[Fiware-cloud] Conslidated highlights from the F2F in Turin

Alex Glikson GLIKSON at il.ibm.com
Tue Oct 4 16:11:45 CEST 2011


Highlights and action items from the F2F at Turin 
General -- main accomplishments: 
Socializing among team members, including additional team members which 
did not attend the kick-off in May. Improved team atmosphere and trust. 
Gained common understanding of the architecture, main assets, interfaces, 
potential integration challenges and gaps -- for the entire WP as a whole, 
as well as for individual GEs 
Identified several significant gaps (see below), some of which may become 
topics for open calls 
Conducted highly productive joint sessions with other WPs (see more 
details below) 
Met in person key representatives of other teams, thus improving future 
collaboration 
Gained better understanding of what other WPs are doing 
Identified and verified potential integration points and assets/GEs that 
can be leveraged
Cloud Edge, Cloud Edge RM
Reviewed the architecture of the Cloud Proxy 
Reviewed end-to-end scenario of provisioning of an application hosted in 
the cloud, having a component designed to run close to the end-user (on a 
Cloud Proxy) 
Refined WP architecture to address the distinction between the Cloud Proxy 
device (running outside of the centralized cloud infrastructure) and the 
management component that would typically run within the centralized 
infrastructure and act as a proxy between the centralized cloud management 
stack and the numerous Cloud Proxy devices located close to the end-uses. 
Identified a gap in providing the above management component. 
Action Item [Serge]: find out whether this gap can be contained (e.g., by 
re-balancing the Cloud Proxy related work between WP4 and WP7). 
IaaS Service Management
It should be possible to design services that comprise components that run 
on Cloud Proxy (and potentially also end-devices -- see comment from I2ND 
interlock), as well as object storage resources. This should be reflected 
in service manifest, as well as handled during provisioning and life cycle 
management of services. It must be studied.
A service composition tool should be part of the ecosystem provided to the 
application developer/provider. Should be considered for the Open Call. A 
Service composition tool should be part of the Portal or a specific tool 
provided by the specific WP within FI-WARE (TBD). IMHO It is not part of 
the SM. 
Scalability and security aspects of Claudia should be considered. The 
security aspects must be applied to all GE. BTW, we need more discussion 
about this topics due to we need to clarify all the security requirements.
How do we store/mngt. the credentials (e.g. EC2 Amazon)? [AI Fernando]
What is the possibility of scaling Claudia? [AI Fernando]
It is possible to update the service manifest dynamically (I suppose yes) 
[AI Fernando]
There are not trivial dependencies between Claudia and IaaS Data Center 
Mngt.
Why do we need to use clone VMs? Can we provide some UCs?
Integration with SLA-SLO. Which type of SLA we are using. How Claudia 
manages the SLAs?
Interoperability between Cloud Standard API (DMTF CIMI) and OCCI through 
OVF. Especify the OVF exchanges messages in common (the same for CDMI 
could be applied) [AI Fernando Andy Alex]
OGF33: Venus-C (venus-c.eu) has a OVF to OCCI tool
France telecom participation in WP4 discussed
Gap related to billing/accounting noted 
PaaS
The current approach for PaaS provisioning is to install middleware 
components into an existing running OS (with Chef), which has been 
installed separately (in our case -- as part of the provisioning of the 
corresponding VM). A unified approach should be considered, in which the 
VM images already include the middleware components pre-installed, and 
they just need to be customized during provisioning
How can we specify the elasticity rules in terms of PaaS? [AI Fernando]
Why do we use Chef for the PaaS? [AI Fernando] 
IaaS DataCenter Resource Management
Reviewed WP architecture, mapped assets and gaps 
A need for common monitoring infrastructure was stressed 
The resource management layer should provide capability of orchestrating 
operations across groups of VMs -- e.g., in a scenario of service 
snapshot. 
We should consider supporting non-virtualized workloads
proximity policies
translate the requirements of the SM to the policies of the CDRM
number of WP7 folks are interested in Network-as-a-Service, something that 
OpenStack/Quantum can provide ? @andy to discuss further with respective 
WP7 people  
Object Storage
Cloud Standard API - CMDI - OVF interoperability 
Action Item [Fernando]: CDMI interface specification of the SM
@andy Further investigations required for CDMI on swift
Self-Service Interfaces
A collaboration-driven approach to Cloud Portal was presented 
Action Item [Alex, Joaquin]: assess whether this can be contained within 
the existing funding. It might make sense to focus on have a 
fully-functional cloud portal as a subject for an Open Call.
where is the user management? permission, accounting, etc?
How to manage the SSO? and how to translate it to each GE?  
Monitoring
Monitoring architecture presented
Noted that Analysis should cover both online (e.g. CEP) and offline (e.g. 
MapReduce)
Noted that Storage should cover both optimized time series storage (e.g. 
rrdtool, graphite) and BigData storage (e.g. Hadoop/HDFS)
Metric sources should have the possibility to be configured (policies) 
from a central location 
Visualisation noted as a deficiency and possibly subject to an open call
Security
Presented extracted security requirements
Andy to continue as mediator between WP4 and WP8
Representative pleased to see WP4?s consideration of requirements
@all Plan to 1st specify WP4 EPICs and then add/modify to account for 
security requirements
Noted that OpenStack keystone can be considered as an asset providing SSO
Interlock with I2ND 
It was mentioned that in some cases a cloud application may include a 
component designed to run at the end-user device (e.g., a mobile phone) -- 
either within the Web browser, or as a 'rich client' component. In a way, 
this can be considered as another 'tier', besides the centralized cloud 
infrastructure and the cloud proxy. This should be taken into account when 
designing the service manifest, as well as during provisioning and service 
life cycle management.
Interest in Network as a Service from WP7 people. Need to understand 
better integration point here.
Interlock with Security
Reviewed security considerations which might apply to Cloud Hosting. How 
to protect the services in the Cloud? 
Action Item [Andy to coordinate]: need to refine our security 
considerations in the HLD (once transferred to the Wiki)
It was noted that the security WP is interested in leveraging the common 
infrastructure for monitoring and CMDB. Cloud Hosting does not intend 
having a centralized CMDB, but having a single monitoring system which can 
be augmented with security probes definitely makes sense. 
Action Item [Alex] need to follow-up on monitoring infrastructure, across 
GEs and WPs
Interlock with Apps
In order to drive business processes related to applications deployed in 
the cloud (e.g., SLAs), resource-related metrics produced by Cloud WP 
components must be accessible by the components of the Apps WP. 
Mashup execution engines might be hosted on the cloud (PaaS would probably 
make more sense than IaaS)
Mashups are composite services. As all other services they should adhere 
to SLAs. To protect end-to-end SLAs of mashups, run-time information about 
performance of the services comprising a mashup should be provided  to the 
Cloud hosting resource optimization GE. Note that this information is 
different from the static USDL data. Kalin [?], took it as an action point 
to verify whether this information can be extracted from the mashup 
execution engine in a reasonable fashion.
We have service manifests (i.e., template) for stand alone services. Do we 
have a reciprocal for mashups? [Kalin, Fernando]
The services comprising mashups should be deployed and running either in 
FiWare or outside of it, but they should exist and run as a prerequisite. 
If services are deployed on FiWare, then resource management GE can 
optimize their placement and resource allocation from a global 
perspective.
There is a strong connection to networking. Mashups' performance 
critically depends on the network performance. Cloud Hosting should be 
network performance aware to manage infrastructure for mashups. 
There is a need from the WP to have WP4 services described as USDL 
Interlock with Data
In order to develop a common monitoring system, 3 components have been 
identified that will be provided by the Data Management WP: pub/sub 
broker, Big Data and Complex Event Processing. Need to follow-up on 
architecture, requirements, interfaces, schedule, etc.
Presented monitoring architecture using GEs as building blocks


-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.fiware.org/private/old-fiware-cloud/attachments/20111004/816a1b0e/attachment.html>


More information about the Old-Fiware-cloud mailing list

You can get more information about our cookies and privacy policies clicking on the following links: Privacy policy   Cookies policy