Background
Echo Nest have a rate limited API. A given application (identified in requests using an API key) can make up to 120 REST calls a minute. The service response includes an estimate of the total number of calls made in the last minute; repeated abuse of the API (exceeding the limit) may cause the API key to be revoked.
When used from a single machine (a web server providing a service to clients) it is easy to control access - the server has full knowledge of the history of requests and can regulate itself correctly.
But I am working on a program where distributed, independent clients make requests in parallel.
In such a case it is much less clear what an optimal solution would be. And in general the problem appears to be undecidable - if over 120 clients, all with no previous history, make an initial request at the same time, then the rate will be exceeded.
But since this is a personal project, and client use is expected to be sporadic (bursty), and my projects have never been hugely successful, that is not expected to be a huge problem. A more likely problem is that there are times when a smaller number of clients want to make many requests as quickly as possible (for example, a client may need, exceptionally, to make several thousand requests when starting for the first time - it is possible two clients would start at around the same time, so they must cooperate to share the available bandwidth).
Given all the above, what are suitable algorithms for the clients so that they rate-limit appropriately? Note that limited cooperation is possible because the API returns the total number of requests in the last minute for all clients.
Current Solution
My current solution (when the question was written - a better approach is given as an answer) is quite simple. Each client has a record of the time the last call was made and the number of calls made in the last minute, as reported by the API, on that call.
If the number of calls is less than 60 (half the limit) the client does not throttle. This allows for fast bursts of small numbers of requests.
Otherwise (ie when there are more previous requests) the client calculates the limiting rate it would need to work at (ie period = 60 / (120 - number of previous requests)) and then waits until the gap between the previous call and the current time exceeds that period (in seconds; 60 seconds in a minute; 120 max requests per minute). This effectively throttles the rate so that, if it were acting alone, it would not exceed the limit.
But the above has problems. If you think it through carefully you'll see that for large numbers of requests a single client oscillates and does not reach maximum throughput (this is partly because of the "initial burst" which will suddenly "fall outside the window" and partly because the algorithm does not make full use of its history). And multiple clients will cooperate to an extent, but I doubt that it is optimal.
Better Solutions
I can imagine a better solution that uses the full local history of the client and models other clients with, say, a Hidden Markov Model. So each client would use the API report to model the other (unknown) clients and adjust its rate accordingly.
I can also imagine an algorithm for a single client that progressively transitions from unlimited behaviour for small bursts to optimal, limited behaviour for many requests without introducing oscillations.
Do such approaches exist? Can anyone provide an implementation or reference? Can anyone think of better heuristics?
I imagine this is a known problem somewhere. In what field? Queuing theory?
I also guess (see comments earlier) that there is no optimal solution and that there may be some lore / tradition / accepted heuristic that works well in practice. I would love to know what... At the moment I am struggling to identify a similar problem in known network protocols (I imagine Perlman would have some beautiful solution if so).
I am also interested (to a lesser degree, for future reference if the program becomes popular) in a solution that requires a central server to aid collaboration.
Disclaimer
This question is not intended to be criticism of Echo Nest at all; their service and conditions of use are great. But the more I think about how best to use this, the more complex/interesting it becomes...
Also, each client has a local cache used to avoid repeating calls.
Updates
Possibly relevant paper.
The above worked, but was very noisy, and the code was a mess. I am now using a simpler approach:
Make a call
From the response, note the limit and count
Calculate
barrier = now() + 60 / max(1, (limit - count))**greedy
On the next call, wait until barrier
The idea is quite simple: that you should wait some length of time proportional to how few requests are left in that minute. For example, if count is 39 and limit is 40 then you wait an entire minute. But if count is zero then you can make a request soon. The greedy parameter is a trade-off - when greater than 1 the "first" calls are made more quickly, but you are more likely hit the limit and end up waiting for 60s.
The performance of this is similar to the approach above, and it's much more robust. It is particularly good when clients are "bursty" as the approach above gets confused trying to estimate linear rates, while this will happily let a client "steal" a few rapid requests when demand is low.
Code here.
After some experimenting, it seems that the most important thing is getting as good an estimate as possible for the upper limit of the current connection rates.
Each client can track their own (local) connection rate using a queue of timestamps. A timestamp is added to the queue on each connection and timestamps older than a minute are discarded. The "long term" (over a minute) average rate is then found from the first and last timestamps and the number of entries (minus one). The "short term" (instantaneous) rate can be found from the times of the last two requests. The upper limit is the maximum of these two values.
Each client can also estimate the external connection rate (from the other clients). The "long term" rate can be found from the number of "used" connections in the last minute, as reported by the server, corrected by the number of local connections (from the queue mentioned above). The "short term" rate can be estimated from the "used" number since the previous request (minus one, for the local connection), scaled by the time difference. Again, the upper limit (maximum of these two values) is used.
Each client computes these two rates (local and external) and then adds them to estimate the upper limit to the total rate of connections to the server. This value is compared with the target rate band, which is currently set to between 80% and 90% of the maximum (0.8 to 0.9 * 120 per minute).
From the difference between the estimated and target rates, each client modifies their own connection rate. This is done by taking the previous delta (time between the last connection and the one before) and scaling it by 1.1 (if the rate exceeds the target) or 0.9 (if the rate is lower than the target). The client then refuses to make a new connection until that scaled delta has passed (by sleeping if a new connected is requested).
Finally, nothing above forces all clients to equally share the bandwidth. So I add an additional 10% to the local rate estimate. This has the effect of preferentially over-estimating the rate for clients that have high rates, which makes them more likely to reduce their rate. In this way the "greedy" clients have a slightly stronger pressure to reduce consumption which, over the long term, appears to be sufficient to keep the distribution of resources balanced.
The important insights are:
By taking the maximum of "long term" and "short term" estimates the system is conservative (and more stable) when additional clients start up.
No client knows the total number of clients (unless it is zero or one), but all clients run the same code so can "trust" each other.
Given the above, you can't make "exact" calculations about what rate to use, but you can make a "constant" correction (in this case, +/- 10% factor) depending on the global rate.
The adjustment to the client connection frequency is made to the delta between the last two connection (adjusting based on the average over the whole minute is too slow and leads to oscillations).
Balanced consumption can be achieved by penalising the greedy clients slightly.
In (limited) experiments this works fairly well (even in the worst case of multiple clients starting at once). The main drawbacks are: (1) it doesn't allow for an initial "burst" (which would improve throughput if the server has few clients and a client has only a few requests); (2) the system does still oscillate over ~ a minute (see below); (3) handling a larger number of clients (in the worst case, eg if they all start at once) requires a larger gain (eg 20% correction instead of 10%) which tends to make the system less stable.
The "used" amount reported by the (test) server, plotted against time (Unix epoch). This is for four clients (coloured), all trying to consume as much data as possible.
The oscillations come from the usual source - corrections lag signal. They are damped by (1) using the upper limit of the rates (predicting long term rate from instantaneous value) and (2) using a target band. This is why an answer informed by someone who understand control theory would be appreciated...
It's not clear to me that estimating local and external rates separately is important (they may help if the short term rate for one is high while the long-term rate for the other is high), but I doubt removing it will improve things.
In conclusion: this is all pretty much as I expected, for this kind of approach. It kind-of works, but because it's a simple feedback-based approach it's only stable within a limited range of parameters. I don't know what alternatives might be possible.
Since you're using the Echonest API, why don't you take advantage of the rate limit headers that are returned with every API call?
In general you get 120 requests per minute. There are three headers that can help you self-regulate your API consumption:
X-Ratelimit-Used
X-Ratelimit-Remaining
X-Ratelimit-Limit
**(Notice the lower-case 'ell' in 'Ratelimit'--the documentation makes you think it should be capitalized, but in practice it is lower case.)
These counts account for calls made by other processes using your API key.
Pretty neat, huh? Well, I'm afraid there is a rub...
That 120-request-per-minute is really an upper bound. You can't count on it. The documentation states that value can fluctuate according to system load. I've seen it as low as 40ish in some calls I've made, and have in some cases seen it go below zero (I really hope that was a bug in the echonest API!)
One approach you can take is to slow things down once utilization (used divided by limit) reaches a certain threshold. Keep in mind though, that on the next call your limit may have been adjusted download significantly enough that 'used' is greater than 'limit'.
This works well up until a point. Since the Echonest doesn't adjust the limit in a predictable mannar, it is hard to avoid 400s in practice.
Here are some links that I've found helpful:
http://blog.echonest.com/post/15242456852/managing-your-api-rate-limit
http://developer.echonest.com/docs/v4/#rate-limits
We are re-implementing(yes from scratch) a web application which is currently in production. We have decided to start doing some performance tests on the new app, to get some early information of the capabilities.
As the old application is currently in production and has a good performance we would like to extract some performance parameters, and then use this parameters as a reference or base goal of the performance of the new application.
Which do you think are the most relevant performance parameters we should be obtaining from the current production application?
Thanks!
Determine what pages are used the most.
Measure a latency histogram for the total time it takes to answer the request. Don't just measure the mean, measure a histogram.
From the histogram you can see how many % of requests have which latency in milliseconds. You can choose to key performance indicators by takes the values for 50% and 95%. This will tell you the average latency and the worst latency (for the worst 10% of requests).
Those two numbers alone will bring you great confidence regarding the experience your users will have.
Throughput does not matter for users, but for capacity planning.
I also recommend that you track the performance values over time and review them twice a year.
Just in case you need an HTTP client, there is weighttp, a multi-threaded client written by the guys from Lighttpd.
It has the same syntax used by ApacheBench, but weighttp lets you use several client worker threads (AB is single-threaded so it cannot saturate a modern SMP Web server).
The answer of "usr" is valid, but you can as well record the minnimum, average and maximum latencies (that's useful to see in which range they play). Here is a public-domain C program to automate all this on a given concurrency range.
Disclamer: I am involved in the development of this project.
Suppose you have a web application, no specific stack (Java/.NET/LAMP/Django/Rails, all good).
How would you decide on which hardware to deploy it? What rules of thumb exist when determining how many machines you need?
How would you formulate parameters such as concurrent users, simultaneous connections, daily hits and DB read/write ratio to a decision on how much, and which, hardware you need?
Any resources on this issue would be very helpful...
Specifically - any hard numbers from real world experience and case studies would be great.
Capacity Planning is quite a detailed and extensive area. You'll need to accept an iterative model with a "Theoretical Baseline > Load Testing > Tuning & Optimizing" approach.
Theory
The first step is to decide on the Business requirements: how many users are expected for peak usage ? Remember - these numbers are usually inaccurate by some margin.
As an example, let's assume that all the peak traffic (at worst case) will be over 4 hours of the day. So if the website expects 100K hits per day, we dont divide that over 24 hours, but over 4 hours instead. So my site now needs to support a peak traffic of 25K hits per hour.
This breaks down to 417 hits per minute, or 7 hits per second. This is on the front end alone.
Add to this the number of internal transactions such as database operations, any file i/o per user, any batch jobs which might run within the system, reports etc.
Tally all these up to get the number of transactions per second, per minute etc that your system needs to support.
This gets further complicated when you have requirements such as "Avg response time must be 3 seconds etc" which means you have to figure in network latency / firewall / proxy etc
Finally - when it comes to choosing hardware, check out the published datasheets from each manufacturer such as Sun, HP, IBM, Windows etc. These detail the maximum transactions per second under test conditions. We usually accept 50% of those peaks under real conditions :)
But ultimately the choice of the hardware is usually a commercial decision.
Also you need to keep a minimum of 2 servers at each tier : web / app / even db for failover clustering.
Load testing
It's recommended to have a separate reference testing environment throughout the project lifecycle and post-launch so you can come back to run dedicated performance tests on the app. Scale this to be a smaller version of production, so if Prod has 4 servers and Ref has 1, then you test for 25% of the peak transactions etc.
Tuning & Optimizing
Too often, people throw some expensive hardware together and expect it all to work beautifully. You'll need to tune the hardware and OS for various parameters such as TCP timeouts etc - these are published by the software vendors, and these have to be done once the software are finalized. Set these tuning params on the Ref env, test and then decide which ones you need to carry over to Production.
Determine your expected load.
Setup a machine and run some tests against it with a Load testing tool.
How close are you if you only accomplished 10% of the peak load with some margin for error then you know you are going to need some load balancing. Design and implement a solution and test again. Make sure you solution is flexible enough to scale.
Trial and error is pretty much the way to go. It really depends on the individual app and usage patterns.
Test your app with a sample load and measure performance and load metrics. DB queries, disk hits, latency, whatever.
Then get an estimate of the expected load when deployed (go ask the domain expert) (you have to consider average load AND spikes).
Multiply the two and add some just to be sure. That's a really rough idea of what you need.
Then implement it, keeping in mind you usually won't scale linearly and you probably won't get the expected load ;)
I want to create a system that delivers user interface response within 100ms, but which requires minutes of computation. Fortunately, I can divide it up into very small pieces, so that I could distribute this to a lot of servers, let's say 1500 servers. The query would be delivered to one of them, which then redistributes to 10-100 other servers, which then redistribute etc., and after doing the math, results propagate back again and are returned by a single server. In other words, something similar to Google Search.
The problem is, what technology should I use? Cloud computing sounds obvious, but the 1500 servers need to be prepared for their task by having task-specific data available. Can this be done using any of the existing cloud computing platforms? Or should I create 1500 different cloud computing applications and upload them all?
Edit: Dedicated physical servers does not make sense, because the average load will be very, very small. Therefore, it also does not make sense, that we run the servers ourselves - it needs to be some kind of shared servers at an external provider.
Edit2: I basically want to buy 30 CPU minutes in total, and I'm willing to spend up to $3000 on it, equivalent to $144,000 per CPU-day. The only criteria is, that those 30 CPU minutes are spread across 1500 responsive servers.
Edit3: I expect the solution to be something like "Use Google Apps, create 1500 apps and deploy them" or "Contact XYZ and write an asp.net script which their service can deploy, and you pay them based on the amount of CPU time you use" or something like that.
Edit4: A low-end webservice provider, offering asp.net at $1/month would actually solve the problem (!) - I could create 1500 accounts, and the latency is ok (I checked), and everything would be ok - except that I need the 1500 accounts to be on different servers, and I don't know any provider that has enough servers that is able to distribute my accounts on different servers. I am fully aware that the latency will differ from server to server, and that some may be unreliable - but that can be solved in software by retrying on different servers.
Edit5: I just tried it and benchmarked a low-end webservice provider at $1/month. They can do the node calculations and deliver results to my laptop in 15ms, if preloaded. Preloading can be done by making a request shortly before the actual performance is needed. If a node does not respond within 15ms, that node's part of the task can be distributed to a number of other servers, of which one will most likely respond within 15ms. Unfortunately, they don't have 1500 servers, and that's why I'm asking here.
[in advance, apologies to the group for using part of the response space for meta-like matters]
From the OP, Lars D:
I do not consider [this] answer to be an answer to the question, because it does not bring me closer to a solution. I know what cloud computing is, and I know that the algorithm can be perfectly split into more than 300,000 servers if needed, although the extra costs wouldn't give much extra performance because of network latency.
Lars,
I sincerely apologize for reading and responding to your question at a naive and generic level. I hope you can see how both the lack of specifity in the question itself, particularly in its original form, and also the somewhat unusual nature of the problem (1) would prompt me respond to the question in like fashion. This, and the fact that such questions on SO typically emanate from hypotheticals by folks who have put but little thought and research into the process, are my excuses for believing that I, a non-practionner [of massively distributed systems], could help your quest. The many similar responses (some of which had the benefits of the extra insight you provided) and also the many remarks and additional questions addressed to you show that I was not alone with this mindset.
(1) Unsual problem: An [apparently] mostly computational process (no mention of distributed/replicated storage structures), very highly paralellizable (1,500 servers), into fifty-millisecondish-sized tasks which collectively provide a sub-second response (? for human consumption?). And yet, a process that would only be required a few times [daily..?].
Enough looking back!
In practical terms, you may consider some of the following to help improve this SO question (or move it to other/alternate questions), and hence foster the help from experts in the domain.
re-posting as a distinct (more specific) question. In fact, probably several questions: eg. on the [likely] poor latency and/or overhead of mapreduce processes, on the current prices (for specific TOS and volume details), on the rack-awareness of distributed processes at various vendors etc.
Change the title
Add details about the process you have at hand (see many questions in the notes of both the question and of many of the responses)
in some of the questions, add tags specific to a give vendor or technique (EC2, Azure...) as this my bring in the possibly not quite unbuyist but helpful all the same, commentary from agents at these companies
Show that you understand that your quest is somewhat of a tall order
Explicitly state that you wish responses from effective practionners of the underlying technologies (maybe also include folks that are "getting their feet wet" with these technologies as well, since with the exception of the physics/high-energy folks and such, who BTW traditionnaly worked with clusters rather than clouds, many of the technologies and practices are relatively new)
Also, I'll be pleased to take the hint from you (with the implicit non-veto from other folks on this page), to delete my response, if you find that doing so will help foster better responses.
-- original response--
Warning: Not all processes or mathematical calculations can readily be split in individual pieces that can then be run in parallel...
Maybe you can check Wikipedia's entry from Cloud Computing, understanding that cloud computing is however not the only architecture which allows parallel computing.
If your process/calculation can efficitively be chunked in parallelizable pieces, maybe you can look into Hadoop, or other implementations of MapReduce, for an general understanding about these parallel processes. Also, (and I believe utilizing the same or similar algorithms), there also exist commercially available frameworks such as EC2 from amazon.
Beware however that the above systems are not particularly well suited for very quick response time. They fare better with hour long (and then some) data/number crunching and similar jobs, rather than minute long calculations such as the one you wish to parallelize so it provides results in 1/10 second.
The above frameworks are generic, in a sense that they could run processes of most any nature (again, the ones that can at least in part be chunked), but there also exist various offerings for specific applications such as searching or DNA matching etc. The search applications in particular can have very short response times (cf Google for example) and BTW this is in part tied to fact that such jobs can very easily and quickly be chunked for parallel processing.
Sorry, but you are expecting too much.
The problem is that you are expecting to pay for processing power only. Yet your primary constraint is latency, and you expect that to come for free. That doesn't work out. You need to figure out what your latency budgets are.
The mere aggregating of data from multiple compute servers will take several milliseconds per level. There will be a gaussian distribution here, so with 1500 servers the slowest server will respond after 3σ. Since there's going to be a need for a hierarchy, the second level with 40 servers , where again you'll be waiting for the slowest server.
Internet roundtrips also add up quickly; that too should take 20 to 30 ms of your latency budget.
Another consideration is that these hypothethical servers will spend much of their time idle. That means they're powered on, drawing electricity yet not generating revenue. Any party with that many idle servers would turn them off, or at the very least in sleep mode just to conserve electricity.
MapReduce is not the solution! Map Reduce is used in Google, Yahoo and Microsoft for creating the indexes out of the huge data (the whole Web!) they have on their disk. This task is enormous and Map Reduce was built to make it happens in hours instead of years, but starting a Master controller of Map Reduce is already 2 seconds, so for your 100ms this is not an option.
Now, from Hadoop you may get advantages out of the distributed file system. It may allow you to distribute the tasks close to where the data is physically, but that's it. BTW: Setting up and managing an Hadoop Distributed File System means controlling your 1500 servers!
Frankly in your budget I don't see any "cloud" service that will allow you to rent 1500 servers. The only viable solution, is renting time on a Grid Computing solution like Sun and IBM are offering, but they want you to commit to hours of CPU from what I know.
BTW: On Amazon EC2 you have a new server up in a couple of minutes that you need to keep for an hour minimum!
Hope you'll find a solution!
I don't get why you would want to do that, only because "Our user interfaces generally aim to do all actions in less than 100ms, and that criteria should also apply to this".
First, 'aim to' != 'have to', its a guideline, why would u introduce these massive process just because of that. Consider 1500 ms x 100 = 150 secs = 2.5 mins. Reducing the 2.5 mins to a few seconds its a much more healthy goal. There is a place for 'we are processing your request' along with an animation.
So my answer to this is - post a modified version of the question with reasonable goals: a few secs, 30-50 servers. I don't have the answer for that one, but the question as posted here feels wrong. Could even be 6-8 multi-processor servers.
Google does it by having a gigantic farm of small Linux servers, networked together. They use a flavor of Linux that they have custom modified for their search algorithms. Costs are software development and cheap PC's.
It would seem that you are indeed expecting at least 1000-fold speedup from distributing your job to a number of computers. That may be ok. Your latency requirement seems tricky, though.
Have you considered the latencies inherent in distributing the job? Essentially the computers would have to be fairly close together in order to not run into speed of light issues. Also, the data center in which the machines would be would again have to be fairly close to your client so that you can get your request to them and back in less than 100 ms. On the same continent, at least.
Also note that any extra latency requires you to have many more nodes in the system. Losing 50% of available computing time to latency or anything else that doesn't parallelize requires you to double the computing capacity of the parallel portions just to keep up.
I doubt a cloud computing system would be the best fit for a problem like this. My impression at least is that the proponents of cloud computing would prefer to not even tell you where your machines are. Certainly I haven't seen any latency terms in the SLAs that are available.
You have conflicting requirements. You're requirement for 100ms latency is directly at odds with your desire to only run your program sporadically.
One of the characteristics of the Google-search type approach you mentioned in your question is that the latency of the cluster is dependent on the slowest node. So you could have 1499 machines respond in under 100ms, but if one machine took longer, say 1s - whether due to a retry, or because it needed to page you application in, or bad connectivity - your whole cluster would take 1s to produce an answer. It's inescapable with this approach.
The only way to achieve the kinds of latencies you're seeking would be to have all of the machines in your cluster keep your program loaded in RAM - along with all the data it needs - all of the time. Having to load your program from disk, or even having to page it in from disk, is going to take well over 100ms. As soon as one of your servers has to hit the disk, it is game over for your 100ms latency requirement.
In a shared server environment, which is what we're talking about here given your cost constraints, it is a near certainty that at least one of your 1500 servers is going to need to hit the disk in order to activate your app.
So you are either going to have to pay enough to convince someone to keep you program active and in memory at all times, or you're going to have to loosen your latency requirements.
Two trains of thought:
a) if those restraints are really, absolutely, truly founded in common sense, and doable in the way you propose in the nth edit, it seems the presupplied data is not huge. So how about trading storage for precomputation to time. How big would the table(s) be? Terabytes are cheap!
b) This sounds a lot like a employer / customer request that is not well founded in common sense. (from my experience)
Let´s assume the 15 minutes of computation time on one core. I guess thats what you say.
For a reasonable amount of money, you can buy a system with 16 proper, 32 hyperthreading cores and 48 GB RAM.
This should bring us in the 30 second range.
Add a dozen Terabytes of storage, and some precomputation.
Maybe a 10x increase is reachable there.
3 secs.
Are 3 secs too slow? If yes, why?
Sounds like you need to utilise an algorithm like MapReduce: Simplified Data Processing on Large Clusters
Wiki.
Check out Parallel computing and related articles in this WikiPedia-article - "Concurrent programming languages, libraries, APIs, and parallel programming models have been created for programming parallel computers." ... http://en.wikipedia.org/wiki/Parallel_computing
Although Cloud Computing is the cool new kid in town, your scenario sounds more like you need a cluster, i.e. how can I use parallelism to solve a problem in a shorter time.
My solution would be:
Understand that if you got a problem that can be solved in n time steps on one cpu, does not guarantee that it can be solved in n/m on m cpus. Actually n/m is the theoretical lower limit. Parallelism is usually forcing you to communicate more and therefore you'll hardly ever achieve this limit.
Parallelize your sequential algorithm, make sure it is still correct and you don't get any race conditions
Find a provider, see what he can offer you in terms of programming languages / APIs (no experience with that)
What you're asking for doesn't exist, for the simple reason that doing this would require having 1500 instances of your application (likely with substantial in-memory data) idle on 1500 machines - consuming resources on all of them. None of the existing cloud computing offerings bill on such a basis. Platforms like App Engine and Azure don't give you direct control over how your application is distributed, while platforms like Amazon's EC2 charge by the instance-hour, at a rate that would cost you over $2000 a day.