BLOG

Application Delivery Beyond Load Balancing | Stay Away From Hardware

Swarna Podila
Posted on Mar 4, 2016 11:07:46 AM

It’s been two weeks (since I joined Avi) or four Silicon Valley weeks!  And if you imagine a proverbial “firehose”, I think it is safe to say Avi unleashes at least 10x more!  So yes, these two weeks have been like a whirlwind for me.
 
As I tread my way back into the world of networking, I realize that traditional approaches to application delivery fail fast in a software-defined world that is ruled by containers, microservices, OpenStack, SDN etc.  Just yesterday, I was looking at messages that we hear constantly from traditional ADC vendors and the differences between the ground reality and the “vendor speak” stood out starkly. 
avi-no-hardware.jpg
 
Enterprises today are adopting software-defined principles to cater to evolving WebScale IT requirements.  Today’s data centers demand WebScale IT requirements such as private cloud automation, agile application deployments on a large scale.  So it got me thinking.  “What would I do, if I were a legacy ADC vendor?”  The answer was simple -- I would resort to sugarcoating these abilities into convincing customers to buy more hardware!! 
 
Brilliant!  Right?  Wrong!
 
So here’s how these thoughts unfolded in my brain: 

Customer Requirement 1: Software-defined Application Delivery

/* If an organization is adopting software-defined application delivery and load balancing model, and an ADC vendor suggests virtual appliances, recognize that it is NOT true!  Virtual appliances are not software-defined solutions.  Remember, it is still an “appliance” because they still suffer from static capacity. You still need to manage and provision them as individual devices */
function (VendorSpeak1)
{
if (requirement = software-defined)
then
#include “MoreAppliances”
}
 
/* Avi Networks is a 100% software-based solution and does not tie enterprises to an appliance-based approach.  This enables enterprises to truly accomplish WebScale IT objectives with a software-only approach that are not restricted to the capacity of the virtual appliance.  With centralized management, Avi’s application service fabric enables enterprises can scale out automatically and elastically without adding operational complexities */
 
function (AviSpeak1)
{
if (requirement = software-defined)
then
#include “use commodity x86 servers”;
}

Customer Requirement 2: Multi-tenancy

/* Enterprises requiring multi-tenancy are driven by requirements such as compliance, application isolation, and reduction of failure domain.  With legacy solutions, it is nearly impossible to dedicate expensive load balancers (physical/virtual) on a per-tenant or a per-application basis. Upgrading large hardware load balancers requires creating an outage window across hundreds or thousands of applications */
 
function (VendorSpeak2)
{
if (requirement = multi-tenancy)
then
#include “MoreHardware”
}
 
/* Avi delivers an appliance service fabric that can spin up per-tenant or even per-app load balancers that enable true application isolation and multitenant application services.  Avi’s centralized control plane and distributed data plane enable clear separation of tenants without one application stepping on another application’s resources.  Enterprises can also shrink their failure domain as they can individually upgrade per-tenant or per-app load balancers */
 
function (AviSpeak2)
{
if (requirement = multi-tenancy)
then
#include “Per-tenant and Per-App Micro-Load Balancers”
}

Customer Requirement 3: Scale and Performance

/* This is my favorite!  Enterprises need to plan for peak utilization; so primary requirements are – (a) ability to scale and (b) performance.  Now these are not absurd requirements.  But what IS absurd is if the said enterprise is forced to spend ~$1 million a year on proprietary hardware to meet their requirements! */
 
function (VendorSpeak3)
{
if (requirement = scale && requirement = performance)
then
#include “ManuallyProvisionMoreHardware”
}
 
/* Enterprises also experience peaks and troughs in app traffic and therefore, should not overcompensate for capacity.  Predictive autoscaling, based on preset rate thresholds should trigger automatic scaling (up/down) of resources without even mandating human intervention.  The Avi Service Engines collect real-time application telemetry, which is analyzed by the Avi Controller to provide insights into each transaction and end user access patterns.  Based on this data, administrators can define rate thresholds to automatically scale up/down the resources without requiring manual intervention */
 
function (AviSpeak3)
{
if (requirement = scale && requirement = performance)
then
#include “AutoScale”
}

Customer Requirement 4: Peak Capacity

/* ‘tis the time of Black Friday (or Cyber Monday).  You are not sure if your application server can hold up to the spike in traffic volume.  Or, consider the example of a health care provider that encourages their doctors to communicate with patients virtually.  What if the application traffic brings the server down?  Would those patients be patient? 
 dontdoit.png
Unplanned outages negatively impact enterprises.  Under-capacity can leave revenue on the table.  When you are planning for zero downtime and ability to absorb peak capacity, your ADC vendor recommends you to procure and pre-provision additional hardware appliances (oh, and did I mention they only recommend the priciest high-end appliances?) – overprovisioning by 5x or even 10x is very common for mission critical applications.  That just means that you are keeping the most expensive assets in your data center underutilized by 90% for the remaining 364 days in the year!
*/
 
function (VendorSpeak4)
{
if (requirement = peak capacity)
      then
#include “MoreHardware”
}
 
/* Avi Vantage Platform enables administrators to define rate thresholds that trigger automatic scaling up of both load balancer and application resources on commodity x86 servers during peak utilization to prevent application outages.  Conversely, administrators can also automatically scale down resources as the traffic goes down.  For the remaining 364 days in the year, these commodity x86 servers are available for your application dev/test environments */
 
function (AviSpeak4)
{
if (requirement = peak capacity)
then
   #include “AutoScale”
}
 
 
// Questions?  Comments? Tweet me @skpodila
 
<div class="hs_followme" style="height:52px"></div>
<script type="text/javascript">
var __hs_fm = {portal: 443964, host: 'hubapi.com', blog: false};
(function(){
var fm = document.createElement('script');
fm.type = 'text/javascript'; fm.async = true;
fm.src = '//static.hubspot.com/js/fm.js';
var s = document.getElementsByTagName('script')[0]; s.parentNode.insertBefore(fm, s);
})();
</script>

Topics: OpenStack, ADC, SDN, Avi Networks, Application Delivery Controller, Microservices

  
New Call-to-action

Subscribe to Email Updates

Recent Posts

Posts by Topic

see all