db performance comparisons (II)

gistfile1.md

This is just about GET & SET, the basic Key-Value store commands.

• Redis is constantly fast and faster than Memcached
• MongoDB is not a good key-value store
• levelUp is only slower than levelDown doing PUT operations
• leveled and levelUp perform almost the same
• multilevel (levelUp rpc) isn't fast enough yet, even with a fake network layer
• multileveled (leveled rpc) is faster than MongoDB doing GET and is slowed down heavily by the network layer
• multilevel-http is not usable at all when performance counts

gistfile2.txt

                      Redis
          15,849 op/s ⨠ set small
          18,625 op/s ⨠ set medium
          18,733 op/s ⨠ set large
          17,634 op/s ⨠ get large
          18,514 op/s ⨠ get medium
          18,345 op/s ⨠ get small

                      MongoDB
          20,900 op/s ⨠ set small
          22,947 op/s ⨠ set medium
          22,035 op/s ⨠ set large
           5,259 op/s ⨠ get large
           5,980 op/s ⨠ get medium
           6,461 op/s ⨠ get small

                      Memcached
          11,644 op/s ⨠ set small
          11,629 op/s ⨠ set medium
          11,484 op/s ⨠ set large
          12,333 op/s ⨠ get large
          12,346 op/s ⨠ get medium
          12,039 op/s ⨠ get small

                      levelUP
          37,840 op/s ⨠ set small
          36,189 op/s ⨠ set medium
          34,063 op/s ⨠ set large
          41,328 op/s ⨠ get large
          45,933 op/s ⨠ get medium
          48,099 op/s ⨠ get small

                      levelDOWN
          69,075 op/s ⨠ set small
          67,202 op/s ⨠ set medium
          61,458 op/s ⨠ set large
          48,181 op/s ⨠ get large
          50,330 op/s ⨠ get medium
          48,918 op/s ⨠ get small

                      leveled
          73,808 op/s ⨠ set small
          72,748 op/s ⨠ set medium
          65,687 op/s ⨠ set large
          60,643 op/s ⨠ get large
          61,294 op/s ⨠ get medium
          63,400 op/s ⨠ get small

                      multilevel
           6,714 op/s ⨠ set small
           6,413 op/s ⨠ set medium
           6,121 op/s ⨠ set large
           8,012 op/s ⨠ get large
           8,214 op/s ⨠ get medium
           8,215 op/s ⨠ get small

                      multilevel (fake network)
          14,457 op/s ⨠ set small
          14,230 op/s ⨠ set medium
          12,659 op/s ⨠ set large
          18,551 op/s ⨠ get large
          18,563 op/s ⨠ get medium
          18,844 op/s ⨠ get small

                      multileveled
           9,851 op/s ⨠ set small
           9,163 op/s ⨠ set medium
           9,603 op/s ⨠ set large
           9,315 op/s ⨠ get large
           9,404 op/s ⨠ get medium
           9,604 op/s ⨠ get small

                      multileveled (fake network)
          28,962 op/s ⨠ set small
          28,335 op/s ⨠ set medium
          27,434 op/s ⨠ set large
          26,322 op/s ⨠ get large
          26,856 op/s ⨠ get medium
          26,361 op/s ⨠ get small

                      multilevel-http
           1,079 op/s ⨠ set small
           1,227 op/s ⨠ set medium
           1,813 op/s ⨠ set large
           1,996 op/s ⨠ get large
           1,635 op/s ⨠ get medium
             978 op/s ⨠ get small

                      Memory
      26,307,413 op/s ⨠ set small
       9,572,282 op/s ⨠ set medium
      14,120,584 op/s ⨠ set large
      22,651,185 op/s ⨠ get small
      33,082,786 op/s ⨠ get medium
      34,338,065 op/s ⨠ get large

What is the last result, memory? The memory of the process itself?
Cheers!

@juliangruber given that the results can vary massively depending on the sizes of the keys & values can you give us some clue about methodology here?

@DTrejo The last result is from storing everything in JavaScript memory

@rvagg those are the strings: https://github.com/juliangruber/multilevel-bench/blob/master/index.js#L1-L5