Budget Fair Queueing (BFQ) Storage-I/O Scheduler

Additional responsiveness results

In this page we report further responsiveness results, with respect to the ones provided in our main result page. In particular, for each device we report our results with xterm and lowriter.

The same comments as in the main result page apply for xterm, so we do not repeat them also in this page. On the other hand, we add an explanation of why lowriter start-up time almost doubles with writes in the background.

PLEXTOR SSD

SSD bash start-up time
Figure 3. xterm start-up time on the Plextor SSD (lower is better).
SSD lowriter start-up time
Figure 4. lowriter start-up time on the Plextor SSD (lower is better).

Both with 10r-seq and 5r5w-seq, BFQ guarantees a start-up time of lowriter comparable to that with an idle device, and dramatically outperforms the other I/O schedulers. In some cases, the latter even make the application fail to start in 60 seconds. With 5r5w-seq, BFQ guarantees a higher start-up time than with 10r-seq. This is mainly due to the combination of the following tricky problems:

HITACHI HDD

HITACHI HDD xterm start-up time
Figure 1. xterm start-up time on the HITACHI HDD (lower is better).
HITACHI HDD xterm start-up time
Figure 2. lowriter start-up time on the HITACHI HDD (lower is better).

With 10r-seq, BFQ guarantees a start-up time comparable to that with an idle device also in case of an HDD, whereas with all the other schedulers the application fails, with any background workload, to start in 60 seconds. With writes in the background, the start-up time of lowriter with BFQ almost doubles, for the same reason as for the HDD.