32 NVRs Needed For ~1000 Avigilon Cameras?
1 NVR for every 33 cameras? 32 total NVRs needed for just over 1000?
Is this the right architecture?
A tweet's been going around with an Avigilon dealer showcasing such a project, see below:
It's not often you get to see 32 @Avigilon NVRs on a site.. pic.twitter.com/E2TVx5KKb4
— TecSec Distributors (@TecSec) December 23, 2013 A couple of things come to mind:
- Only ~33 cameras per recorder is quite low. That low density likely increases cost.
- With so many NVRs, it would likely be much less expensive, more efficient and easier to manage with external storage (SAN, NAS, etc.).
Specifically, the MSRP for those NVRs (with the lowest amount of storage - 3TB) is $5,160 and it gets much more expensive with more storage, excerpt:
And that, of course, does not include the cost of VMS licenses.
What approach would you take with this many cameras?
Here are some common approaches and rationales for managing large numbers of cameras:
- More powerful servers managing 100+ cameras each. Rather than using NVR appliances, purchase COTS servers with multiple CPUs, higher amounts of RAM, etc. For example, the Avigilon NVRs come with a single CPU and 6GB RAM (see specifications). More powerful COTS servers can reduce total number of boxes needed and total cost.
- Storage centralized in a SAN or NAS instead of storing inside of each NVR appliance (i.e., one central 'pool' of say 300TB vs 10TBs inside each NVR appliance). This is easier to manage and reduces inefficiencies as storage is shared amongst all servers. For example, you do not need to worry about any single NVR needing more storage or wasting storage on a particular NVR.
- VMSes running on virtual machines often allow for more cameras to be recorded/managed on a single server (i.e., multiple VMS instances, one per VM) as well as facilitating transfer of that VM to a new machine (for service, fault tolerance or expansion).
With regards to Avigilon's NVR, they are specified to 32MB/s (i.e., 256Mb/s) throughput and/or 128 cameras. At 33 cameras per NVR, that implies an average bandwith consumption of 8 Mb/s, which is quite a lot even for full frame rate, multi megapixel continuously recorded video (unless it's JPEG2000 and then....)
Avigilon's employees and partners have gone berzerk on twitter, mostly with ad hominin attacks (calling us 'such a dick', etc.) but have not offered any specific facts or technical details that would explain the decision.
1. Avigilon will not record to NAS
2. 256 Mb/sec standard advise by lots of VMS companies
3. 8 Mb/sec about right for H.264 specially at night time
I personally never put more then 24-30 cams per server ( 2MP at 30 frames )
Virtual Machines choice is up to customer not up to me
I offer they decide
there is a potential for this having been a install specified by an engineer... in that case spec may have called for avigilon nvrs vs. the cots servers you mention... sometimes you can substitute with an equal or greater product... without much for details I am just speculating...
Without knowing the application and customers concerns, objectives and requirements it impossible to say whether this is a good design for this application.
We would not by default design or engineer a system like this but we have been asked by some of our customers to provide solutions like this and believe it or not there is some very valid logic/reasoning for requesting solutions like this
Considerations
Why choose only Avigilon for servers/storage, cams and software? Customer wants complete turn key solution from one vendor "one throat to choke / no finger pointing / one call for service". You may find it weird but our largest customers are the ones that request this the most for standardization, efficiency and support.
Why not consolidate servers storage? Customer wants to minimize / limit the exposure / affected devices in any single piece of equipment failure (including servers and storage).
Why so few cameras per server? Customer has cameras set to high frame rates and bandwidth reducing the number of cameras per server or System is designed for worst case bitrates Or customer plans on doing failover or camera additions in the future and has budget now for servers... or bla bla bla.
These are some possible reasons behind a design like this - but...
Again, without knowing the application and customers concerns, objectives and requirements it impossible to say whether this is a good design for this application.
John what you know is enough to speculate - that's it.
I will say this - we have servers from Avigilon running over 100 megapixel IP cameras on a single server and we have some running over 180 analog cameras on a single server. However we have some running 30 or fewer cameras as well. We have applications where we go with servers straight from dell and create a few VMs on them and utilize centralized SAN for storage. We also have applications with customer provided storage and servers. Crap - we actually have Avigilon running on $400 pcs from Walmart with 40 analog cams and a few IP no problem. The fact here is all of these applications we utilized the appropriate servers and storage based on the specific requirements, concerns and budgets for each customer. Every one of these customers are happy as well.
I see no reason why Avigilon or the integrator should try to explain the application on this site simply because you decided to post an article. This does not give Avigilons competitors any type of advantage either. Trying to use this speculative, incomplete ponderance of an article to gain a competitive advantage would be ignorant.
I know as well as you by now that you do these articles for fun probably when you are bored or have nothing else to write about - to see what responses you will get. This holds no real value to any of your users (other than entertainment) because it is completely speculative. It's a "bait" article - good bait though trying to get a company or your users to defend a product because of your negative speculation...
As this is a solely speculative discussion and you are speculating this is a bad design I'm simply stating there could be good (relative) reason behind this design. I do not know the installer or the application. I'm just providing some insight into possibilities other than the only negative possibilities laid out by John. As "proper" design is relative AND dependent on so many factors that are not provided in this discussion we have absolutely no clue who is closer in the ppossibilities laid out.
Ultimately, if the customer is happy with the end solutuon and within budget everything else is a moot point.
Lots of Avigilon people are saying John is merely speculating when saying that 32 NVRs running 1000 cameras is not the most optimum architecture for this deployment.
Please provide ANY reasoning that makes sense for utilizing such an architecture in THIS particular install.
Thank you.
Everything with THIS is speculation because we do not have any details other than camera count and the number of servers.
It's easy to give a preferred design method based on camera count alone but literally impossible to know wether it is acceptable or preferred without the rest of the solution details / requirements and the customer preference and input.
Good integrators design systems based on each customers requirements and preferences. We have customers that have us do things we would never recommend or design but that's the way they want it - so we do it.
Without knowing the complete project requirements and the customers preference and input noone can speak to what's right for THIS application - period
Your question reveals both your inability to understand previous comments and your ignorance when it comes to system design and customer input.
There are 2 potential architectures being considered:
- Dozens of NVR appliances (OEMed from Dell)
- A Few COTS servers + network based storage
You and Alex have done a great job of making it seem that these two approaches are equally valid for a 1000 camera, single site project. This is false.
For this scenario, regardless of the unknowns, the benefits of a few COTS servers + network based storage over NVR appliances are myriad and significant - much lower product cost, less space needed, lower power consumption, greater scalability, less storage wasted, etc.
Please refute this specifically. The only potential viable argument I have heard so far is Alex's implication that Avigilon's technical flaws are what prevents this (COTS servers + network based storage) from being pursued.
So you know John, with only the information provided, we would lead with a design as you suggest utilizing more robust servers and a centralized SAN. So we agree that superficially this appears to be a better architecture for an application like this and one even we would suggest and present initially. But just because this would be our initial design does not mean that our customer would agree or accept the design. Therein lies our disagreement and what I think is one of your weakest mental attributes: how to deal with customers / end users in the real world - they matter even if they disagree and ultimately we are providing a solution for them so we have to listen to them and build our design around our knowledge and experience along with our customers specific requirements, concerns, preferences and input. Sometimes the solutions we implement are radically different than what we have designed or preferred because of our customers standards, requirements and preferences. The fact is we do not know any other details about this implementation and this customer so even though we have our design preferences that we agree on does not in any way allow us to assert with absolute authority that it would be best for this application and customer.
If this was the first reply from the Avigilon camp, I imagine this whole string would never have occured.
Happy Holidays everyone! :)
Thank you for acknowledging the fundamental point here - that for the core details we know, "robust servers and a centralized SAN" is a better approach. This is why it's a topic worth discussing and an important point for surveillance professionals to understand when considering large scale systems.
You presume that the customer 'disagreed' with what we both agree on and that the customer pushed the Avigilon dealer to go with all Avigilon NVR appliances.
This is ironic since you simultaneously keep harping how we don't know enough but evidently you know enough to confidently assert that it was the customer's choice to go with all Avigilon NVRs rather the dealer who advocated it. Do you see the catch-22 you are in?
It could be either one (the customer demanded or the dealer pushed for it) but given who benefits financially in selling more Avigilon NVRs vs COTS servers + SANs, it is far more likely that the dealer was the one who advocated this approach.
John, your pen (keyboard) is mightier than your mind. I presume absolutely nothing - you should take off your negative filter and read my posts again. I have never said I thought one thing over another as you have - I simply provided an alternate to your speculation. Again, I know it's tough for you to understand but speculation is just that and holds no true value to your users other than entertainment. Actually I know you know it and you like to stir things up - good for your site traffic but bad for actual content value.
What's truly great is Avigilon refuses to speak to you because of your tactics and you just can't stand it - it's hilarious. You think you should be able to force a manufacturer to defend themselves because you choose to speculate over a few twitter feeds and then you twist it around with your own twitter feeds... Classy John - and professional because we know security professionals make their decisions based on the speculation and rantings of a "non-biased" guy like yourself.
IPVTMZ is here. At this rate your average user will change from end users and security / other professionals to teenage girls - umm and me because I'm a sucker for this crap obviously.
Your whole counterargument above is premised on the customer disagreeing with (both of our) recommendations for "robust servers and a centralized SAN" approach. Read your comment again and you will see. All I am saying is that of those two 'speculations', mine is far more likely as it follows the financial incentives. You can feel free to disagree but let's leave it at that.
That said, you are a rabid Avigilon fanboi who only ever objects to IPVM content when it is critical about Avigilon. When we criticize other companies in the same way, like the recent Genetec and Panasonic project, nothing...
Keep on popping away, as you acknowledge, you are a sucker for defending Avigilon.
I would like to wish everyone a Merry Christmas! Hope you all are having a great day and have the opportunity to spend valuable quality time with all those that are important in your life.
John, Sean: I think it would be helpful to members if you could identify the material facts in this case.
On first reading all I see is
- A rack with what appear to be 13 double height nvrs
- A negative statement regarding how infrequently one sees 32 nvrs
Regressing thru the twitter thread I see a statement about 1000 cameras by someone different than the one making the 32 nvr statement. I can't really say I saw much 'showcasing'.
Are there other tweets/conversations that are relevant?
Thx Marty!
Perhaps John could (or has already) reach(ed) out to JP for clarification. Maybe even offer him a temp login for a roundtable discussion??
On the bright side even Ryno_Video admits "It's not often you get to see 32 Avigilon NVRs on a site.", so such low density configurations for whatever reasons may be not very common.
P.S. If the 32 nvrs are in a single rack we're talking about one of these puppies:
Now that's a rack!
FWIW only three of the NVRs seem to be on...
I have to agree with some of the undisclosed about "Real World". Sometimes providing the right solution isn't what is going to get you the job, because unfortunately someone else will give the customer what they want, even if it's wrong.
As both a Dell direct partner and an Avigilon dealer, I can tell you that there is not all that much difference in cost to me between the two servers. While I read most if these comments I did not read them all (this seems to be a hot topic), did anyone consider that maybe the dealer decided to utilize Avigilon servers for the sheer fact that Avigilon would have not have a leg to stand on if there were performance issues surrounding the servers? I can't tell you how many times i have installed servers that meet or exceed the "Recommended" specs only to have a manufacturer question them right out of the box if we were discussing performance issues. The next thing they question is the network itself, which is usually dropped once I ask them which if the 3 Cisco CCIE's in my office they want to discuss that with.
I have my first medium scale Avigilon install within the next 30 days (365 streams), I will let you know how it goes. I hope all I am hearing here about Avigilon is just a bunch if noise, but if it is not, the engagement will be short lived.
Some VMS systems can only handle video databases of a certain size. The larger the database gets, the slower playback and retrieval time gets. And then some just have a maximum limit, like any SQL or Oracle database. You can get more powerful machines, maybe, but some databases only scale to certain sizes, so you have to spread out overall size across multiple machines. (And more powerful servers and SANS sometimes cost almost as much as multiple smaller servers.) 33 cameras maybe not seem like a lot of cameras that would make a sizable database, but if the client requires a very long retention time (60 days... 90 days.... 180 days), that's a large databse for just a few cameras.
IPVMU Certified | 12/27/13 07:06pm
[Remove]
IPVMU Certified | | 12/28/13 02:54pm
Elaboration: When an email thread at work would go "south" or turn bad in back and forth conversation and one party wanted out of the replies, they would reply to all "Remove" to be left out. More than a request to opt out of the conversation, it was generally a commentary on how bad the conversation had degenerated.
I've never sold an Avigilon NVR, always use COTS and buy the software separetelly, a bit more cost effective for me and the customer with pretty much the same warranty.
Without knowing the application it's hard to judge if the hardware investment was justified. Also I'm assuming with a 1,000 camera deployment at least a couple of those 32 servers are set for standby failover.
Of course Avigilon OEM's the DELL Rack Servers for their HD RACK NVR's. 24x7 4 Business Hour Critical support worldwide, and for many companies this will mean a lot.
If we have a local client, where we are writing and supporting an remote/onsite service support agreement, and they had a absolute minimum camera count, and a absolute MAX budget, I would consider an alternative server, one that we have extensive IT experience with and would match, (actually nearly double) the 32MBytes/Sec recommended max on the Avigilon Server.
We've tested the Intel SR2xxxx servers, with LSI controllers and RAID10 configurations and have exceeded 64Mbytes continuous write speeds.
I believe I could place 100 Cameras on each of 10 servers, and have Each Camera to have a secondary/tertiary to split any one server that failed and divided it's 100 cameras with 11 going to each of the remaining 9 servers....
We have this in miniature in place......
But we sure like taking a server out of the box, and getting 5 minute setups, so doing my own server would be carefully decided, and done as a last resort to meet the needs of the client. Of course we assume the tech risks, but having been doing that in the IT world for 25+ years, it would work as expected.
I've re-read the posts, and we've all made assumptions about cameras based on our own experiences. Does everyone know that Avigilon has a 16 and 29MP JP2K camera at 2IPS high and compression that can create 50mbits and more in bandwidth. Assuming the installation is going with the PRO series cameras, I suppose 30 cameras could stress an NVR, and based on the retention requirements. The math might be perfect.
Newest Discussions
| Discussion | Posts | Latest |
|---|---|---|
|
Started by
David Little
|
6
|
2 minutes by Undisclosed Manufacturer #2 |
|
Started by
Undisclosed Integrator #1
|
22
|
less than a minute by Undisclosed Integrator #1 |
|
Started by
John Honovich
|
14
|
less than a minute by John Honovich |
|
Started by
Brian Rhodes
|
1
|
less than a minute by Brian Rhodes |
|
Started by
Undisclosed Manufacturer #1
|
8
|
26 minutes by Undisclosed Manufacturer #1 |
