Hardware?
Cable
term...
And we can assume most of us that it's VMWare based no matter who is
selling it.
So being a hardware snob and "Cloud Computing" providers rarely post
this info.
Who has the better backbone of hardware? And is anyone actually using
something unique out there that makes them a cut above the rest?
Virtual machines are great, this I can't argue; but if it's built on
hardware that fails every day, what is the actual gain. To me it seems
you just magnified your issues 10 fold since the providers will be
swapping hardware the more customers they share the platform with and
would eventually exhaust their "turn up" SLAs as they are too busy
keeping up swapping out crap servers. The same could be asked about
the their back up hardware and architecture.
My curious thought for the day...
Masanori TANAKA
/M
MARK CONLEY
> Date: Thu, 9 Oct 2008 14:14:55 -0700
> From: mta...@itochu.net
> To: cloud-c...@googlegroups.com
> Subject: Re: Hardware?
Moshref
Why DDN Storage??
> </html
Dave Graham
MARK CONLEY
From: mos...@comcast.net
David Sofocleous
Dave Graham
Moshref
Mark,
Thank you for your input, HDS, will provide you the same performance, but not only that, they offer very good scalability, with data bases, DB2. SAP, SQL,
DDN say: Safely use SATA drives for critical applications
For critical application, you use SAS, not SATA!!! SATA won’t give you performance!
Furthermore they don’t offer Thin Provisioning, .
Isn’t that CC major goal is to run data bases, and of course Apps on top of them for clients, and user.
How about Price performance that is the key for CC offering. I believe DDN will be too high!
DDN is targeting huge Data Center implementation.
Rgds,
Moshref
</html
Jan Klincewicz
wisdom still in purchasing hardware from Tier 1 vendors. As I am now a
Systems Engineer in the virtualization space I can GUARANTEE your life
will be easier getting support for a complex software environment if
your servers are well-tested and popular.
Even with a vendor lilke HP, a componenet like Virtual Connect will be
less known by typical support Engineers than a common Cisco switch.
You can save a lot of acquisition costs going with generic hardware
but unless you have the internal resources of a Google or Amazon, you
will likely require multiple vendors to interact at some point
> <http://www.byteandswitch.com/blog.asp?blog_sectionid=673&doc_id=165534&WT.s
> vl=blogger1_1> &doc_id=165534&WT.svl=blogger1_1
--
Sent from my mobile device
Cheers,
Jan
MARK CONLEY
Thank you for your input, HDS, will provide you the same performance, but not only that, they offer very good scalability, with data bases, DB2. SAP, SQL,
DDN say: Safely use SATA drives for critical applications
Would you consider nuclear simulation mission critical? What about broadcasting the Olympics to the entire country of China?
Isn’t that CC major goal is to run data bases, and of course Apps on top of them for clients, and user.
Some. Check our Nirvanix I don't think database is what they are running!
Stop trying to sell me on your current outlook on storage and read what I said about unstructured data being the bulk of what's being created.
HDS Netapp EMC.... were not designed for these environments period. Show me a single benchmark were HDS or anyone, puts up anywhere near 6GB/s write performance from a single system.
You are comparing apples and oranges.
If you need DB storage Netapp would be my choice.
How about Price performance that is the key for CC offering. I believe DDN will be too high!
Are you just guessing?
Furthermore they don’t offer Thin Provisioning, .
Not usefull in VOD post or broadcast space.
DDN is targeting huge Data Center implementation.
Nope. Just environments that need to scale beyound 100TB and need performance in large unstructured data enviromnents.
Mark,
For critical application, you use SAS, not SATA!!! SATA won’t give you performance!
Rgds,
ed.d...@icoda.co.uk
Sent from my BlackBerry® wireless device
-----Original Message-----
From: "Jan Klincewicz" <jan.kli...@gmail.com>
Date: Fri, 10 Oct 2008 10:02:40
Jan Klincewicz
There are dozens of brilliant new start-ups and small innovative shops out there, but when things need to work in grand scale (as CC implies) small incompatibilities are multiplied a thousand fold.
I hate referring my customers to a Hardware Compatibility List as often as I do, especially knowing so many great products probably WOULD interact well, but as an SE who is supposed to make things work in a one-day proof-of-concept, I would be crazy to take a chance on something esoteric, no matter how promising.
I suppose there is a great feeling of liberation among providers who have the technical wherewithal to "roll their own" and run Open Source Xen and CentOS on white-boxes, but when you have an SLA, and things go badly, it's comforting to have a 24x7 support agreement.
Cheers,
Jan
Cheers,
Jan
John Silva
WEB2.O infrastructure or SaaS/Cloud/HaaS, is 3PAR Utility Storage. - www.3PAR.com
These guys ROCK!
Their solutions does exactly what it says it on the cover, them seem
to have a cult like following with their customers proud to tell
prospects about their 3PAR Utility storage purchase.
If your looking for a new storage infrastructure then consider these
guys - they kick butt against the big guys like EMC etc..
Jan Klincewicz
It's been a while since any new hardware vendors have emerged. I was hoping for Fabric7 to succeed, but alas ..
http://search.sys-con.com/node/368244
even a well-financed startup with great talent and ideas can get squashed in this industry.
Cheers,
Jan
Cheers,
Jan
John Silva
Dave Graham
isn't it supposed to be "EMC = Extra Margin Corp"? ;)
i definitely have a bit of a different view than you on the relative strengths and weaknesses of EMC, but that's not something to talk about on this list. ;) happy to enjoin that conversation offline.
as far as where I'd position storage in the listing below, I'd definitely see storage virtualization as being the key component to a WELL developed IaaS strategy within CC. To reiterate what I've stated before, the actual physical storage is less of a concern than the policies/software/technology that manage data flow and movement. IBM's SVC, EMC's Invista/Rainfinity, etc. etc. exist BEFORE the physical storage layer (they're part of IaaS by concept; and yes, they could be considered "storage devices") as front end virtualization layers to underlying storage arrays (HDS, HP-Lefthand, 3Par, Pillar, DDN, et al.) The problem with using singular arrays with singular technologies present has to do with scalability and management; something that CC sometimes takes for granted.
cheers,
Dave Graham
978.239.2489
Dan Phillips
How about Software innovators like Asigra???
Our Agentless Back-up and Recovery Platform is second to none.
Especially in the Virtual Space.
Basically, I think there’s a lot of great dialog here. Thanks for inviting me and putting up with my shameless plug for my company’s software platform. But the truth of the matter is that IaaS and PaaS and SaaS will all converge. Economies of scale can now be achieved because the software and hardware can now support completely virtual environments. At the end of the day we have to understand that the technology that we pour our souls into is but a means to an economic end. And, the other fact is that very very little of the technology that’s been created to date provides a ROA (Return on Assets) sufficient to warrant any company owning it.
So, now that environments can be created with sufficient SLA’s to support Enterprise businesses, thus allowing the CFO and CIO’s to make the decision that through aggressive SLA management they can support the needs of their company’s employees through a fully outsourced technology environment (not meaning fully outsourced including the people necessary to run it) we’ll begin to see more and more companies moving their IT infrastructure over to an IaaS framework.
Yes, it’s my opinion and I know what opinions are like…..
Thanks again for letting me a part of this super group!
All the best,
Dan
Paul Harrington
10GE+Blades+RDMA modules like: iSER or IB+SDP:
A good example hardware platform:
The HP BL495c with dual 10GE
http://h18004.www1.hp.com/products/servers/proliant-bl/c-class/495c-g5/index.html
On Thursday, October 09, 2008, at 05:41PM, "Dave Graham" <opteron....@gmail.com> wrote:
>i just thought I'd make the following point:
>
>when it comes to hardware within CC, in my personal opinion, storage is the
>easiest thing to take care of. anyone and their mom can come up with
>storage systems (yup, even EMC can do that... :) ) but what really drives
>the innovation is HOW the data is stored and HOW it's optimized in flight
>for storage. (in other words, this is less about CORE and more about EDGE).
>
>for example, one of the great vendors I get a chance to work with is Netex (
>netex.com). They're not a core storage company; rather they focus on how to
>optimize data IN FLIGHT from source to target (WAN acceleration, in this
>case). Secondarily, when it comes to HOW the data is stored, policy based
>engines that handle the distribution and representation work to the hosts
>(or Cloud OS, if you will) abstract the entire storage hardware layer from
>the end user. Honestly, do you really CARE what type of RAID is being used
>on S3? what's more important is where that data is (can i get to it) and
>where can I retrieve it. API integrated storage engines will provide this
>layer of attachment. Code your cloud application directly to the API and
>let it do the dirty work of allocating and distributing the storage.
>
>btw, that's the principle behind storage virtualization as well; just
>without the API hooks.
>
>Dave Graham
>Flickerdown Data Systems
>1207 Main St. #2
>Holden, MA 01520
>978.239.2489
>
>
>On Thu, Oct 9, 2008 at 8:09 PM, Moshref <mos...@comcast.net> wrote:
>
>> Why DDN Storage??
>>
>>
>>
>> *From:* cloud-c...@googlegroups.com [mailto:
>> cloud-c...@googlegroups.com] *On Behalf Of *MARK CONLEY
>> *Sent:* Thursday, October 09, 2008 3:13 PM
>> *To:* cloud-c...@googlegroups.com
>> *Subject:* RE: Hardware?
Dave Graham
would love to know more. again, looking at storage within the cloud doesn't preclude the need to have redundant copies of that information. archetypically, this data is "sprayed" about the cloud infrastructure to ensure that no single copy of data exists in a vacuum.
how does your product unit unite IaaS and PaaS?
how can we best leverage your technology and IP?
cheers,
KennyO
of all this cloud/IaaS stuff is that HW just doesn't matter that much
anymore. (Full disclosure: I'm speaking as a software vendor). When
stuff fails, infrastructure management just works around it, and moves
the SW image (and data and network) somewhere else.
The conversation needs to get away from HW performance/reliability
fixation, and onto a software performance/reliability fixation.
Kenneth Campbell
I work with a company that offers "Cloud Computing" also; I'm not in
that group (I just love watching Synergy rebottled) so I won't sell
you a thing or pass on info that you don't want shared. Imagine every
CC vendor is trolling this group anyways, so that may be a mute point.
But I think hardware does become a factor for a few reason:
- Key server vendors; Dell, IBM, HP, SuperMicro and Sun
Imagine this is where I'd say most of the CC service providers get
the hardware from, believe Amazon is using internally built boxes
(anyone want to confirm or deny?) So when it starts to come crunch
time on the competitors to bring in customers I'd jack up server
prices and now CC vendors have to seek a cheaper vendor. This is the
same for any hosting platform, any of us that has been around has had
to make this transition at one point at another. This is where
software meets hardware; you can have the slickest software out there
that you have running on IBM servers but now your going to have to
move to HP because your finance people got a amazing deal on hardware
from them. Now there has to be code rewrites, development time and the
transition/painful late nights of stuff you missed dealing with
unforeseen outages with angry customers. I can't imagine that company
isn't going to pass these costs onto the customer, so it is relevant.
Likewise how much is a company willing to continue to get gouged on
server prices to avoid this?
- Hardware limitations. A POD is a POD, but everything down to NICs
have limitations and quirks to work through. What is the virtual
network interface throughput you gain from one vendors application
compared to the other? How will a hardware change effect effect your
service?
- Blades vs. Individual servers
Never been a fan of blade computing, doesn't help that vendors get a
better commission at certain companies pushing these things; but if
you are using Blades and a failure occurs on the backplane does that
company have enough available spares to cover that loss?
KennyO, agree software performance/reliability is key and is something
that should be covered. My concern on this thread was that you can
have the slickest software out there and customers can love it. But
Moore's & Murphy's laws and all out greed will eventually apply to CC.
Just curious how many vendors have tested and run their CC software on
multiple types of hardware with success or our they still focused on
just using one vendor so they can get to market quicker and make a
buck on this years buzz word? And would you be willing to bank your
infrastructure on this.
Jan Klincewicz
platform is necesarry to support SLAs even if it is transparent to the
end-user. Additionally, supportability is a huge issue for complex
software platforms. It is hardly irrlelevant to SOMEONE.
--
Masanori TANAKA
From my perspective, the most important thing of the platform for Cloud
Computing would be massive scalability. I suppose HP, IBM and some other
Blade servers should be good for general purpose in the enterprise arena.
However if we see a platform of Cloud Computing, it has to be totally tuned
as if F1(Formula One) car. Means what we need is flexibility of
configuration and know-how to do it.
Rackable server is open x86 server. It is not different with other
vendors, if we see an individual part. But the beauty is that Rackable
server is perfectly optimized for Cloud Computing data center
environment and workloads, such as power, performance, dense and price.
I do not think that extremely high-dense server with a lot of power
usage might work in the Cloud. We may need higher balance from multiple
aspects.
/M
Simon Plant
» The conversation needs to get away from HW performance/reliability
Simon Plant
Sent via BlackBerry from T-Mobile
Moshref
many great discussion about SW implementation, specifically CC Apps, and
Data base in this form,
But HW , including Storage, Networking, can't be ignored,
Rgds,
Moshref
Michael Fehse
Eric K.
There is no *best* hardware provider/platform in the cloud space. It
all depends on what the particular requirements are, and the external
limitations being imposed are.
For example, traditional blades may be appropriate in some
environments, but for hyperscale environments (thousands or millions
of servers), they tend to be too expensive, too power hungry, and have
too many touch points to manage (integrated ethernet, FC, etc,,,).
Even traditional servers are too expensive for some when looking at
the massive volumes they are purchasing.
I've always felt that the most important thing is to try to keep
absolutes out of any decision making process. Nothing you purchase
will ever be 100% of what you need - it's all about figuring out what
the best option is at the time you make it.
Eric
On Oct 15, 3:39 am, Michael Fehse <m...@mikefehse.com> wrote:
> That is a problem of "role".
> As a user of a cloud i do not want to be embarrassed with HW.
> As a developer of cloud enabled SW ... see above.
> As a provider of a cloud, well that is a totally different story. Now
> HW matters a lot.
>
> I remember all these discussions from the old GRID days. We had to be
> very strict in describing the roles...
>
> Best regards,
> On Oct 15, 2008, at 4:55 AM, Moshref wrote:
>
>
>
>
>
> > With all due respect, CC is total solution with HW, and SW . It has
> > been
> > many great discussion about SW implementation, specifically CC Apps,
> > and
> > Data base in this form,
> > But HW , including Storage, Networking, can't be ignored,
>
> > Rgds,
> > Moshref
>
> > -----Original Message-----
> > From: cloud-c...@googlegroups.com
> > [mailto:cloud-c...@googlegroups.com] On Behalf Of KennyO
> > Sent: Monday, October 13, 2008 2:53 PM
> > To: Cloud Computing
> > Subject: Re: Hardware?
>
> > I've read all of these responses, and I just don't get it. The point
> > of all this cloud/IaaS stuff is that HW just doesn't matter that much
> > anymore. (Full disclosure: I'm speaking as a software vendor). When
> > stuff fails, infrastructure management just works around it, and moves
> > the SW image (and data and network) somewhere else.
>
> > The conversation needs to get away from HW performance/reliability
>
> - Show quoted text -
Laurent Therond
infrastructure. Yet, high availability often implies redundancy, which in
turn implies some degree of waste.
I'll have to look at Rackable in more detail.
Rod Boothby
You are right about a need to focus on adding business value.
However, that doesn't necessarily mean new architectures. New architectures implies re writing your application. Cloud computing does not necessarily require you to dump every legacy application you have and rebuild. If it did, it will take a very long time for most enterprise users to adopt it.
Simon Plant
Whilst I totally agree that we aren't going to rip and replace all the investments we've made, I disagree the architectures won't change because of the Cloud. COTS/packaged software today doesn't fare that well in the cloud, its not written to take advantage of the scalable/burstable nature and requires a heck of a lot of configuration and testing to join existing kit. The Cloud benefits from dynamic configuration and scripted entry into a compute group.
Whilst we can host apps written yesterday in the cloud today, new apps we create shouldn't take the old design approaches. A good example is the Gigaspaces webinar for EC2 which recommends a spaces based architecture where components are all on one tier and scale horizontally instead of traditional 3 tier layered architectures.
Hadoop is another example of coding to frameworks which have inherent scaling.
More a today/tomorrow thing I agree for new projects, but a new science for us to learn none the less.
Do others have examples for the group to study and a precis of the benefits and capabilities?
Best,
Simon Plant
Sent via BlackBerry from T-Mobile
From: "Rod Boothby" <rod.b...@gmail.com>
Date: Wed, 15 Oct 2008 16:02:31 -0700
To: <cloud-c...@googlegroups.com>
Subject: Re: Hardware?
Jan Klincewicz
I USED to be in the blade business until I realized customers are beginning to purchase less hardware opting instead to virtualize what they have.
I agree with you that nothing will ever be 100% of what you need. I have seen many environments with thousands of servers (though never one with millions.)
I suppose with visualization that is not far off though.
I agree with SOME of your generalizations concerning blades (they are expensive, proprietary, and have complex touch points) but ANY networked, SAN connected server
will require the same components (NICS, HBAs etc..)
I do argue strenuously that blades are not "power hungry" as you suggest, certainly no more so than most standard x86 platforms. Sharing of power supplies and well-engineered enclosure
cooling and thermal monitoring provide redundant hardware resources FAR more efficiently than traditional 1U servers. Also the resiliency allowed by abstracting server "personalities" on a per-slot basis by virtualizing MAC addresses and Word-Wide Names make fail-over much more efficient.
Cheers,
Jan
--
Cheers,
Jan
Matt Rudnick
Reminder: New England Cloud Users Group Monthly Meeting
Speakers: Christien Rioux, Chief Scientist and Co-Founder of Veracode
Geir Magnusson Jr, VP of Engineering & Co-Founder of 10gen
Time: Wednesday, October 22nd, 6:30pm, Meet & Greet 6:30-7, Meeting: 7-9
Location: Papa Razzi Wellesley (private room in back), 16 Washington St, Wellesley, MA 02481
--------------------
Speaker Bios:
Geir Magnusson Jr, VP of Engineering & Co-Founder of 10gen
With a diverse background as both a technical executive and an internationally known leader in open source software, Geir comes to us from Joost, where he led software development and had responsibility for platform architecture, implementation and delivery. Prior to Joost, he held positions as Director of Middleware Architecture and Open Source Technology at Intel, and Vice President of Products and Strategy at Gluecode, an open source application server startup that was acquired by IBM. He was VP, Engineering and Chief Architect of Adeptra, an innovating communications service provider, and as CTO, guided technology FitLinxx during it's rapid growth years. Geir began his commercial career as an architectural and product lead at Bloomberg Financial Markets where he developed the company's real-time, multi-platform financial data delivery and presentation system. He is also a member of the board of advisors for WSO2, an open source SOA middleware vendor. In addition to his commercial software experience, Geir is a Director of the Apache Software Foundation, currently represents the ASF on the Executive Committee of the Java Community Process, the organization that governs the evolution of the Java platform, and helped found major open source projects, including Apache Geronimo and Apache Harmony, and has been recognized for his work in the Java ecosystem through a Google-O'Reilly Open Source Award.
---
Christien Rioux, Co-Founder and Chief Scientist at Veracode
Christien is responsible for the technical vision and design of Veracode’s advanced security technology. Working with the engineering team, his primary role is the design of new algorithms and security analysis techniques. Before founding Veracode, Mr. Rioux founded @stake, a security consultancy, as well as L0pht Heavy Industries, a renowned security think tank. Mr. Rioux was a research scientist at @stake, where he was responsible for developing new software analysis techniques and for applying cutting edge research to solve difficult security problems. He also led and managed the development for a new enterprise security product in 2000 known as the SmartRisk Analyzer (SRA), a binary analysis tool and its patented algorithms, and has been responsible for its growth and development for the past five years. At L0pht, Mr. Rioux was a senior developer. He co-authored the best-selling Windows password auditing tool @stake LC (L0phtCrack) and the AntiSniff network intrusion detection system. His other activities with L0pht included significant security research, publication work and public speaking engagements. Mr. Rioux is also responsible for numerous security advisories in many applications, operating systems and environments. He is recognized as an authority in the areas of Windows product vulnerability assessment, application optimization and program analysis. His background includes 23 years of computer programming and software engineering experience on a wide range of platforms and for numerous companies, including financial institutions, mechanical engineering firms, educational institutions and multimedia groups. He graduated from the Massachusetts Institute of Technology in 1998, with a Bachelor’s Degree in Computer Science.
questions? call matt @ 617 851 6712
jimmy
(applications. middleware, ect). It is important that the hardware be
the best fit for the solution. In cloud computing, the user
experience is removed from the hardware. The best definition I can
find is ..."Packing of computing resources (usually in the form
compute or storage grids) which deliver the end product or experience
via the internet ." As such the hardware usually does not include
many of the expensive requirements that traditional enterprise
hardware has and available is managed at a laver above the hardware.
(i.e in the grid1) So the hardware should focus on things like power
efficiency and density as an environment requires.
Regards
Jimmy