Low read performance detected in the tests

[ikopylov]

The test was done on a cluster with 1 node with 1 disk with 1 vdisk. Other important settings: max_blob_size: 1tb, max_data_in_blob: 1000000000 and timestamp_period: 30s. Read test was done with bobp with 8 threads.

Test plan:

1. Write 2000000 data (50kb each) with bobp with 32 threads
2. Read that data (bad perf here)
3. Overwrite 2000000 data (write them with the same ids)
4. Read that data again (even smaller perf here)
5. Overwrite 2000000 data (write them with the same ids)
6. Read that data again (even smaller perf here)

Script for single test iteration (write + read):

bobp -b put -l 51200 -c 2000000 -h 10.5.7.195 -t 32
sleep 30
echo "$item - no opt\n" >> ./optimize_result
bobp -b get -l 51200 -c 2000000 -h 10.5.7.195 -t 8 >> ./optimize_result
sleep 30

Results:

Total statistics, elapsed: 8112.940817929s
avg total: 246 rps | total err: 0
get: 12327.42 kb/s | resp time   0.03 ms
--
Total statistics, elapsed: 19838.394783048s
avg total: 100 rps | total err: 0
get: 5041.34 kb/s | resp time   0.08 ms
--
Total statistics, elapsed: 30376.709618961s
avg total: 65 rps | total err: 0
get: 3292.40 kb/s | resp time   0.12 ms

Rps and throughput is too low for this test. HDD linear read performance: 143.60 MB/sec. We need to investigate the source of the problem