Re: [cso] [irtf-discuss] a new research proposal

[Greg Bernstein <[e-mail]>]

Hi all,
as to "Why isn't today's load balancing not good enough?", there are 
several types of load balancing: those that take place within data 
centers and those that take place between data centers. Within data 
centers we are seeing more "network aware" capabilities, e.g., virtual 
switches are now offering flow statistics [OpenVSwitch 
<http://openvswitch.org/>], and work on VM placements within data 
centers [1].  Between data centers the network capabilities tend not to 
be known, this is particularly true for technologies below the IP 
layer.  Without some type of interaction between application and network 
layer, one could never know, for example, that additional wavelengths 
(10Gbps or 40Gbps) may be available between two locations.
Note that in general there have been many theoretical optimization 
results reported in this area [2-4].

"I think it is important to justify how a tighter integration provides 
also benefits to the network operators, not only to the applications."  
This is key, if it doesn't benefit network operators it will not be 
implemented, if it doesn't benefit the application providers (and end 
users) it will not be used!

Best Regards

Greg B.

References:
[1] X. Meng, V. Pappas, and L. Zhang, "Improving the Scalability of Data Center Networks with Traffic-aware Virtual Machine Placement," in 2010 Proceedings IEEE INFOCOM, San Diego, CA, USA, 2010, pp. 1-9.
[2] S. Das and J. Kangasharju, "Evaluation of network impact of content distribution mechanisms," in Proceedings of the 1st international conference on Scalable information systems, Hong Kong, 2006, p. 35.
[3] S. Ratnasamy, M. Handley, R. Karp, and S. Shenker, "Topologically-aware overlay construction and server selection," in INFOCOM 2002. Twenty-First Annual Joint Conference of the IEEE Computer and Communications Societies. Proceedings. IEEE, 2002, vol. 3, pp. 1190-1199 vol.3.
[4] V. N. Padmanabhan and G. M. Voelker, with Lili Qiu, "On the placement of Web server replicas," in INFOCOM 2001. Twentieth Annual Joint Conference of the IEEE Computer and Communications Societies. Proceedings. IEEE, 2001, vol. 3, pp. 1587-1596 vol.3.



On 3/21/2011 2:40 PM, Leeyoung wrote:
> Hi Moritz,
>
> Thanks for your review of and interest in the draft and providing a set of excellent discussion points and questions.
>
> As you indicated, the benefit of cross application/network optimization applied to network operators as well as application providers. One of the key problems today is the lack of application-awareness by the networks and vice versa. So I believe the proposal would benefit both application and network and ultimately the end-users, especially for the mission-critical application users.
>
> In regard to the question why today's load balancing is not good enough, I think the question depends on the kind of application we are dealing with. For web-based applications including streaming video, etc, I think today's load balancing works well in most of cases. On the other hands, for mission-critical applications, today's load balancing may not be good enough to meet the requirement of critical end-user applications such as remote medical surgery, real-time interactive 3D video applications, mission critical financial transaction services, etc. All these application should be provided with some level of resource guarantee above and beyond "best effort."
>
> For your question on why not reusing ALTO mechanisms, I am for it actually if the ALTO mechanisms can provide solutions to the requirements. One thing ALTO may be lacking is the exchange of network abstraction information --- no network operators want to reveal detail network topology and resource (e.g., available bandwidth) information. So we need to investigate network abstraction mechanism and "virtual" network topology exchange, etc., as part of the research. But I agree with you that, we can reuse existing mechanisms where possible.
>
> Thanks,
> Young
>
> -----Original Message-----
> From: [e-mail] [mailto:[e-mail]] On Behalf Of Steiner, Moritz (Moritz)
> Sent: Monday, March 21, 2011 3:17 PM
> To: [e-mail]
> Subject: Re: [irtf-discuss] a new research proposal
>
> Hello Young,
>
> I read your proposal with a lot of interest. It certainly makes a lot of sense to tie all type of resources together: computing, storage and network resources.
> You raise a couple of very interesting questions: How to enhance the quality of application experience? How to enhance the resilience of the running applications? What implications is the load balancing on the application level going to have on the network engineering happening at a lower level at different time scales?
> Besides those points I think it is important to justify how a tighter integration provides also benefits to the network operators, not only to the applications. Why isn't today's load balancing not good enough?
> As you have stated the scope of ALTO is much more limited than what you're proposing, but wouldn't it make sense to reutilize some concepts introduced by ALTO, such as the network map and cost map?
>
> Best
> Moritz
>
>
>
>


-- 
===================================================
Dr Greg Bernstein, Grotto Networking [phone]