idle everywhere. The one link that works hard is Array A's CL1-A and CL2-A pair.
Array C's two MP blades ran at about 40% all day while it served about 100 IOPS, so their load came from internal work rather than hosts, such as snapshot or data-reduction housekeeping; 129 snapshot pairs were removed from it this week. Array B's blades ran at 9% while serving 32,000 IOPS. Array A's two front-end ports averaged 550–600 MiB/s and peaked at 2.7 GiB/s, close to a 32G FC link's limit, while every parity group stayed at or below 1% busy.
40%
Array C MP blades
Average busy at about 100 IOPS; peak 59%
9%
Array B MP blades
At 32,000 IOPS of 512-byte I/O
2.7GiB/s
Array A port peak
CL1-A; about 90% of a 32G FC link one way
≤1%
Parity group busy
All seven groups, all day
Figure 1 · MP blade utilization
Processor load does not follow host I/O: Array C works hardest with the least to do
Every cell uses the same scale, 0 to 100% busy, and runs 01:00 to 01:00 with faint gridlines at 07:00, 13:00 and 19:00. Line: 10-minute average. Shading: the highest 1-minute value in each 10 minutes. Array C's data ends at 17:10, when it stopped reporting to VSP 360.
Figure 2 · Port throughput
Two ports carry nearly all the bandwidth; the 32,000 IOPS on Array B fit in 9 MiB/s
Every cell uses the same scale, 0 to 3,052 MiB/s, the one-way rate of a 32G Fibre Channel link (3,200 MB/s). iSCSI ports run at 10 GbE (1,192 MiB/s) and NVMe/TCP at 100 GbE. Ports that never exceeded 5 MiB/s are left out. Same time span and gridlines as Figure 1.
Figure 3 · Parity group busy %
The flash is not the bottleneck: no parity group was more than 1% busy at any point
Every cell uses the same scale, 0 to 100% busy, 5-minute samples. Values are whole percentages, so 0–1% is the full range seen. Same time span and gridlines as Figure 1.