[DRBD-user] low performance on 'Write Tests' using tiobench

Chun Tian (binghe) tianchunbinghe at gmail.com
Thu Mar 2 09:45:30 CET 2006

Note: "permalinks" may not be as permanent as we would like,
direct links of old sources may well be a few messages off.


Hi,

We try to use DRBD in some reason. After some tests using tiobench
like DRBD's performance page, we found the write rate is too low,
almost 1/2 in Sequential Writes, 1/4 in Random Writes.

We use two HP-DL140, with 2.8GHz P4 CPU*2 (with HyperThreads), 4GB
memory and 2 SATA Maxtor hard-disk (we just use one of the two on each
machine), and got these results:

1. Sequential Writes

Identifier	File Size	Blk Size	Num Thr	Rate	Maximum CPU%	Avg
Latency	Max Latency	Lat% >2s	Lat% >10s	CPU Eff
2.6.15.4	6144	4096	4	35.19	112.000%	1.004	16665.66	0.00515	0.00000	31
2.6.15.4-dis	6144	4096	4	17.58	54.070%	1.881	283191.21	0.01303	0.00134	33
2.6.15.4-prot-c	6144	4096	4	19.21	65.180%	1.607	683948.76	0.00648	0.00134	29
2.6.15.4	6144	4096	8	34.13	221.600%	1.905	45493.9	0.01908	0.00019	15
2.6.15.4-dis	6144	4096	8	18.48	115.400%	3.204	348416.72	0.04895	0.00178	16
2.6.15.4-prot-c	6144	4096	8	19.18	149.900%	2.952	575454.06	0.03472	0.00248	13
2.6.15.4	6144	4096	16	35.29	461.600%	3.176	40471.59	0.08113	0.00051	8
2.6.15.4-dis	6144	4096	16	17.40	241.100%	6.184	354021.31	0.07319	0.01087	7
2.6.15.4-prot-c	6144	4096	16	18.78	332.300%	5.087	635343.88	0.05341	0.00826	6

2. Random Writes

Identifier	File Size	Blk Size	Num Thr	Rate	Maximum CPU%	Avg
Latency	Max Latency	Lat% >2s	Lat% >10s	CPU Eff
2.6.15.4	6144	4096	4	0.59	1.032%	10.097	3155.68	0.00000	0.00000	57
2.6.15.4-dis	6144	4096	4	0.12	0.454%	6.362	388.50	0.00000	0.00000	26
2.6.15.4-prot-c	6144	4096	4	0.15	0.583%	6.715	436.39	0.00000	0.00000	26
2.6.15.4	6144	4096	8	0.59	2.596%	1.400	437.01	0.00000	0.00000	23
2.6.15.4-dis	6144	4096	8	0.13	2.661%	0.072	32.24	0.00000	0.00000	5
2.6.15.4-prot-c	6144	4096	8	0.15	1.541%	7.065	493.70	0.00000	0.00000	10
2.6.15.4	6144	4096	16	0.65	4.094%	0.049	0.26	0.00000	0.00000	16
2.6.15.4-dis	6144	4096	16	0.13	3.657%	1.359	277.98	0.00000	0.00000	3
2.6.15.4-prot-c	6144	4096	16	0.17	5.794%	2.486	494.66	0.00000	0.00000	3

We care the 'rate' field more, and found that after using DRBD, either
connect or disconnect, the 'rate' took down from [35] to [17], and
[0.6] to [0.15] ... so we cannot REPLAY the results on DRBD's page...
very confused.

Who can tell me the possible reason, thanks very much!



More information about the drbd-user mailing list