[openstack-dev] Quantum Performance
Juergen Brendel (jbrendel)
jbrendel at cisco.com
Wed Nov 7 18:28:24 UTC 2012
Dan,
Yes, that's exactly what it's going to be like. Sorry, I'm new top OpenStack development, so I did the fix already, but still need to figure out the gerrit side of things.
Yes, the code checks for the presence of a specialized count function and uses it if the plugin provides it.
I hope to have this available sometime today for review.
Juergen
From: Dan Wendlandt [mailto:dan at nicira.com]
Sent: Thursday, November 08, 2012 7:21 AM
To: OpenStack Development Mailing List
Subject: Re: [openstack-dev] Quantum Performance
On Wed, Nov 7, 2012 at 10:12 AM, Dan Wendlandt <dan at nicira.com<mailto:dan at nicira.com>> wrote:
I'd like to see something like this automated as part of the smoke tests, otherwise its easy for performance regressions to sneak in. I am not aware of this existing for a Folsom-based setup yet.
This seems to be the bug we're using to track: https://bugs.launchpad.net/quantum/+bug/1075369. I don't see a linked review on gerrit though. Have you not yet posted it?
Actually, looking at the code, the change is pretty simple. The only real trick is to handle the fact that a true "count_x" resource is not part of the plugin API, but would make sense given quotas. Since we can't change the plugin API after the fact for Folsom, it probably makes sense to have the quota code check if a count_x method exists, and if not, fall back to using the existing method. For any plugin that inherits from db_base_plugin_v2, we can provide default count_x methods that use sqlalchemy.
Dan
Dan
On Wed, Nov 7, 2012 at 7:56 AM, Debojyoti Dutta <ddutta at gmail.com<mailto:ddutta at gmail.com>> wrote:
In fact this points to the fact that we need a Geekbench equivalent for openstack so that folks can submit performance numbers for their scaled setups.
For starters, we could have simple metrics like the test here (for quantum) and repeat this for nova too (for starting a bunch of VMs).
The submitted numbers will also be a metric for new fixes/features which might have performance sideeffects
debo
On Wed, Nov 7, 2012 at 7:20 AM, Edgar Magana (eperdomo) <eperdomo at cisco.com<mailto:eperdomo at cisco.com>> wrote:
Team,
We ran a comparative tests between current Quantum code and our proposed fix for this bug and the results are incredible, in current quantum code it takes around 4-5 seconds to create a network when already have over 600 networks created and almost 7 second when we have over 1K networks. See this log:
Tue Nov 6 23:39:52 PST 2012
Created a new network:
+---------------------------+--------------------------------------+
| Field | Value |
+---------------------------+--------------------------------------+
| admin_state_up | True |
| id | e87b2f27-9308-4001-b180-228ff5ab6479 |
| name | net621 |
| provider:network_type | local |
| provider:physical_network | |
| provider:segmentation_id | |
| router:external | False |
| shared | False |
| status | ACTIVE |
| subnets | |
| tenant_id | b248e18b173b4ea99873e102a0715dcd |
+---------------------------+--------------------------------------+
Tue Nov 6 23:39:57 PST 2012
++++
Wed Nov 7 00:17:46 PST 2012
Created a new network:
+---------------------------+--------------------------------------+
| Field | Value |
+---------------------------+--------------------------------------+
| admin_state_up | True |
| id | 2138b910-8996-40b7-a9b4-ac090e126a2b |
| name | net1032 |
| provider:network_type | local |
| provider:physical_network | |
| provider:segmentation_id | |
| router:external | False |
| shared | False |
| status | ACTIVE |
| subnets | |
| tenant_id | b248e18b173b4ea99873e102a0715dcd |
+---------------------------+--------------------------------------+
Wed Nov 7 00:17:53 PST 2012
Wed Nov 7 00:17:53 PST 2012
Created a new network:
+---------------------------+--------------------------------------+
| Field | Value |
+---------------------------+--------------------------------------+
| admin_state_up | True |
| id | 29530fb3-1a1a-4c88-b9e4-954988bb8ccb |
| name | net1033 |
| provider:network_type | local |
| provider:physical_network | |
| provider:segmentation_id | |
| router:external | False |
| shared | False |
| status | ACTIVE |
| subnets | |
| tenant_id | b248e18b173b4ea99873e102a0715dcd |
+---------------------------+--------------------------------------+
Wed Nov 7 00:18:00 PST 2012
With the changes on the count call we have a constant performance regardless the number of networks created:
Wed Nov 7 02:48:59 PST 2012
Created a new network:
+---------------------------+--------------------------------------+
| Field | Value |
+---------------------------+--------------------------------------+
| admin_state_up | True |
| id | 024aa164-cb7b-4683-a5ae-41f11a582529 |
| name | net3847 |
| provider:network_type | local |
| provider:physical_network | |
| provider:segmentation_id | |
| router:external | False |
| shared | False |
| status | ACTIVE |
| subnets | |
| tenant_id | 5baba10045db44cdad235fff1d5e59b1 |
+---------------------------+--------------------------------------+
Wed Nov 7 02:49:00 PST 2012
Wed Nov 7 02:49:00 PST 2012
Created a new network:
+---------------------------+--------------------------------------+
| Field | Value |
+---------------------------+--------------------------------------+
| admin_state_up | True |
| id | 91e7eaa3-bdc4-4705-a8b5-2b63500fd066 |
| name | net3848 |
| provider:network_type | local |
| provider:physical_network | |
| provider:segmentation_id | |
| router:external | False |
| shared | False |
| status | ACTIVE |
| subnets | |
| tenant_id | 5baba10045db44cdad235fff1d5e59b1 |
+---------------------------+--------------------------------------+
Wed Nov 7 02:49:01 PST 2012
Wed Nov 7 02:49:01 PST 2012
Created a new network:
+---------------------------+--------------------------------------+
| Field | Value |
+---------------------------+--------------------------------------+
| admin_state_up | True |
| id | 19fd1785-77c2-4526-9fa6-5a2c8c01b3a4 |
| name | net3849 |
| provider:network_type | local |
| provider:physical_network | |
| provider:segmentation_id | |
| router:external | False |
| shared | False |
| status | ACTIVE |
| subnets | |
| tenant_id | 5baba10045db44cdad235fff1d5e59b1 |
+---------------------------+--------------------------------------+
Wed Nov 7 02:49:01 PST 2012
We will push this code changes ASAP.
Thanks,
Edgar
From: Edgar Magana <eperdomo at cisco.com<mailto:eperdomo at cisco.com>>
Reply-To: OpenStack List <openstack-dev at lists.openstack.org<mailto:openstack-dev at lists.openstack.org>>
Date: Tuesday, November 6, 2012 9:19 AM
To: OpenStack List <openstack-dev at lists.openstack.org<mailto:openstack-dev at lists.openstack.org>>, "gkotton at redhat.com<mailto:gkotton at redhat.com>" <gkotton at redhat.com<mailto:gkotton at redhat.com>>
Subject: Re: [openstack-dev] Quantum Performance
Hi Folks,
I just want to give you an update on this topic, we opened a bug against this behavior:
https://bugs.launchpad.net/quantum/+bug/1075369
Bug Description:
In quantum when a new network is created and system checks for user's quota, instead of getting a count from DB object, it returns all objects from db and locally counts them all. It has performance implications when the number of objects increases.
We will submit the patch ASAP.
Thanks,
Edgar
From: Salvatore Orlando <sorlando at nicira.com<mailto:sorlando at nicira.com>>
Reply-To: OpenStack List <openstack-dev at lists.openstack.org<mailto:openstack-dev at lists.openstack.org>>
Date: Wednesday, October 31, 2012 12:54 AM
To: "gkotton at redhat.com<mailto:gkotton at redhat.com>" <gkotton at redhat.com<mailto:gkotton at redhat.com>>, OpenStack List <openstack-dev at lists.openstack.org<mailto:openstack-dev at lists.openstack.org>>
Subject: Re: [openstack-dev] Quantum Performance
Edgar,
My suspicion here goes to the IP allocation mechanisms.
Hence, I'd try to run the following tests (they should be easily scriptable):
1) Create networks no subnets
2) Create networks, with subnet having a small CIDR (say /28)
3) Create networks with larger CIDRs (say /24, /20, /16 etc)
What behaviour do you observe in the above three cases?
Also, It might be worth shutting down agent to pinpoint whether the problem is in Quantum service itself on in the interactions with the various agents.
Salvatore
PS: thanks for doing these tests - They've been lying on my TODO list for way too much time!
If you don't hate me too much for the service insertion story, I'll be glad to buy you a drink at the next summit :)
On 31 October 2012 08:37, Gary Kotton <gkotton at redhat.com<mailto:gkotton at redhat.com>> wrote:
On 10/31/2012 09:48 AM, Edgar Magana (eperdomo) wrote:
Hi Folks,
I am running OpenStack Folsom with Quantum/OVS plugin. I am running some performance tests for Quantum, basically I am creating 5000 network with one subnet each one of them.
After the first ~50 - 70 networks the system response time is slower and slower, to the point that it could take up to ~6 - 7 seconds to create just a network and the same for the subnets.
Hi,
Is the problem you are seeing with the Quantum service or the agents? Off the bat I would say that it is with the Quantum service. Are the calls done in parallel or are they done sequentially?
There are a number of things that we should do to isolate this:
1. Check the internal logic of the network creation. Basically there are 3 stages:
i. provider network treatment
ii network create
iii. l3 notification
It would be interesting to see who eats the most time.
2. Need to profile database access
Thanks
Gary
I would like to know if somebody else has experimented this kind of behavior in Quantum or even in nova-network.
Thanks,
Edgar
_______________________________________________
OpenStack-dev mailing list
OpenStack-dev at lists.openstack.org<mailto:OpenStack-dev at lists.openstack.org>http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
_______________________________________________
OpenStack-dev mailing list
OpenStack-dev at lists.openstack.org<mailto:OpenStack-dev at lists.openstack.org>
http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
_______________________________________________
OpenStack-dev mailing list
OpenStack-dev at lists.openstack.org<mailto:OpenStack-dev at lists.openstack.org>
http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
--
-Debo~
_______________________________________________
OpenStack-dev mailing list
OpenStack-dev at lists.openstack.org<mailto:OpenStack-dev at lists.openstack.org>
http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
--
~~~~~~~~~~~~~~~~~~~~~~~~~~~
Dan Wendlandt
Nicira, Inc: www.nicira.com<http://www.nicira.com>
twitter: danwendlandt
~~~~~~~~~~~~~~~~~~~~~~~~~~~
--
~~~~~~~~~~~~~~~~~~~~~~~~~~~
Dan Wendlandt
Nicira, Inc: www.nicira.com<http://www.nicira.com>
twitter: danwendlandt
~~~~~~~~~~~~~~~~~~~~~~~~~~~
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openstack.org/pipermail/openstack-dev/attachments/20121107/371d17e0/attachment.html>
More information about the OpenStack-dev
mailing list