[Gluster-users] GlusterFS performance

Yavor Marinov ymarinov at neterra.net
Thu May 23 14:16:42 UTC 2013


Here is the output of volume profile info after 1.1Gb file is copied 
over mounted glusterfs on client:

[root at gfs1 data]# gluster volume profile test info
Brick: 93.123.32.41:/data
-------------------------
Cumulative Stats:
    Block Size:             131072b+
  No. of Reads:                    0
No. of Writes:                 8192
  %-latency   Avg-latency   Min-Latency   Max-Latency   No. of 
calls         Fop
  ---------   -----------   -----------   ----------- 
------------        ----
       0.00       0.00 us       0.00 us       0.00 us 1     RELEASE
       0.02     114.00 us      35.00 us     208.00 us 5      LOOKUP
       0.26    9306.00 us    9306.00 us    9306.00 us 1      CREATE
       0.75   26260.00 us   26260.00 us   26260.00 us 1       FLUSH
      98.97     424.91 us     165.00 us   14056.00 us 8192       WRITE

     Duration: 18352 seconds
    Data Read: 0 bytes
Data Written: 1073741824 bytes

Interval 14 Stats:
    Block Size:             131072b+
  No. of Reads:                    0
No. of Writes:                 5353
  %-latency   Avg-latency   Min-Latency   Max-Latency   No. of 
calls         Fop
  ---------   -----------   -----------   ----------- 
------------        ----
       0.00       0.00 us       0.00 us       0.00 us 1     RELEASE
       1.22   26260.00 us   26260.00 us   26260.00 us 1       FLUSH
      98.78     397.60 us     169.00 us   13423.00 us 5353       WRITE

     Duration: 172 seconds
    Data Read: 0 bytes
Data Written: 701628416 bytes

[root at gfs1 data]#


---
Find out about our new Cloud service - Cloudware.bg 
<http://cloudware.bg/?utm_source=email&utm_medium=signature&utm_content=link&utm_campaign=newwebsite>
Access anywhere. Manage it yourself. Pay as you go.
------------------------------------------------------------------------
*Yavor Marinov*
System Administrator

Neterra Ltd.
Telephone: +359 2 975 16 16
Fax: +359 2 975 34 36
Mobile: +359 888 610 048
www.neterra.net <http://www.neterra.net>


On 05/23/2013 04:54 PM, Michael Brown wrote:
> That's only a single data point (one LOOKUP call) - this tells you 
> nothing.
>
> You need to check after it's been running (and processing traffic) for 
> a while.
>
> For example, here's the stats off one of my bricks (an SSD):
> Brick: fearless2:/export/bricks/500117310007a84c/glusterdata
> ------------------------------------------------------------
> Cumulative Stats:
>    Block Size:                 32b+ 64b+                 128b+
>  No. of Reads:                    0 0                     1
> No. of Writes:                    1 5634                  4252
>
>    Block Size:                256b+ 512b+                2048b+
>  No. of Reads:                    0 1                     0
> No. of Writes:                  343 24                     1
>
>    Block Size:               4096b+ 8192b+               16384b+
>  No. of Reads:                    8 7                    10
> No. of Writes:                    4 0                     0
>
>    Block Size:              32768b+ 65536b+              131072b+
>  No. of Reads:                   25 165                   436
> No. of Writes:                    2 7                    36
>
>  %-latency   Avg-latency   Min-Latency   Max-Latency No. of 
> calls         Fop
>  ---------   -----------   -----------   ----------- 
> ------------        ----
>       0.00       0.00 us       0.00 us       0.00 us            
> 120      FORGET
>       0.00       0.00 us       0.00 us       0.00 us           
> 2940     RELEASE
>       0.00       0.00 us       0.00 us       0.00 us           4554  
> RELEASEDIR
>       0.00     104.00 us     104.00 us     104.00 us              1    
> TRUNCATE
>       0.00     125.00 us     110.00 us     140.00 us              
> 2     XATTROP
>       0.01      12.62 us       8.00 us      25.00 us            
> 146      ACCESS
>       0.04      81.77 us      65.00 us     112.00 us             60    
> SETXATTR
>       0.05      30.99 us      25.00 us      57.00 us            
> 212     SETATTR
>       0.06      12.73 us       8.00 us      59.00 us            
> 574     INODELK
>       0.09     188.43 us     140.00 us     244.00 us             
> 60      CREATE
>       0.11      25.24 us      15.00 us     149.00 us            
> 533      STATFS
>       0.12     260.72 us     206.00 us     430.00 us             
> 60       MKDIR
>       0.26      11.34 us       5.00 us     127.00 us           
> 2925       FLUSH
>       0.27      15.14 us       7.00 us      90.00 us           
> 2274     ENTRYLK
>       0.36     102.52 us      81.00 us     161.00 us            
> 442       RMDIR
>       0.63      27.61 us      17.00 us     606.00 us           
> 2880        OPEN
>       0.76     171.58 us      91.00 us    5691.00 us            
> 555      UNLINK
>       0.87      22.66 us       8.00 us     469.00 us           
> 4812     READDIR
>       0.87      24.37 us      10.00 us    1302.00 us           
> 4506        STAT
>       0.94      61.67 us      16.00 us     194.00 us           1917    
> GETXATTR
>       1.06      51.20 us      10.00 us     224.00 us           
> 2600       FSTAT
>       1.14      31.46 us      18.00 us    1016.00 us           
> 4554     OPENDIR
>       2.56      31.19 us      18.00 us    4373.00 us          
> 10304       WRITE
>       2.58     417.28 us      15.00 us    1860.00 us            
> 776        READ
>       3.64      17.26 us       6.00 us    4824.00 us          26507    
> FINODELK
>      24.03     146.42 us      49.00 us    9854.00 us          20622    
> FXATTROP
>      26.67     652.70 us      42.00 us   89705.00 us           5134    
> READDIRP
>      32.86     128.19 us       9.00 us    4617.00 us          
> 32204      LOOKUP
>
>
> On 13-05-23 09:03 AM, Yavor Marinov wrote:
>> I've just enabled profiling of the volume and this is the information 
>> from the profile info printed:
>>
>> [root at gfs1 ~]# gluster volume profile test info
>> Brick: 93.123.32.41:/data
>> -------------------------
>> Cumulative Stats:
>>  %-latency   Avg-latency   Min-Latency   Max-Latency   No. of 
>> calls         Fop
>>  ---------   -----------   -----------   ----------- 
>> ------------        ----
>>     100.00     148.00 us     148.00 us     148.00 us              
>> 1      LOOKUP
>>
>>     Duration: 13950 seconds
>>    Data Read: 0 bytes
>> Data Written: 0 bytes
>>
>> Interval 4 Stats:
>>
>>     Duration: 7910 seconds
>>    Data Read: 0 bytes
>> Data Written: 0 bytes
>>
>> [root at gfs1 ~]#
>>
>> Anything here that might be useful ?
>>
>>
>> ---
>> Find out about our new Cloud service - Cloudware.bg 
>> <http://cloudware.bg/?utm_source=email&utm_medium=signature&utm_content=link&utm_campaign=newwebsite>
>> Access anywhere. Manage it yourself. Pay as you go.
>> ------------------------------------------------------------------------
>> *Yavor Marinov*
>> System Administrator
>>
>> Neterra Ltd.
>> Telephone: +359 2 975 16 16
>> Fax: +359 2 975 34 36
>> Mobile: +359 888 610 048
>> www.neterra.net <http://www.neterra.net>
>>
>>
>> On 05/23/2013 01:10 PM, Явор Маринов wrote:
>>> I've made a mistake we are using 30Mbit connectivity on all of the 
>>> nodes. Below is a iperf test between the node and the client
>>>
>>> [root at gfs4 ~]# iperf -c 93.123.32.41
>>> ------------------------------------------------------------
>>> Client connecting to 93.123.32.41, TCP port 5001
>>> TCP window size: 23.2 KByte (default)
>>> ------------------------------------------------------------
>>> [  3] local 93.123.32.44 port 49838 connected with 93.123.32.41 port 
>>> 5001
>>> [ ID] Interval       Transfer     Bandwidth
>>> [  3]  0.0-10.1 sec  49.9 MBytes  41.5 Mbits/sec
>>> [root at gfs4 ~]#
>>>
>>> But when trying to copy a 1Gb file on the client's mounted volume 
>>> the speed between the client and the node is ~500kb/s
>>>
>>>
>>> ---
>>> Find out about our new Cloud service - Cloudware.bg 
>>> <http://cloudware.bg/?utm_source=email&utm_medium=signature&utm_content=link&utm_campaign=newwebsite>
>>> Access anywhere. Manage it yourself. Pay as you go.
>>> ------------------------------------------------------------------------
>>> *Yavor Marinov*
>>> System Administrator
>>>
>>> Neterra Ltd.
>>> Telephone: +359 2 975 16 16
>>> Fax: +359 2 975 34 36
>>> Mobile: +359 888 610 048
>>> www.neterra.net <http://www.neterra.net>
>>>
>>>
>>> On 05/23/2013 12:16 PM, Nux! wrote:
>>>> On 23.05.2013 09:41, Явор Маринов wrote:
>>>>> Thanks for your reply.
>>>>>
>>>>> No matter how many nodes (currently the volume is only with its own
>>>>> node) the speed is really slow. For testing purposes, i made a volume
>>>>> with only one node, without any replication - however the speed is
>>>>> still ~500kb/s. The cloud servers are limited to 30Gbit/s but still
>>>>> the traffic when writing to the node is ~500kb/s
>>>>>
>>>>> i'm using 3.3.1 glusterfsd with kernel 2.6.18-348.el5xen and i need
>>>>> to know if the the problem is within the kernel.
>>>>
>>>> I don't think it is a problem with gluster; I never used el5 for 
>>>> this, but I doubt there's an inherent problem with it either. That 
>>>> speed limit looks odd to me and I think it's somewhere in your setup.
>>>> Have you done any actual speed tests in the VMs?
>>>>
>>>
>>>
>>>
>>> _______________________________________________
>>> Gluster-users mailing list
>>> Gluster-users at gluster.org
>>> http://supercolony.gluster.org/mailman/listinfo/gluster-users
>>
>>
>>
>> _______________________________________________
>> Gluster-users mailing list
>> Gluster-users at gluster.org
>> http://supercolony.gluster.org/mailman/listinfo/gluster-users
>
>
> -- 
> Michael Brown               | `One of the main causes of the fall of
> Systems Consultant          | the Roman Empire was that, lacking zero,
> Net Direct Inc.             | they had no way to indicate successful
> ☎: +1 519 883 1172 x5106    | termination of their C programs.' - Firth

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://supercolony.gluster.org/pipermail/gluster-users/attachments/20130523/ca8c6028/attachment.html>


More information about the Gluster-users mailing list