Dilip Sathaye
2013-Apr-09 06:32 UTC
[Lustre-discuss] Querry on expected throughput of FC HBA
Dear All,
I have a query on expected throughput from 8gb FC HBAs. We have a lustre
2.3 setup with 7 OSTs on 8gb FC . The OSTs are mapped in such a way that
one FC port has one OST assigned. During our benchmarks using IOR ( 32
clients on IB FDR ) , with -B option , we seem to getting saturated at
1.3GB/s throughput. This roughly maps to 185MB/s throughput per FC HBA port
for OSTs. Other tests like multiple dd , from multiple systems are also
pointing to approximate 185MB/s per FC port on OST. I am not sure if this
is normal or much less ? Sample output of lfs is as given below.
lfs getstripe .
.
stripe_count: 7 stripe_size: 1048576 stripe_offset: -1
./t
lmm_stripe_count: 7
lmm_stripe_size: 1048576
lmm_layout_gen: 0
lmm_stripe_offset: 5
obdidx objid objid group
5 784762 0xbf97a 0
6 785240 0xbfb58 0
0 784632 0xbf8f8 0
2 785619 0xbfcd3 0
4 784596 0xbf8d4 0
1 785144 0xbfaf8 0
3 785079 0xbfab7 0
./f2
lmm_stripe_count: 7
lmm_stripe_size: 1048576
lmm_layout_gen: 0
lmm_stripe_offset: 2
obdidx objid objid group
2 785617 0xbfcd1 0
4 784594 0xbf8d2 0
1 785142 0xbfaf6 0
3 785078 0xbfab6 0
5 784761 0xbf979 0
6 785239 0xbfb57 0
0 784631 0xbf8f7 0
./f1
lmm_stripe_count: 7
lmm_stripe_size: 1048576
lmm_layout_gen: 0
lmm_stripe_offset: 3
obdidx objid objid group
3 785076 0xbfab4 0
5 784759 0xbf977 0
6 785237 0xbfb55 0
0 784630 0xbf8f6 0
2 785616 0xbfcd0 0
4 784593 0xbf8d1 0
1 785141 0xbfaf5 0
Regards
Dilip
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
http://lists.lustre.org/pipermail/lustre-discuss/attachments/20130409/9c6af529/attachment.html