Note: "permalinks" may not be as permanent as we would like,
direct links of old sources may well be a few messages off.
/ 2004-04-28 11:12:39 -0400 \ Seriy: > We've setup two machines with 3ware 8506 controllers running RAID 10: > > Controller SCSI ID: 0 > Monitor version: ME7X 1.01.00.038 > Firmware version: FE7S 1.05.00.063 > BIOS version: BE7X 1.08.00.048 > PCB version: Rev5 > Achip version: 3.20 > Pchip version: 1.30-66 > Model: 8506-4LP > Serial number: F14402A3460382 > Unit count: 1 > Unit 0 > Status: OK > Capacity: 240.06 GB (468879104 blocks) maybe you want to enable this one? > Write Cache: Not In Use > Configuration: Striped Mirrors 64K (RAID 10) > Stripe size: 64K (65536 bytes) > Subunit count: 2 > > > Network cards are Intel PRO/1000 MT Server adapter with the Intel's > drivers version 5.2.30.1. drbd version 0.6.11. Kernel is 2.4.19. > > Everything seems to work, but the connection is terribly slow: > > version: 0.6.11 (api:64/proto:62) > > 0: cs:SyncingAll st:Primary/Secondary ns:13939248 nr:0 dw:61378768 > dr:15169750 pe:215 ua:0 > [========>...........] sync'ed: 42.2% (17763/30725)M > finish: 2:02:48h speed: 2,486 (3,122) K/sec > you definetely want to use 0.6.12 !! though this has nothing to do with the issue at hand. you are sure the raid is not rebuilding at the same time? is the device used at the same time by any applications? > Here is the config: > > drbd: pre-parsed needs update, parsing /etc/drbd.conf > ................................. > parsed configuration NOT written to /var/lib/drbd/drbd.conf.parsed > CONF=(\ > __hash__34 \ > 'drbd0:protocol' 'C' \ > 'drbd0:fsckcmd' '/bin/true' \ > 'drbd0:disk' ' --disk-size=31463302 --do-panic' \ > 'drbd0:net' ' --sync-rate=100M --sync-min=100M setting min that high is not usefull, but will add latency to all other processes making the box feel very slow. set to 3M or 10M. syncer will always sync at the highes possible throughput available (but not more than max of course). > --tl-size=1024 --sync-nice=19' \ sync nice 19 while setting sync min 100M is complete nonsense. set to 0 or -10. > 'drbd0:on main1:device' '/dev/nb0' \ > 'drbd0:on main1:disk' '/dev/sda2' \ > 'drbd0:on main1:address' '10.10.0.1' \ > 'drbd0:on main1:port' '7788' \ > 'drbd0:HOSTS' ' main1 main2 ' \ > 'drbd0:on main2:device' '/dev/nb0' \ > 'drbd0:on main2:disk' '/dev/sda2' \ > 'drbd0:on main2:address' '10.10.0.2' \ > 'drbd0:on main2:port' '7788' \ > 'drbd0:skip-wait' 'no' \ > 'drbd0:load-only' 'no' \ > 'drbd0:inittimeout' '-0' \ > 'drbd0:incon-degr-cmd' '0' \ > ) > > > I'm wondering if anybody has a similar configuration, especially 3ware > controllers and similar network cards. > > Please advise if anybody has any suggestions, > Thanks a lot. to find your bottleneck, which is most likely NOT drbd itself, you may want to check: what results does bonnie++ (just for example. or any other benchmark utility you want to use) give, on a - non-drbd partition on the same hardware - unconnected drbd partition - connected (not syncing) drbd partition measure the performance and latency (rtt) of your network, especially when saturated. Lars Ellenberg