Dragnet in Network Circle

Thursday, January 28.

It was cool in Silicon Valley, there was a slight drizzle and a chill in the air.
We were working the day watch out of the city. The boss is Captain Glenn. My partner’s Ram. He’s a good player. I am Detective Sergeant Joe Friday.
We got a call about an ongoing acquisition at the Network Circle. We checked it out.
The acquiring CEO was still talking. One of the engineers being acquired spoke with us.
“It is terrible, officer.”
“Just the facts, ma’am.”
“We have good workstations, half of Wall Street runs on our systems; the Java franchise is not skimpy either …” she said
“ Shh ..” said Ram, “Let us listen to what Larry has to say for his acquisition …”
“… has a fabulous installed base, a fabulous group of customers, a wonderful pipeline of technology”, said Larry, “ … and we just have to do a better job of taking this wonderful engineering output and delivering it to customers…” He continued ” … we are very good at saying no, … and will halt … practice of filling special orders for unprofitable systems with unusual features. We don’t want to sell products that we lose money on.”
Jonathan was charged with mismanaging an icon. He was last seen tweeting a #Haiku from his cell – “Financial crisis/Stalled too many customers/CEO no more …”

Larry was found innocent. He rushed out of the courtroom to continue his fight against the cloud windmills. The judge said he wished he did acquisitions as good.

As closing remarks, the Judge was troubled by a new development he had heard and asked if there are any good definitions of this new trend called “Cloud Computing”. We said we will pound the streets hard, shake the informants and will get back as soon as we find any signs … On second thought, I should have pointed him to my blog.

Cloud 2.0/Cloud TNG/Cloud Reloaded – Top 5

As we muddle through the various Cloud models and technologies, we also should keep an eye on what the next generation would be. That frame and context gives us a direction for our current workings. Here my top 5 on Cloud 2.0. What says thee ?

1. Hypervisor-native CPUs : The hypervisor will move into the CPU microcode – at the appropriate slot. The main reason for this are the new powerful chips – how would we ever manage a system with four octo core chips ?
   1. The hyperthreading was an attempt at this, but it was at the wrong layer.
2. Network-native Cloud Protocols : For now, the Cloud OS layer and the network are not leveraging each other at all. I firmly believe that current cloud protocols like the gossip, adjacency-awareness and others would eventually embed into the network. The protocols will be useful not only for the Cloud OS layer but also for the “Big data” architectures (see below)
3. Network services will morph as appliances : Many of the network services like security, acceleration,file services and so forth will become software appliances – with the ability to scale, distribute and elastic, as needed by the cloud infrastructure. With the advent of powerful processors, almost all network services will move either into the virtualization layer; except the services that need the highest wire-speed performance, which will move into the TOR (Top Of Rack) and EOR (End Of rack) switches.
   • [Update 1] An interesting blog by Allan in GigaOm “Where are the Network Appliances?”. He ends the blog with a poignant statement ” … networking architectures will change to a degree we’ve not seen in well over a decade.” … true indeed …
4. Software primitives will move into PaaS : Currently PaaS is an all or nothing proposition. This will change. An orthogonally extensible, non-proprietary PaaS will evolve with embedded primitives like actors, STM, functional programming et al. Look at languages like Scala, Erlang and Ruby as well as frameworks like Akka.
5. NOSQL based DaaS will evolve : Currently the BigTable, Big Data, NOSQL technologies are used only by a few esoteric applications. This will change. the NOSQL (NotOnly SQL) technologies will form a substrate for the Cloud architectures as data stores – temporary, permanent, cache and in-memory.

I will add more … What are your thoughts on Cloud 2.0 ?

A Simple Minded Cloud Reference Architecture – Part II

After my original blog on this topic, I have a newer diagram (not necessarily better or even advanced ;o)) which depicts the cloud layers. Thoughts ?

Cloud Stack

Sherlock Holmes – A quick review

Saw the movie this weekend. I am a Jeremy Brett* fan and (still) think that he is THE personification of Sherlock Holmes.
But time has changed, and am fine with Robert Downey Jr., and Jude Law. And I do think that Sir Arthur Ignatius Conan Doyle would be proud the way the new movie turned out ….

• Enjoyed the movie – Well made & an intelligent film
  • Of course, follows the traditional American movie making style (explains everything) as opposed to the British ones (more cerebral)
• Not your fathers “arm-chair” detective !
• Lots of action – I think Guy Ritchie has chosen a balance between smartness and action – enough to begin a new franchise !
  • Of course, the story is all new
  • May be he should redo all the SH books!
  • And I shudder to think what Guy will do with Poirot !
• Didn’t hear the words “Elementary, Dear Watson” !
• In fact, the major change is Dr.Watson, a totally different personality – for starters, old Dr.Watson is a gentleman – he would never sock Holmes !
• I think Robert Downey Jr. captured Holmes’ eccentricity very well – not the edgy and melancholic portrayal by Jeremy Brett
  • He has the crispiness that we saw in Jeremy Brett as well
• He is very action-oriented – may be he will develop more of a pensive personality in the future films
  • Granted – even Conan Doyle pictures Sherlock as athletic, not a couch potatoe ;o)
• The sound track is excellent – should be – it is by Hans Zimmer !
  • I have the Dark Knight Audio CD
  • And downloaded the Sherlock Holmes CD yesterday
  • Even while watching the movie, I occasionally noticed the sound tracks – it stands out in many places
  • Will buy the DVD when it comes out, just for the music …
• Am going to take the little Kaushik sometime soon (He has read all the Sherlock Holmes books – should be interesting to hear his comparison )

* Trivia : Jeremy Brett was born Peter Jeremy William Huggins. Becuase of his career choice, his father forbade him to use the family name. So Jeremy chose the name of the manufacturer of the suit he was wearing – Brett & Co !

Ref: http://british-tv.suite101.com/article.cfm/jeremy_brett_as_sherlock_holmes

Data Cloud vs. Data in the Cloud

Due to various reasons, I did some quick analysis on the cloud data systems and technologies. Some thoughts :

First a picture of the Data Cloud stack. It has become a little complex, I hope the clip arts of Style 1450 add some diversions ;o) And the background picture is the Exit Glacier (Alaska)

Data Cloud Stack & Technologies

Am still working on this blog …

Reference Links:
[1] http://nosql-database.org/
[2] http://internetmindmap.com/database_software
[3] SNIA

Virtualization and Cloud Computing – Road Ahead – Trends ‘10

I was thinking of writing a cloud computing prediction list, but realized that it is better to talk about trends, which are more relevant in the ever changing world of cloud computing. Also it is easier to write as :

• Trends are vectors than a single snapshot point in time, thus giving one a thick gray region
• And the fact that trends need not be linear gives one the latitude to say things that are closer to reality …

First we need a trends model and then a mapping of the trajectory and locus cloud domain onto that model – both shown below (of course, you can click on the pictures to get the larger view):

Technology Trends Model

The Model:

The technologies can be viewed in the traditional X-Y axes, Technology Adoption vs. Maturity.

• Quadrant I is where the maturity and adoption are high, resulting in stable technologies.
• In Quadrant II, we have mature technologies, solving effectively niche problems but not yet mainstream. The extreme is the top left corner, where we have complex technologies solving extremely special problems very well.
• Then we have Quadrant IV, where the domain drives the technologies. Many times the requirement is very fast that standards are preempted by the users. And the extreme case is the bottom right corner, when there are complex domains where technologies can never mature.

Usually technologies take three different routes:

1. A linear trajectory, when the technology matures in sync with the adoption
2. A trajectory where the technology matures and then veers into the mainstream
3. A trajectory where the users force standardization and maturity in a technology

Cloud/Virtualization Mapping into the trends Model:

Let us map some of our favorite cloud technologies onto this model:

• Virtualization as a substrate for business architectures:
  • Virtualization as a technology is in sync with the adoption – at least at the present state. Which means the advances are absorbed by the adoption. In the very near future, we will see virtualization based solutions for traditional systems problems, like DR and HA. I remember designing elaborate systems for HA (with multi-cast active-active processes) which are all irrelevant now.
  • I did study the VMware HA and FT (as part of VCP) and they are right steps. Still not there yet, but just usable now. This is where the user adoption and influence can help to mature the product.
• Virtualization-native security
  • There is a big gap in the security mechanisms at the virtualization layer. Vmware has technologies like security zones and vmsafe. But we need prevalent programming models as well as tools to effectively implement security – at different levels.
  • Actually this is the case with virtualization-native network – switching and routing – as well
• Hybrid Cloud
  • While technologies to implement public clouds are maturing; and strong public clouds are emerging, there are no real solutions for the hybrid clouds. Again, a trajectory where the users will determine the maturity and force standardization of the technologies.
• Cloud Storage
  • In my humble opinion, cloud storage is maturing more than adoption. As Mark Carlson points out in the SNIA paper, network based storage protocols (iSCSI, NFS, FCoE, and so forth) are standards and are implemented with interoperability. The big table type of storage is just developing (more below) and the Cloud based storage is working through it’s paces
• No SQL key-Value Stores
  • Actually I am lumping key value stores as well as non-structured stores (like bigTable and HFS) in this category. Again, quoting Mark Carlson, these are table space storage focusing mainly on scalability (rather than features). They are solving niche problems very well, but not yet mainstream. They will follow the path of maturing technology driving the adoption.
• IWM (Intelligent Workload Management) based on predictive analytics
  • I am sure Amazon, Microsoft, Google, Rackspace, Savvis and others have implemented some form of IWM; but the algorithmics and programming models are not yet available as mainstream packages. I believe that this is a domain that will be driven by the users.
  • As virtualization becomes norm, the cloud infrastructure control and management planes will get complex and that is when we need the quants to help us out!
  • I am working on a few pragmatic ideas on this, which I will share in another blog, in the near future.
• PaaS
  • No prediction (whether trends or not) will be complete without PaaS; so I left the best for the last !
  • And thanks to insights from Gary Berger and James Urquhart, some of my thinking in this area were clarified …
  • So far, PaaS is a “my way or highway” i.e. an all encompassing full programming environment; which is fine but not the chosen path for many including the best of the breed mode.
  • This is the year PaaS is going to lose it’s shine and meet the contenders. I think time has come for virtualization based composable abstractions as part of a PaaS platform.
  • An orthogonally extensible PaaS platform that can be part of (and integrate well with) normal private clouds (and public clouds!) will emerge
• Cloud Builders
  • I predict that this year the role of cloud builders and scalability architects would be more clear and relevant. We understand application architects and developers very well, but in the era of clouds there are also new distinct roles that need to fulfilled.

So … what thinks thee ? What generates intensity and motivation for you in the cloud business ? (I assume you are in the cloud business, otherwise you will not reach here ;o) …

Anatomy of a Cloud Concept Lab Stack

For the last few months we have been building a cloud concept lab based on Cisco’s UCS system (for compute), Nexus 5K/7K (for network) and NetApp/EMC(Storage). And during the Christmas season, we have a chance to show off our work. I will write more about the interesting projects we are working on; but for now, some pictures …

First our Compute POD – 6 X UCS Chassis with 8 blades each, connected to 6120s and then to Nexus 5Ks. As you can see we take our equipment seriously ! And Santa is here !

UCS Compute POD

UCS Closeup - Santa Claus is coming to the Cloud town !

A lab is known by the tools it keeps — A glimpse at our tools cache ;o)

A Lab is known by the tools it keeps !

Our network stack – Nexus, Catalyst, ASA and ACE

Network Stack - Nexus 7K, Cat 6500 & ASA

The storage stack – NetApp for now, and we are off to a VCE stack as well

And of course that is me & Rohit’s Flip camera

Just me !

Now onto the gory details – which I will add as a set of future posts – but to get started:

Cloud Concept Lab Topology

What is more important is not the stack one has, but what one does with it. And we have a set of concept projects …:

• One interesting project is the Cloud provider API implementation (vCloud/OCCI with our own Cloud OS – in JRuby) and then showcasing Enterprise Extension into a service provider cloud (with extended network and compute semantics) ! A few of our pragmas :
  • Store context as an in-memory cmdb (with a disk based persistence backup)
    • memcached coupled with Cassandra
  • A very lightweight compute stack in terms of management and control layers
    • Control by Cloud OS talking directly to UCSM (UCS manager) for compute and network management layer (for network)
    • Stateless cloud instances with intelligence and inferences in the Cloud OS
  • …

More details in the next posts …

Till then Happy Holidays ….

P.S: The pictures were taken mostly by Rohit and Ram using Ram’s Canon Rebel XTI / Tamrom 17 – 50 /2.8 lens

VCP4 Study Plan

As I had mentioned in my last blog, I got thru VCP-410. My secret – good teachers for my VSphere 4.0 Fasttrack class. Jim McCullough and Brian Watrous did a good job.

I also followed a study plan – after the class. Some thoughts below :

• Good Practical experience
  • Have a running system, good experience in solving various problems in a running virtualization infrastructure is essential. No doubt, the vmware course (FastTrack, for me) also rounds of the actual working experience – as one will not get a chance to work through all the facets of the product.
• Read a good book – to get a slightly different view and narration technique
  • The VMware documents, while are very good compared to other tech documents, are not exactly a long read without getting monotonous
  • I read thru the book “Mastering VMware vSphere 4.0” by Scott. Rounds off the knowledge; also gets one’s attention as the book says the things in a different way, different order
• Read VMware Tech Notes
  • I found the Virtual Networking very useful
• Take on-line sample tests
  • I like to take the sample tests early and fail ! This way, one gets a feel for how the material will manifest in the test, also as one reads the documentation, the questions one failed will stand out !
  • My first choices are Simon Long and the VMware site
    • When you are at Simon’s site, don’t do the Maximums yet ! It totally confused me until I wrote some notes
• Now the time has come for the VMware documentation work !
  • First download the book bundle and unzip them into a directory (I actually printed them and bound them at Kinkos)
    • The URL will change. I used the URL  http://www.vmware.com/support/pubs/vs_pages/vsp_pubs_esxi40_u1_e_vc40_u1.html
    • The URL has the link “Download pdf versions of all documents”
  • Also keep the pdf copy of the Exam Blue print (download from the site http://mylearn.vmware.com/portals/certification/) and make sure you have covered all the sections, as you read the various documentation
  • The order I followed:
    • ESX & vCenter Setup Guide
    • vSphere Basic System Administration
    • iSCSI SAN Configuration Guide
    • Fibre Channel SAN Configuration Guide
    • Resource Management Guide (Pay special attention to this guide. It has lots of good information)
    • Upgrade Guide (important, may not so obvious stuff here)
    • Availability Guide
• Next comes the VMware materials you got during the VMware course. Read them through to refresh all the good work you have done during the course
• Now go back to the sample exams and work them until you have > 85%
• (Dday – 3) Finally curl up with the study notes from Simon and the VMware Maximums.
  • Also my notes, if that helps. Or write your own notes on areas you are not that comfortable with.
  • This is also the time to do the “vSphere 4 Configuration Maximum Exams”. First do the exams referring to the notes.
  • And don’t overdo the Maximums. “They will come you …” when you need them
• (DDay – 1) Get a good night sleep
• (DDay) Refresh the VMware Maximums list, and the final notes
• You will ace the test in no time …
• Test taking tips
  • There are 85 questions for 90 minutes.
  • They have review check box which makes it easier to mark questions that you have some doubt on
  • You can also review all questions after you have gone thru all the 85 questions
  • My suggestion is breadth first – answer all questions you are sure of and mark the ones you have some doubt on. DOn’t dwell on any question too long
  • After the firat pass, depending on the time available (there is a countdown clock) allocate time for the ones you have marked for review
  • Usually your first gut feel is the right answer; but sometimes as you analyze the question, newer insights will change the answer
• Good luck …

VCP4 Notes

Finally passed the VCP (VMware Certified Professional) today – clocked at 413/500 ! The toughest, for me, was to remember the myriad of vSphere maximum numbers. Some notes I had written down to remember a few things easily:

VCenter Ports

80, 389 (LDAP), 443, 636(Linked Mode), 8080, 8443
902/903 – For vClient to show VM Console

vSphere Network Switch Maximums

A quick diagram made it easier … First thing I did, after settling down for the exam, was to draw this on the erasable board.

ESX Partition sizes

I couldn’t remember them, so made a list in an order that was easy for me to remember !

Partition	Size
/	5 GB
Swap	600 MB/1600 MB
/boot	1.25 GB (1100MB for /boot and rest for vmkcore)
Esxconsole..vmdk	1200 MB
Optional
/home	512
/tmp	1024
/usr
/var/log	2000

Maximums

Same story here. Too many mixed numbers all over the map. So I created a sorted list and the (kind of) ER diagram mentioned earlier !

Number	Attributes that have this number as value
4
	iSCSI initiators/Host
	Failovers/Cluster
8
	HBAs/Host
	iSCSI paths to a LUN
16
	HBA Ports
	DistSw/vCenter
20
	VCPUs/Core
32
	Extents/Volume
	Hosts/Cluster
	FC Paths to a LUN
	e1000 NIC ports/Host
40
	VMs/HA Closter (> 8 Hosts)
50%
	Failover as % of Cluster
61
	Static Targets/iSCSI Adapter port
64
	Dynamic Targets/iSCSI Adapter port
	Logical Processors/Host
	Hosts/DistSw
100
	VMs/HA Cluster(<= 8 hosts)
	Hosts/DC
248
	StdSw/Host
256
	VMs / Volume
	Volumes/Host
	LUNs/Host
	Targets/HBA
	VMs/Host DRS Cluster
320
	VMs/Host
512
	VCPUs/Host
	PortGroups /vSS
	dPortGroups/vCenter
	Resource Pools/Cluster
1024
	Paths to a host
	Children / Resource Pool
1280
	VMs/DRS Cluster
4088
	Ports/StdSs
4096
	StdSwitch/vDSwitch ports per Host
	Resource Pools/Host
6000
	vDSS ports/vCenter

NAS with iSCSI for vSphere

Needed to buy a shared storage for a small vSphere 4.0 sandbox – here are my observations.

• Was looking for block storage device – so that I can use vmfs. Barring that, I was Ok with NFS
• We already have netApp et al in our concept lab. So wanted a low budget solution that will hold 2 to 4 ST31000340NS (Seagate 1TB, SATA Barracuda) disks
• iSCSI looked like the best option for block storage
• After some digging around, the choice was between NetGear ReadyNas, Synology 209 and the QNAP 210
• Lower end NetGear boxes don’t support iSCSI
• Both Synology and QNAP are based on Marvell 6281 chip and with 256MB memory (BTW, check out the plug computer which also is based on the 6281 ! Very interesting … A topic for another day …)
  • Synology has the 209j and the 209+II. The 209j is way under powered and the 209+II has better CPU and more memory, but prices a little high
• Both SY209 and the QNAP210 have iSCSI target interface but looks like the QNAP is more advanced. (Quite possible that both are based on the Linux iSCSI substrate !
• So finally I ordered 2 of the QNAP 210s
• BTW, an interesting contender is the Windows Storage Server with iSCSI ! Adding the iSCSI control plane to Win2K8R2 is a brilliant move by Microsoft !
• I will blog the agonies and ecstasies after setting these up in my vSphere 4.0 sandbox …