Princeton S* Network Systems - Latest Comments

Re: Blog Name -> Dirty Slate Design

dealzbydesign — Mon, 23 Nov 2009 01:11:57 -0000

What's in a name? Hmmm, sounds intriguing at first. But..."...the ultimate goal should always be impact...". Very interesting now.

Re: CoralCDN Lesson: Fair-sharing bandwidth via admission control

mfreed — Fri, 20 Nov 2009 19:47:10 -0000

Rade,

Thanks for the comment. The website for hamilton.ie seems offline, so I can't currently access your paper (Google HTML cache is a bit hard to read).

That said, I haven't thought much about formally formulating the problem, much actually solving it, but I could imagine specifying the function something like this:

For a given node:
maximize \sum_{\forall_i} u (c_i)
subject to
c_i >= 0 // allocated capacity to site i is non-negative
c_i <= d_i // site i's allocated capacity <= demand
\sum_{\forall_i} c_i <= C // allocated capacity doesn't exceed node's total capacity

and model the utility function u as some concave function (say log x).

Now, this doesn't say anything about performance, but it does say that, for a given node, it's better to allocate some marginal bandwidth to a new domain rather than allocate more resources to one existing one (subject to available capacity), given the concave utility. Now if you extend this over all nodes, reasoning about a site's utility function taking as input \sum capacity c_i,j \forall node's j, subject to all available capacities, that might be a reasonable formulation.

Your thoughts are welcome.

Re: Computing for Global Development

isagenix01 — Fri, 20 Nov 2009 02:41:10 -0000

such a great article. Computing for Global Development Workshop. With great local innovations from the Honey Bee Network. Isagenix

Re: CoralCDN Lesson: Fair-sharing bandwidth via admission control

Rade Stanojevic — Thu, 19 Nov 2009 10:45:36 -0000

The UCSD paper on Distributed Rate Limiting deals with the problem of controlling the aggregate usage of a service (service = domain, in your context) across a set of distributed nodes so that performance (reject rate) at each node is equalized. Their solution indeed uses gossiping to collect the aggregate (accounting) usage that is used as a feedback on when to discard packets/requests.

In this paper, we show that in fact one can solve the same problem, without collecting the aggregate info)and without any centralized controller). Namely, if every node updates its local limit based on its performance (QoS) and the performance of its few neighbors, eventually the system should converge to the state at which performance is equalized across all nodes.

However, the problem you are facing in CoralCDN with multiple domains sharing finite resources seems way more challenging. Actually, it is not obvious to me what the objective would be and how to formalize the problem of bandwidth sharing in such context. Do you have any thoughts on this? I suspect that for appropriate performance objective, one might not need per node/domain accounting, but could rely on some adaptive decentralized controller to drive the system to the desired state with minimal communication overhead.

Re: Blog Name -> Dirty Slate Design

webdesigneruk — Wed, 18 Nov 2009 10:35:52 -0000

Great ideas, really enjoyed reading this

Re: CoralCDN Lesson: Fixing overlooked assumptions in DHTs

adslviettelvietteladsl — Mon, 16 Nov 2009 06:01:31 -0000

Hi, nice to meet you, my site provide DSL service, i hope you can get some information interesting in my site .

Re: CoralCDN Lesson: The great naming conflation of the Web

trusts — Fri, 13 Nov 2009 20:08:35 -0000

Mike, that's a good point. I was listening to the radio last evening, and there is a group trying to do a .GAY top level domain. I thought it was interesting, because the person who is promoting that idea owns a for-profit business that plans on purchasing most of the popular domain names if .GAY gets the go ahead.

Re: CoralCDN Lesson: The great naming conflation of the Web

mfreed — Fri, 13 Nov 2009 19:27:30 -0000

I'm not too worried about that. Much is just domain squatting anyway...

Re: Post-doc opportunity

Junaid Chaudhry — Mon, 09 Nov 2009 06:07:14 -0000

G'day Mike
I am working in the Computer Science and Engineering department at the University of Hail (Saudi Arabia) and have a very keep interest in Self Management and Self Organizing Systems. I have worked for about 6 years in u-Frontier project in South Korea. I wonder if there is any opportunity where we can remotely collaborate ? I have had a chance to work on Collaborative Processing in Sensor networks project (sponsored by Ministry of Science and Technology, South Korea), u-Frontier (U2: System Integration Team) where we worked on domestic gateway using prosyst technologies and developed bundles for elderly people and childcare, Autonomic Middleware for Network management in ubiquitous networks and my last depotation was at University of Trento in OKKAM project (sponsored by EU comission) with emphasis on Entity Lifecycle Management.

I think I can be useful. Please page me (if interested) on jDOTchaudhryATuohDOTeduDOTsa

Thanks.

Re: CoralCDN Lesson: The great naming conflation of the Web

trusts — Fri, 06 Nov 2009 22:13:45 -0000

Do you think it's true that the internet is slated to run out of domain names next year?

Re: The Design of CoralCDN

Artezanal — Mon, 02 Nov 2009 18:15:39 -0000

This topic is very good and is very interesting, I hope that you write more about it, thanks.

http://artezanalnet.com.br/blog/

Re: CoralCDN Lesson: Fair-sharing bandwidth via admission control

nyccondoman — Wed, 28 Oct 2009 10:53:10 -0000

Thank you for the article. I was unaware of CoralCDN before reading this.

Re: Blog Name -> Dirty Slate Design

ftasatellite — Wed, 21 Oct 2009 20:55:31 -0000

Dirty Slate is definitely a interesting name.

Re: CoralCDN Lesson: The great naming conflation of the Web

mfreed — Wed, 07 Oct 2009 11:20:14 -0000

Hi oobx,

Actually, the cookie issue is much less a security issue if you are a website that is trying to explicitly use CoralCDN for cached content. You should just specify that your code uses the full origin name when setting cookies: www.yoursite.com.nyud.net, instead of just setting a default of the domain.tld (i.e., nyud.net) for "ease of use". This is good security practice anyway: the principle of least privilege and all. Then a user from evil.com.nyud.net can't read cookies set to www.yoursite.com.nyud.net, as it fails the same origin policy check.

The problem I raise above is more when a website is being accessed by a Coralized URL and they are not similarly security conscious, so that they default to using the domain.tld, instead of the full origin name.

Let me know if that assuages your concern.

Re: CoralCDN Lesson: The great naming conflation of the Web

oobx — Sat, 03 Oct 2009 00:42:04 -0000

I was just turned on to Coral and was pretty jazzed about it until I read your post. Google app engine as a CDN is what led me to Coral. GAE should not be subject to such security issues. I've not read your other posts; so please excuse my ignorance of firecoral, etc.

In trying to comprehend the scope of the security issues you raise, I conclude that only cookies set by nyud.net-cached content are vulnerable. So, I just use coral cache for images and truly static content.

But, what's to stop evildoer from linking to my script that sets cookies? Nothing. But, how would he gain the trust of the user in order for the user to click on the nyud.net link? Then, how would evildoer track that click and convince the user to go to the malicious site to hijack data?

Coral CDN sounds like a great asset for bandwidth-poor folks. I hope you can improve upon it. As is, it seems very workable so long as developers understand the caveats such as security and the potential to skew statistics.

Thanks for raising the issue.

Re: Blog Name -> Dirty Slate Design

amexi — Sun, 16 Aug 2009 03:58:15 -0000

Hi mike..
It's a good and fresh name...

Re: CoralCDN Lesson: Fixing overlooked assumptions in DHTs

viettel_adsl — Tue, 04 Aug 2009 07:58:30 -0000

Hi! I'm Jake Bunce, the manager of Viettel ISP at http://www.adslviettel.com, and I think your post is awesome. It's hard to find quality information like this, I'm glad i found this, thanks for the valuable information.
But I have a suggestion: In my opinion the posts font and size is not the best typo for read. It is very uncomfortable.
Anyway, good work!

Re: CoralCDN Lesson: Fair-sharing bandwidth via admission control

advnet — Mon, 03 Aug 2009 10:13:51 -0000

very nice article Thanks For Sharing nice Chart also its Really know about This....

Re: CoralCDN Lesson: Fair-sharing bandwidth via admission control

Technology forum — Thu, 09 Jul 2009 23:43:18 -0000

thanks for this article.it helped in my project

Re: Security mechanisms in CoralCDN (and some attacks)

Stephen Omondi — Fri, 03 Jul 2009 03:00:23 -0000

Hello, I am an undergrad student in Kenya in my final year and woking on the research topic of Distributed System Security (Authorization and Authentication. I am not well versed with what a project involves and really wish to get help on the same. Could we be in touch so you can give me some guidance. I would really like to get some light shed on this area.

Re: Blog Name -> Dirty Slate Design

Media Contour — Wed, 01 Jul 2009 15:37:52 -0000

Dirty Slate is a very interesting and catchy name for your blog. Wiping the slate clean is often the best way to get a fresh start, allowing your ideas and creativity to flow unhampered. Great insight!

Re: Blog Name -> Dirty Slate Design

Adffiliate — Sun, 28 Jun 2009 02:57:47 -0000

"Dirty Slate design " VERY INTELLIGENT UNIQUE NAME

I hope it will be one of the most successful projects
Cheers

Re: Firecoral @ IPTPS

Adffiliate — Sun, 28 Jun 2009 02:29:56 -0000

Good News
Waiting the firefox extention to be available for download free vectors

Re: CoralCDN Lesson: Accepting conservatively and serving liberally

mfreed — Tue, 02 Jun 2009 20:57:03 -0000

Hi Om,

Well, anytime you appear to be a public-facing proxy cache, you are subject to DMCA notices. And sometimes even when you aren't, like when you are actually a printer. The short response is that yes, we've had to deal with DMCA notices. The typical answer I give appears in our FAQ:

CoralCDN does not provide archival storage of content, like google.com's cache or archive.org. Much like a web cache or "content accelerator" at ISPs, CoralCDN only keeps data temporarily in its file caches, either until the data expires or it is evicted (as may occur for unpopular data). As described above, CoralCDN will serve data for some maximum fixed period (24 hours) before checking back with the origin website. If the content at that site has changed, CoralCDN will fetch the new content afresh, replacing the old. If the origin site is no longer online or the particular content returns some HTTP error message, CoralCDN will only serve the old data for a short time (24 hours).

Thus, if you believe that a website is making infringing content available, please direct any notices to that particular website. Recall that CoralCDN's naming method makes it obvious the identity of the origin website: A Coralized URL of the form http://www.example.com.nyud.net/ corresponds to an origin site http://www.example.com/ .

If/When that origin site complies with the notice, the content in question will naturally be removed from CoralCDN's caches through purely automated technical means in at most 24 hours.

My understanding is that the DMCA requires that content be taken down with some reasonable time period, which CoralCDN's expiry times satisfy. In the past, this explanation and the resulting system behavior has satisfied content owners.

Re: CoralCDN Lesson: Accepting conservatively and serving liberally

Omprakash Gnawali — Mon, 01 Jun 2009 22:53:13 -0000

If a site has gone down in response to DMCA take-down notice, can "serving liberally" result in proxies also being subject to DMCA notices? The technical argument here says that the original server did not return the "GONE" response but sending "GONE" response is probably not legally mandated.