[DRBD-user] ganeti + drbd 8.0.7 = slow?!

Rudolph Bott rb at megabit.net
Wed Apr 9 17:27:54 CEST 2008

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


Hey list,

I set up a two-node-xen-cluster using Debian etch and ganeti. Everything works fine so far (setting up DomUs, failover between the two nodes etc).

Both nodes are equal in hardware-terms, Xeon QuadCore with 4GB RAM, 3ware 9650, 4-disk-RAID5 array. Both Dom0's have 1 core assigned and 512MB memory. The rough throughput of the controller seems to be around 90-100MB/sec.

FYI, ganeti uses LVM and puts DRBD replication on top between the cluster nodes to replicate DomU storage across the cluster. The initial sync already is quite slow... according to /proc/drbd somewhere around 2-5MB/sec. Using a dd if=/dev/zero of=/tmp/somefile to copy ~3GB of data leaves me with 
5-10mb/sec

this is way to slow for whatever use :/ On top of that, the DomU becomes quite unresponsive/sluggish after a lot of I/O has happened (like a dd run). e.g. it takes a couple of seconds just to 'ls' a folders content (containing only one file!). This behaviour stops after a short while but still it's 
strange.

The drbd-link between the two nodes used to be a cross-linked connection with e1000 nics on both sides. I've replaced it with a switched connection but that didn't change anything either (someone told me about possible timing problems with cross-linked e1000 nics).

Mit freundlichen Grüßen / with kind regards

Rudolph Bott
-- 
Megabit Informationstechnik GmbH
Karstr.25  41068 Moenchengladbach  Tel:02161/30898-0  Fax:-18
AG MG HRB 10141, GF: Dipl.-Ing. Thomas Tillig, Michael Benten



More information about the drbd-user mailing list