Blog posts

Summary: Split/Merge-System Support for Elastic Execution in Virtual Middleboxes, NSDI’13

Split/Merge presents an abstraction for virtual middleboxes that splits the state among replicas and dynamically re-balances both existing and new flows. They classify the middleboxes’ state into internal and external, and the external state is further divided into partitioned and coherent (global.. E.g., stats counter). All the three types of state are stored in each replica— the internal state need not be synchronized while the external state must be consistent (strong or eventual) among all replicas.  They create an application level library called Freeflow that supports creating and destroying of VMs while balancing the load (scaling-out and -in). They show the performance improvement by using an IDS and two other synthetic middleboxes. Section 2 gives an excellent anatomy of a typical middlebox and the overview of these three middleboxes used.

Split/Merge has four components. One, an application library that handles state related challenges (state classification and partitioned state must be moved between replicas etc). Second, an SDN controller to reflect the forwarding path changes (to route the traffic to appropriate replica). Third, an orchestrator that implements an elasticity policy which decides when to scale and when to migrate flows. Four, a VM manager that implements an actual creation and deletion of middlebox replicas.


Ugh! Another government to control the Internet :-(

Before going further, know that I have no relation with the ISPs nor that I care for them. Net neutrality (or I would call it ‘Internet socialism’) is not what many people think what it is for the future.

Enforcing net neutrality is not going to help but bring more danger to the Internet users! As computer networking person, as I explain later who is governing the Internet fairness, net neutrality is not going to bring fairness for the Internet users.

Now, how is that? You see, the big players who support net neutrality are Microsoft, Google, Facebook, Netflix etc. These are the content providers who benefit by regulating the rules for broken Internet. They are frightened and anxious that they cannot compete with other ISPs such as AT&T, Verizon and T-Mobile, because they think that these telecom providers further their own content. While the aforementioned industry giants are already a monopoly in their own respective disciplines, they want to be competitive in every field that they want to be and the latter giants are not letting them to do so. This has eventually become a big political drama and influenced normal people who are not aware of what Internet is and how it was built from the ground.

The Internet is a complex ecosystem that is built from protocols and standards that are carefully designed by the researchers and industry folks. Now what you are craving for is “Fairness”, a fair flow of Internet content which is not at all guaranteed by net neutrality. Then what is controlling this fairness? not the FCC; not the industry; not the users. The Internet in itself is governing its fairness. The Internet is a layered stack, where each protocol strives to make sure this fairness is achieved among all the users — In particular, a transport protocol called TCP (Transmission control protocol) prevents any end-point in taking advantage of the data transmission. It simply restrict the amount of content that an end-point can transmit over the wire. This one protocol has its own research for at least from the beginning of the invention of the Internet and is still active. Similarly, there are several other protocols and algorithms that the Internet is fortified of. Now let me go to what exactly you might be thinking in your mind — that what if an industry giant like AT&T goes viral and create it’s own Internet protocols? Let me tell you how that is not possible.

Wait for more ….. In Construction :))

Summary: Probius-Automated approach for VNF and service chain analysis, SOSR’18

The paper argues that there is lot of performance uncertainty in NFV environments and there is no systematic framework to identify the root cause of performance problems in VNFs. Although, there is some prior work on improving the performance of VNFs, they are focused on individual elements of NFV infrastructure. The performance problems are mainly at: VNF services, virtual switch, hypervisor, hardware. Moreover, they show that the throughput changes significantly with different network chains or adding more NFs, which makes it hard for network operators to distinguish which VNF is causing the problem.

So, they take an approach that systematically understand the NFV platforms and behaviours of NFV service chains. Then, they create various service chains based on some configurations and resource constraints, which provides insights about performance issues before the VNFs are deployed. Next, they build a comprehensive monitoring and tracing tool for all the NFV entities. Finally, they develop a performance analyzer to understand the root cause of performance issues using the collected traces.

The service chain builder generates all possible combinations of service chains while VNF manager creates these service chains based on all possible resource constraints. The workload manager emulates network workloads based on the given configurations from network operator. The configurations from operators are range of resource constraints, workload configurations, VNF configurations such as type and CPU memory constraints, and lastly service chaining policies such as which VNFs are together or followed by which.

To understand the NFV environment, they examine the VNFs workflow/relation with the underlying hypervisor and service chains. This investigation extracts important tracing points at all entities of an NFV platform, that affect the performance of VNFs. They use three monitors to collect these features and build graph based representation to maintain the workflow synchronization. After collecting the data, they first filter the outliers and then identify the suspicious service chains with anomaly detection (Cook’s distance, regression analysis). Finally, they analyse the performance individual VNFs as well as difference combinations of service chains and reason about common causes of performance issues.

change number of ticks matplotlib

I know many graduate students spend most of their time on plotting :)) And often you would have to change the frequency of xticks or yticks or their labels. Here is what I have to change the number of ticks.

Use two lines below to change the number of xticks or yticks in matplotlib in python.

start, end = ax.get_xlim()
ax.xaxis.set_ticks(np.arange(start, end, stepsize))

How Oceans Carry our Internet Traffic

There is really interesting stuff about Internet that I didn’t know before. Even with expensive and robust communications satellites today, 99% of the Internet traffic among countries and continents is carried by cables beneath the ocean floors. It seems that there will be no replacement to submarine cable communications in near future. Look here for more details.

ACM Latex: Remove copyright, reference format and doi which is footer and headers… etc

Use the following commands to remove the above info.


If your abstract is missing, you should put your abstract before \maketitle