[DRBD-user] DRBD performance (only uses 40% of a GE link)

Igor Neves igor at 3gnt.net
Mon Mar 3 13:21:49 CET 2008

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


Christian Balzer wrote:
> Hello,
>
> On Mon, 3 Mar 2008 09:34:30 +0100 Florian Haas wrote:
>   
>> And you are perfectly certain you know what you're doing in setting
>> use-bmbv?
>>
>>     
> No, but a question from me about this last Thursday went unanswered so far
> and the documentation suggests that it speeds things up.
> So for the sake of determining the most effective setup...
>
>   
>> And you are also perfectly certain you want to set max-buffers != 
>> max-epoch-size?
>>
>>     
> Nothing I read suggested that they need to equal, but changing both to
> 2048 made no difference. Not that I expected any, given that a test with
> protocol B gave pretty much the same results as C.
>
>   
>> And you have a good reason to disable the sndbuf-size 512k you had
>> already set?
>>
>>     
> Yeah, since it did not have any noticeable effect on the performance.
>
> It did test more than just the config posted, but the results did not vary
> considerably. 
> Right now I'm doing a run with use-bmbv off, max buffers and epoch at 8192
> and sndbuf at 1MB. ethstats hovers around:
> ---
> eth1:  446.41 Mb/s In     6.09 Mb/s Out -   9123.4 p/s In    4778.3 p/s Out
> ---
> Which suggest the results will be slightly better, but still it's just
> using half of the available bandwidth. 
> What is the most throughput anybody has achieved on writes with Protocol
> C and what settings (and network hardware) did they use? 
>   

I have 4 NIC Intel Ethernet cluster like you, and i take out of it
115MBytes write. The controller behind it gives me about 160MBytes
writing, but  I get the Gbit Ethernet limit, i have 4 Gbit Ethernet
bonding but i cant get out that, even in network tests, because of the
load balancing algorithm used by the switch.

I never manage to get on 4Gbit Ethernet trunk, with one switch in the
middle, for the same mac address. For this you have to buy a switch that
does the same algorithm that Linux bonding driver does (mode 0), or LACP
with the same algorithm. If anyone knows one, please tell me, i would
like to test it too.

For different mac address's in network tests i get about 3.3Gbit,
without tuning (MTU).

Good luck.
>   
>> Just my €.02.
>>
>>     
> Just my 2yen. ^_^
>
> Christian
>
> P.S.
> The results just came back:
> ---
> Version  1.03       ------Sequential Output------ --Sequential Input- --Random-
>                     -Per Chr- --Block-- -Rewrite- -Per Chr- --Block-- --Seeks--
> Machine        Size K/sec %CP K/sec %CP K/sec %CP K/sec %CP K/sec %CP  /sec %CP
> borg00a      50000M           54917  17 47675  15           410592  55 392.4   1
>                     ------Sequential Create------ --------Random Create--------
>                     -Create-- --Read--- -Delete-- -Create-- --Read--- -Delete--
>               files  /sec %CP  /sec %CP  /sec %CP  /sec %CP  /sec %CP  /sec %CP
>                 128 79514  95 +++++ +++ 65768  76 78165  94 +++++ +++ 81634 100
> ---
> So yes, slightly better, but still nothing to write home about.
> Which of the changes above you think brought this about and will cranking
> up things to even more insane levels get me to about 80MB/s (which is what
> I could settle for)?
>
>   
>> On Saturday 01 March 2008 06:49:30 Christian Balzer wrote:
>>     
>>> Hello,
>>>
>>> I'm trying to build a HA cluster here. Each node has 8 2.66GHz cpu
>>> cores, 24GB RAM and 8 1TB SATA drives behind a LSI (Fusion MPT) SAS
>>> 1068E controller. Interconnection is via one of 4 1GE interfaces,
>>> directly. Kernel is 2.6.22.18 and DRBD is 8.0.11, the storage device
>>> in question is a 3TB MD RAID5 spread across all 8 drives. The native
>>> results for this device using ext3 and bonnie for benchmarking are:
>>> ---
>>>  Version  1.03       ------Sequential Output------ --Sequential Input-
>>> --Random- -Per Chr- --Block-- -Rewrite- -Per Chr- --Block-- --Seeks--
>>> Machine        Size K/sec %CP K/sec %CP K/sec %CP K/sec %CP K/sec
>>> %CP  /sec %CP borg00a      50000M           120486  36 87998
>>> 17           535665  44 390.9   1 ------Sequential Create------
>>> --------Random Create-------- -Create-- --Read--- -Delete-- -Create--
>>> --Read--- -Delete-- files  /sec %CP /sec %CP  /sec %CP  /sec %CP  /sec
>>> %CP  /sec %CP 128 74265  90 +++++ +++ 83659 100 71540  88 +++++ +++
>>> 81619  99 ---
>>>
>>> The same test done on the resulting (UpToDate) drbd device:
>>> ---
>>> Version  1.03       ------Sequential Output------ --Sequential Input-
>>> --Random- -Per Chr- --Block-- -Rewrite- -Per Chr- --Block-- --Seeks--
>>> Machine        Size K/sec %CP K/sec %CP K/sec %CP K/sec %CP K/sec
>>> %CP  /sec %CP borg00a      50000M           41801  13 39659
>>> 11           413367  37 397.7   1 ------Sequential Create------
>>> --------Random Create-------- -Create-- --Read--- -Delete-- -Create--
>>> --Read--- -Delete-- files  /sec %CP /sec %CP  /sec %CP  /sec %CP  /sec
>>> %CP  /sec %CP 128 78847  95 +++++ +++ 86936  99 78722  95 +++++ +++
>>> 63054  76 ---
>>>       
>
>
>   


-- 
Igor Neves <igor.neves at 3gnt.net>
3GNTW - Tecnologias de Informação, Lda
 
 SIP: igor at 3gnt.net	JID: igor at 3gnt.net 
 ICQ: 249075444		MSN: igor at 3gnt.net
 TLM: 00351914503611	PSTN: 00351252377120





More information about the drbd-user mailing list