Posts Tagged ‘Elastic Computing’

Lady Ga-Ga or: How I Learned to Stop Worrying and Love the Facebook

January 30, 2010

The western world ended quite suddenly.

The news, and pictures, about Lady Ga-Ga actually being a man, were first reported by Steve Jobs as he presented Apple’s new iPlot gadget at a secret location.

127 journalists immediately tweeted the story , and it was soon re-tweeted by 13,068 followers.

The tweets were automatically converted 1675,042  LinkedIn notification which turned into automatic 300,000 WordPress Updates.

Than Google picked the news up and sent alerts to 1,020,068 Lady Ga-Ga followers and 1,002,900,3 day traders.

However, the big problem started as the new automatic “Google Alert” to “FaceBook comments” mechanism kicked in.

Since Facebook comments are automatically generting Tweeter alerts ,a vicious positive feedback cycle was created.

Tweeter->LinkedIn->WordPress->Google->Facebook->Tweeter.

Soon, 95% of the computing power of the western world was targeted at breaking the (false) news to the same people again and again.

When New York  lost its electric power, due to the high consumption by data center. Google decided to cancel Google wave and create a super algorithm to solve the problem.

They took five of their Nobel prize winners, who have been working on JavaScript optimizations, and asked them to solve the problem.

Google Geniuses quickly realized the problem is similar to solving the “ipartite graphs with no induced cycle of length > 6″  problem, but just when they were ready to solve it, the network on their Android t-Mobile crashed. The only person to hear about Amazon’s EC2 explosion  was President Obama, with his secure Blackberry.

As San Francisco,Tel Aviv, Rome and London lost all electric power the mob started rioting the food supplies. Unfortunately they starved after two days because all of the food was organic.

Luckily , China was saved, as Google decided to block them, or vice versa.

Bar Refaeli, DNA Sequencing and Cloud Computing

December 7, 2009

Much like Bar Refaeli and Leonardo DiCaprio, DNA Sequencing and cloud computing go hand in hand together.

I had a  very  interesting conversation with a friend yesterday about DNA Sequencing and cloud computing.

My friend is leading one of the largest cancer genome research projects in the world (and  yes, he is extremely  bright).

It appears that there is a great progress in DNA sequencing technology, based on chemical process. The pace is much faster than Moore’s law. As a result the budgets are shifting from the chemistry side to the computational side.

In the past, the budget would be 90% for biology and 10% for analyzing the data coming our of the DNA.

As the sequencing costs have fallen by orders of magnitude there is more and more data ( a single patient genome data is one TeraByte).

The more data , the more computing power needed to analyze it and hence the budget split becomes 50-50.

Each computation can take up to 24 hours, running on 100 cores mini grid.

In theory, such tasks are great for cloud computing IAAS (Infra Structure as a Service) platforms or even PAAS (Platform as a service) solutions with Map-Redux capabilities.This EC2 Bioinformatics post provide interesting examples.

In practice there are three main challenges

  1. Since Cancer research facilities need this server power everyday, it is cheaper for them to build the solutions internally.
  1. To make things even more challenging, the highest cost in most clouds is the bandwidth in and out of the cloud. It would cost $150 to store one patient data on Amazon S3, but $170-$100 to transfer it into S3.
  1. Even if the cost gap can  be mitigated, there can be regulatory problems with privacy of patients data.After all its one person entire DNA we speak about. Encryption would probably be too expensive, but spiting and randomizing the data can probably solve this hurdle.

So, where do clouds make most sense for this kind of biological research ?

One use case is the testing of new improved  algorithm. Then, the researchers want to run the algorithm on all the existing data, not just the new one.

They need to compare the results  of the new algorithm with the old algorithms on same data set.They also need to finish the paper on time for the submission deadline :).

In such scenarios there is a huge burst of computation,needed on static data, at a very short period of time.Moreover,  if the data can be stored on shared cloud, and used by researchers form across the world, than data transport would not be so expensive in the overall calculation.

These ideas are fascinating and hopefully would drive new solutions, cures and treatments for cancer.

Why Won’t The Big Big Giant Eat You for Lunch ?

October 11, 2009
Oh, man! We killed Mr. Burns! Mr. Burns is gonna be so mad! – Homer Simpson
Big Giant Stepping On A Small Company

Giant Stepping On A Small Company

One of the most annoying questions I had to answer in last couple of years was “Why can’t Cisco\IBM\Microsoft\VMWARE\HP easily copy what you do?”

To some extent, it is another variation of the annoying “What’s your intellectual property?”

Both of the questions are studied in first year MBA courses  .They seem to make sense at a first glance, but I would try to show they are highly overrated questions.

The underlying assumption is that BBG (Big Big Giant) can use its amazing resources, huge capital , loyal customer base and brand to kill any small company if the small company does not have a great barrier to entry, which is typically a technological one.

Having worked in few BBG’s and couple of Start-Up I beg to differ. The giants tend to fail themselves.

StartUp Beats Big Big Giant Corporate

StartUp Beats Big Big Giant Corporate

Lets start with some questions:

  • Why was Sun unable to succeed with its own firewall (SunScreen?) when it tried to stop OEM’ing Check Point’s?
  • Why was Check Point repeatedly unable  to take the SOHO firewall market  (FireWall-First, Small-Office, Safe@Home,Secure-1) ? NetScreen took if from CP. Then Fortinet did the same thing for NetScreen.
  • Why does Microsoft still lack  significant footprint in the Firewall business?
  • Why does Microsoft  ten billion dollars research budget fails to copy Google’s search algorithm for ten years?
  • How come Google Video lost to YouTube?
  • Why is VMWARE leading over Microsoft in virtualization?  Microsoft acquired Connectix in 2003. Connectix virtualization technology was almost as good as VMWARE’s at the time. Today there is a big gap in market share.
  • How come IBM , with years of building super computers does not have an elastic cloud solution?
  • How does small Riverbed perform  so well among the networking giants?
  • What was the huge intellectual property in Windows that OS2 lacked?

Here is a hint to my proposed answer to why giants fail, details to follow on part II :

  1. Time.
  2. Focus.
  3. Execution.
  4. Constraints.
  5. Culture.
  6. Investors.
  7. Golden Cage Syndrome.

An American-Israeli Sushi-Humus Grill-Bar

September 25, 2009

In my May post A Mexican, a Sicilian and a Greek Walk Into a Restaurant I discussed  how three different restaurants,in the same location, next to our start-up office, went out of business one after the other. Just after I posted it, a fourth restaurant opened in the same location with the brilliant name “Grey, an Israeli-American Bar- Diner”.

Guess what happened last week to the American-Israeli diner that served  sushi, Asian noodles, middle-eastern grilled meats, American burgers and breakfasts.

Restaurants,like start-ups, should  be very focused in their messaging and efforts. It is very hard to market excellence in five different areas.We had to decide whether  to deliver our cloud solution as a service (“Public Cloud”) or an on premise product (“Private Cloud”) . Both options were viable, and there were good arguments for each road. Despite the similarities and temptation it would have been a mistake to aim at both markets at very beginning.

Even “The Cheesecake factory” focuses on a single  item (with 40 variations :) ) from its huge menu – when it comes to branding. Personally, this is also the only item I Like to eat there. And yes, I know the picture is from another Deli …

New York style Cheesecake with strawberries

New York style Cheesecake with strawberries

The unique requirements of cloud-based enterprise applications

February 9, 2009

We just published on IT Structures web site the white paper I was working on.

Not So Virtual Cloud, But Virtually Nice

Not So Virtual Cloud, But Virtually Nice

Here is the abstract and the full paper can be found here. If you want to get to the technical part, jump to the requirements section below.

  • The unique requirements of cloud-based on-demand multi-tenant applications
  • Limitations of existing building blocks in virtualization and enterprise software technologies
  • Introducing an intelligent technology layer to provide automation of environment setup & provisioning, elasticity, resource allocation and scalability

The Challenge: Virtual Labs for Sales and Training

The days of “blind” purchasing of enterprise software and hardware solutions based on vendor promises alone are a thing of the past.

Customers have universally adopted a “try before you buy” approach, demanding not only a generic evaluation of the solution prior to purchase, but also a proof-of-concept (POC) implementation using their own data, integrated with their own applications and in their own environment. Equally, customers
want to invest the minimum effort in such POCs, whose setup is often more time- and resource-consuming than the actual evaluation process.

Vendors consequently find themselves providing POCs and pilot projects with a significant increase in cost of sales and a lengthened sales cycle: tying up hardware inventory, wasting sales engineers’ time at customer premises and inflating travel costs. The same often applies to post-sales training, where the vendor must provide staff for training and the cost is borne by either the vendor or the buyer, or both.

Thankfully, the convergence of virtualization and cloud computing is making POCs, interactive demos and postsale training easier and more accessible, at least in theory.

Since any network environment, server or application can run as a VM, and since cloud infrastructure can run such VMs (as well as real hardware) on demand as a service, it is logical that the two can be combined to deliver scalable, multi-tenant, on-demand provisioning and management of virtualized POCs, demos and training. Such a solution would deliver “virtual engagement” of customers during pre- and post-sales stages and reduce the expensive, lengthy real-world sales processes.

Unfortunately, although the base infrastructure and building-block components are available, assembling them to deliver virtual sales engagement and training is not at all straightforward. This is where IT Structures steps in.

This white paper explains the complex requirements for on-demand virtual engagement delivered as a cloud based service, and how IT Structures developed its ground-breaking orchestration technology in order provide it in a scalable, flexible model.

The Requirements

Cloud-based solutions must fulfill at least all the requirements expected from traditional data center management tools, software-as–a-service solutions and modern virtualization environments.

The core requirements are:
1. Complexity and Realism – The ability to build and run any enterprise application or appliance in a multi server
environment, with a complex networking topology that can be connected to the internet and to on premise
data centers.

2. Instant Gratification – Trying out a new environment should be fast and easy. As a result, the performance of the system must be excellent and it must not require any dedicated client installation. In an elastic production environment it is critical to have a frictionless solution because of the extremely frequent changes.

3. Multi-Tenant and Tiered – the system must support multiple software vendors working at the same time;
it must allow multiple enterprise customers to work at the same time on an identical but separate copy of the environment. The system must ensure the complete privacy and security for each user. The service must ensure that failures are confined to a specific environment and do not propagate across the system.

4. Replication – The system must be able to replicate a template of an IT environment and create hundreds of new customized running instances on the fly. This is critical for production, training and demo solutions and is at the core of the cloud concept.

5. Internet Enabled – All functionality must be available over the internet. The service must allow secure access to environments over the web on the one hand, and simulate private networks on the other hand. All instances should run concurrently and be accessible in the cloud.

6.Self Service – The service is geared towards both non-technical as well as technical users. It must abstract complex, composite IT operations into simple, web-based, single-click business operations.

7. Availability - The service must be able to recover from failures automatically, maintain exceptional uptime and provide self-healing and recovery functionality across all its components. Even when certain tasks fail, the service should optimize its resources to provide the highest service levels to the maximal number of
customers.

To read the way we achieve the implementation you can get the full paper or just send me an email.

Hardware, Software and (Virtual) Appliances Myths – Part Three

December 9, 2008

San francisco Virtual

In Part One I examined some myths about hardware and software appliances and showed appliances are mainly packaged software components.In  Part Two I described why hardware appliances became so successful in the last years and where.

In this part I’ll try to show how virtual appliances combine the best of both worlds.They combine the benefits of both software and hardware appliances with the extreme flexibility of virtualized computing.

Looking back to 2002, Check Point released SecurePlatform – an appliance on a CD, also known internally by the cool name “Black CD”. At the time, Check Point “real” hardware offering was not very successful and it relied on Nokia appliances to compete with Cisco and NetScreen appliances.

NetScreen appliances and appliances in general became more and more successful . Nokia produced excellent appliances as well, but they were typcalliy sold at a very high premium , chiefly for the brand.

SecurePlatform was invented  in order to offer the customers a cheaper option. SecurePaltform is a basically a bootable CD that one inserts into any x86 servers that formats the hard drive and installs a secure, shrunk down, Linux operating system with all of Check Point software products pre-installed.

The idea is to get most of the “real” appliance advantages (ease of install, drivers, secure OS, fast boot time,optimized performance) with the advantages of sofwatre ( flexibility, modularity, familiar shell and interfaces) at a very cheap hardware price (customer can choose his box and use x86 agreements and discounts).It also allows the customer to grow capcity easily without complex upgrades.

Overtime SecurePlatform became very successful and turned in to the customers’ favorite deployment choice. While in 2003 it still lacked a lot of appliance features ( image management, backup and recovery, web based interface), those were added along the years.

It is important to note that SecurePlatform based appliances, like other CD appliances,  still had some gaps from other appliances.

1. The form factor is still of a standard PC. With 1U servers becoming the norm it was less of an issue, but the number of network interfaces was still a problem in some cases.

2. Keeping up with driver computability with all the x86 vendors was very hard. When Dell\HP\Lenovo release a new firmware\driver they don’t bother to update anyone and back porting Linux based device drivers is not fun at all. The implications are that the appliance is not as generic as would seem.

3. There is no single point of support for hardware+software.

4. There is no “real” hardware acceleration, if it is really needed.

To overcome some of these, in 2005, Check Point started selling hardware appliances, based on SecurePlatform as another alternative.

Virtual Appliances are the next generation in the same concept.

Because the hypervisor presents a standard “hardware” API to the operating system, most of the compatibility issues are solved by the hypervisor manufacturers. Because the appliance is packed as a standard virtual machines, there is no need for the reboot\format\install procedure.

Ducati Motorcycle

Ducati Motorcycle

Of course, since the appliane is a virtual machine the customer enjoys great flexibility, not found in regular appliances or even “CD Appliances”

  • High Availability and load balancing across physical server (e.g Vmotion)
  • Full control over memory and CPU allocation in real time
  • Easy provisioning , tracking and backup which are appliance independent
  • Consolidating many appliances to one physical server while maintaining modular design and software independence
  • The appliance can be used “inside” hypervisors, so there is no need to move traffic from the bus to the network
  • Form factor and port density are less of an issue , since the switches and routers are virtual as well

To make the creation of virtual appliances easier, companies like Rpath, are providing an easy to use software to handle a lot of the work Check Point, NetScreen and other vendors and to redo to create their own appliances.

Some problems still remain open, mainly the lack of standard central management to control appliances from different vendors. I’m guessing one start-up or another is working on the problem.Hardware acceleration is lacking, but it would be probably be solved by future developments in the core virtualization companies.And no one needs hardware acceleration anyway :)

To summarize, it seems that virtual appliances turn software into the king again.They combine software advantages and overcome its shortcomings.

In a cloud based world, there is a good chance it will become the favorite deployment vehicle.

Virtual Clouds – How Gartner’s Top 10 Strategic Technologies for 2009 Consolidate

October 20, 2008

Gartner just published in their blog the  Top 10 Strategic Technologies for 2009 .

The list is  Virtualization, Business Intelligence, Cloud Computing, Green IT, Unified Communications, Social Software and Social Networking, Web Oriented Architecture,  Enterprise Mashups,  Specialized Systems, Servers – Beyond Blade.

The interesting point , In my opinion, is that many of these technologies are actually supporting each other, making the trend even stronger. I’ll describe why this is so and than use my company to give a subjective example.

I believe virtualization, cloud computing, Web Oriented Architecture and enterprise mashups have a  a great synergy.

Virtualization (#1) key strength is in abstraction. It removes the coupling of hardware and software.

Cloud computing (#3) takes the abstraction to the next level. Now, no hardware is needed at all.

The problem with most clouds is that they do not allow reuse of existing enterprise applications.However, Virtual Clouds can run any application from the data center , but do it on on the internet, on demand. Basically, if you have a cloud of VMWARE or Hyper-V servers you could move application between cloud and Enterprise data center on demand.

To make it more interesting, the simple fact that clouds are on the net (#7) makes them the Ideal to create enterprise mashup (#8).

With the right security and networking in place it is possible to to create hybrid enterprise applications which have one leg in the cloud and one leg in the virtual cloud.

In IT Structures we have built a virtual cloud to support the business application of virtual sales. Our service offers collaboration environment ( #6) for sales engineers and ISV  to run proof of concepts for enterprise applications in the cloud. We are using virtual private networking (VPN) technology to connect clouds and private data centers.

Their are Clouds in the Horizon - Good Ones

There are Clouds in the Horizon - Good Ones

The cool thing is that because of virtualization it is much easier to replicate, provision and allocate resource in a multi-tenant environment while keeping the environments separated. Building a service , rather than a product uses economies of scale to  reuse resources during dead hours.

The cloud location over the web means that Proof Of Concepts can be accessed by vendors, IT, executives and contractors as opposed to the traditional closed garden approach. The on demand nature lets the POC start in five minutes, which is a win-win for both the vendors and the enterprises.

Creating a virtual cloud is not trivial, the security, storgae, performance,networking and elasticity are really really hard to obtain.But once it is done, it can offer many  revolutionary new services. To wrap up,  Gartner is right on target this time. The only thing they got wrong is that they published Just three technologies this year :)

Cloud Computing To Save the Economy ?

October 17, 2008

Douglas Gourlay from Cisco makes a good point on cloud computing and economical depression .

Similar in some aspects to my previous  post Can you make money writing algorithms ? Part II.

The basic concept is quite accurate, cloud computing is best suited where capital is sparse and change is frequent. He asks what will it make to turn current hosting to cloud computing one. My guess would it takes a very strong software to turn them in that direction. The type of software a service provider would not be able to write, but really needs a very strong start-up or an ISV to develop. It is much more complicated than just integration of open source and commercial software ( trust me, I do support in my spare time : ).

I’m actually busy these days writing a detailed technical white paper on what it takes to create a virtual machines cloud to run real enterprise applications. I’ll update when its out and ready. Hints : Scale, Frequency of Change, Networking, And highly flexible optimization system that bridges business, technical and product gaps and can be changed instantly.Stay tuned.

Cloud Computing, Wine Walk, Pride Parade and Micro Macro Economics

July 4, 2008
  • Here are some really unrelated insights from my two weeks business travel in the states.
  • 1. The dollar is so weak that a haircut in New York, 45th street, 5th Avenue Is just $11. In Tel Aviv’s old north it is $18. In Nes Ziona it is $12.
  • 2. The real cost of hair cut in New York is $13. One dollar extra because the Israeli barber assumed I was a stupid German tourist .He said to his friend, in Hebrew – “You can take $12 from this guy”. Another dollar was for a tip strongly demanded, although the whole thing tookless than five minutes…
  • 3. For $15 you can eat the best ceviche in the world in Fresca, new Peruvian restaurant in the trendy Fillmore street , San Francisco. That does not even buy one a breakfast at Nir Zuk’s Noa bistro in Jaffa.
  • 4. Cloud computing is the new buzz. A year and half ago it was an early rumor. Now I was in three separate conferences in one week and all tried to focus around it – Structures08, velocity08, Gartner Israel.
  • 5. Internet Explorer 8 is going to have 6 concurrent HTTP requests for each host. Hurray. This simple change can speed up the entire Internet by 20%. In 2010 they might even support HTTP pipelining. And we thought computes would have Artificial intelligence by 2010.
  • 6. Everyone knows TCP is broken for the modern web, everyone knows how to fix it. No one is working on HTTP over UDP and CDN’s are still making a fortune.
  • 7. The San Mateo Wine Walk is a really nice event. Where else would you get 10 glasses of wine for $30 ? and where else can you see college kids performing “Born to Be Wild” and 60 year old White, Mexican and Afro American ladies in shorts shouting at them and dancing ?
  • 8. Steve Souders, Yahoo’s former chief Performance Officer now works for Google. Quite Symbolic. And he is a very nice person as well.
  • 9. 50% of the overall cost of running a data center is actually the cost of building it. Turns out that the physical and power infrastructure is very expensive.
  • 10. There were 10 firemen marching in San Francisco 2008 pride parade. There were about 40 chefs. 70% of Linux kernel contributors are working for commercial firms.

Privacy, Security and Elastic Computing

February 1, 2008

There is an interesting contest going on in SmugMug image sharing site,  you can get 600$ if you can find a security hole in their system.

This is the result of an an interesting debate if security and privacy are separated and how privacy and probability are related.

The core of the issue is that images that are marked “private” are actually public URL’s which can be easily enumerated. While SmugMug offers stronger mechanism for access control, I do believe this one creates a false sense of security.

SmuMug is a great site and it seems the people who make it are really innovative and smart. However, in the end, the question is how much would it cost to break it, assuming there is one evil person who wants to abuse it.

The surprising answer is 2535$.

 I’ll demonstrate by assuming there is one evil person in the world who hates SmugMug for being so cool and successful.

This person decides to spend his hard earned money to create a publicity nightmare.
Lets assume there are 1 Million real picture out of the 250 Million possible URL’s (the actucal number does not really matter).

He spends 500$ (100*0.01$/HR*5000HR) to get 100 servers from Amazon EC2 and use them for 2.08 days. Each server can send 50,000 HTTP requests per hour.
After 2 days the evil person knows exactly the links to the one million “private” pictures ( 50*50,000*100 = 250,000,000 ).

He needs to pay 10$ for bandwidth for the pictures ( 1M * 0.1MB * 0.0001$/MB).
The non existing links would cost 25$ ( 250,000,000 *0.0001$/MB *0.001
MB).

Total cost is 535$ to get all the pictures.
BTW, since SmugMug is using amazon’s S3, bandwidth cost would probably be 0$ since bandwidth between S3 and EC2 is free )

In order to find the interesting ones he uses Amazon Mechanical Turk. He pays 0.01$ for 5 images classification (a HIT) so the total cost would be 2000$ (1M * 0.01$/ 5).

Now the evil hacker can post top 1000 photos in Flicker and get his evil wish fulfilled (2535$ cost)

To make matters worse, a cheap evil person can accomplish the same task with a zero cost, using JavaScript & open web sites. This is very early in the morning, so I might have missed some of the calculations, but the order of magnitude seems fine.

So, I suggest SmugMug keep doing the great work they are doing, but also invest the time and effort to fix this issue.

The fact no one has complained so far, is merely because the attack didn’t take place so far. Security through obscurity does not work in the long run.

It is a shame that one evil person can cause so much work and harm to so many good people, but that’s life.


Follow

Get every new post delivered to your Inbox.

Join 137 other followers