Since Neo4j works primarily in memory, I was wondering if it would be advantageous to enable hugepages (https://www.kernel.org/doc/Documentation/vm/hugetlbpage.txt) in my Linux Kernel, and then XX:+UseLargePages or maybe -XX:+UseHugeTLBFS in the (OpenJDK 8) JVM ?
If so, what rule of thumb should I use to decide how many hugepages to configured?
The Neo4j Performance Guide (http://neo4j.com/docs/stable/performance-guide.html) does not mention this, and Google didn't turn up anyone else discussing it (in the first couple of search pages anyway), so I thought I'd ask.
I'm wrestling to get acceptable performance from my new Neo4j instance (2.3.2-community). Any little bit will help. I want to know if this is worth trying before I bring down the database to change JVM flags... I'm hoping someone else has done some experiments along these lines already.
Since Neo4j does its own file paging and doesn't rely on the OS to do this, it should be advantageous or at least not hurt. Huge pages will reduce the probability of TLB cache misses when you use a large amount of memory, which Neo4j often would like to do when there's a lot of data stored in it.
However, Neo4j does not directly use hugepages even though it could and it would be a nice addition. This means you have to rely on transparent huge pages and whatever features the JVM provides. The transparent huge pages can cause more-or-less short stalls when smaller pages are merged.
If you have a representative staging environment then I advise you to make the changes there first, and measure their effect.
Transparent huge pages are mostly a problem for programs that use mmap because I think it can lead to changing the size of the unit of IO, which will make the hard-pagefault latency much higher. I'm not entirely sure about this, though, so please correct me if I'm wrong.
The JVM actually does use mmap for telemetry and tooling, through a file in /tmp so make sure this directory is mounted on tmpfs to avoid gnarly IO stalls, for instance during safe-points (!!!). Always do this even if you don't use huge pages.
Also make sure you are using the latest Linux kernel and the latest Java version.
You may be able to squeeze some percentage points out of it with tuning G1, but this is a bit of a black art.
Related
The project I am working on currently uses Neo4j community. Currently we process 1-5M vertices with 5-20M edges but we aim to handle a volume of 10-20M vertices w/ 50-100M edges.
We are discussing the idea of switching to a graph database open source project that would enable us to scale in these proportion. Currently our mind is set on Janusgraph with Cassandra.
We have some questions regarding the capabilities and development of Janusgraph, we ould be glad if someone could answer! (Maybe Misha Brukman or Aaron Ploetz?)
On Janusgraph capabilities:
We did some experiments using the Janusgraph ready-to-use docker image, queries being issued through a java program. The java program and docker image are run on the same machine. At the magnitude of 10k-20k vertices with 50k-100k edges inserted, a query to with all the vertices possessing a give property takes 8 to 10 seconds (mean time over 10 identical queries, time elapsed before and after the command in the java program). The command itself is really simple:
g.V().has("secText", "some text").inE().outV();
Moreover, the docker image seems to break down when I try to insert more record (extending towards 100k vertices).
We wonder if it's due to the limited nature of the docker image or if there is any problem or if it could be normal? Anyway it seems really, really slow.
We set up a 2 nodes Cassandra cluster (on 2 different VMs) with Janusgraph on town, again the results were quite slow.
From what I read on the Internet, people seem to use Janusgraph deployment with millions of vertices in production, so I guess they can execute simple queries in matter of milliseconds. What is the secret there? Do you need like 128GB of RAM for the whole thing to perform correctly? Or maybe there is a guide a good practices to follow that I am unaware of? I tried my best using Janusgraph official documentation and user comments on forums like here but that ain't much I'm afraid :/
On Janusgraph future:
Janusgraph seemed to evolve quite quickly over the first years (like 2016-2018) but the past few monthes I didn't see much activity from the Janusgraph community, except for the release of version 0.5 a few monthes ago. For example, no meeting since last year.
So I'm wondering: is Janusgraph on the right tracks to last and be maintained for many years to come. Did things slow down a bit because of COVID or is there a thing?
Is backward compatibility considered in Janusgraph? From what I can read in the docs, many things have changed from version 0.2/0.3 to 0.4 and 0.5. Many are to come like, for example, Cassandra Thrift and embedded being deprecated. So, in a production environment where we can't always afford to update version every year, let aside the code modification in a case where some component is deprecated, does Janusgraph dev think of achieving some backward compatibility soon, or maybe should we still wait for the 1.0 version for that?
Thank you for reading all this and I am looking forward to all the answers you can give me :) have a nice day!
Mael
JanusGraph with Cassandra has design limitations at the storage layer which makes performance slow. In practice, its a large, scaleable, but slow graph database that offers the replication and redundancy benefits of Cassandra.
Cassandra shards data and is very good at distributing data randomly across the cluster, however this destroys data locality which is needed to make traversals fast and efficient. JanusGraph also supports several back-end storage options in addition to Cassandra, which means its not tightly tuned to any particular storage architecture.
Memory can make a difference, so verify how much memory you have allocated to the JVM on each node, use G1GC and disable swap. The VisualVM is helpful to profile your memory headroom.
Hello I know this might be late but please tell me. Are you accessing all the vertices for analysis or transactional queries ? OLAP or OLTP ? because how many vertices and edges you query and how you do that has a major effect. for example do you tell Janusgraph to return a vertex that have millions of edges with all those edges in one shot or only few of them. this is referred to as the hot vertex ( a vertex that has a lot of edges that cant be stored on one server instance ).
And if so, how. I'm talking about this 4GB Patch.
On the face of it, it seems like a pretty nifty idea: on Windows, each 32-bit application normally only has access to 2GB of address space, but if you have 64-bit Windows, you can enable a little flag to allow a 32-bit application to access the full 4GB. The page gives some examples of applications that might benefit from it.
HOWEVER, most applications seem to assume that memory allocation is always successful. Some applications do check if allocations are successful, but even then can at best quit gracefully on failure. I've never in my (short) life come across an application that could fail a memory allocation and still keep going with no loss of functionality or impact on correctness, and I have a feeling that such applications are from extremely rare to essentially non-existent in the realm of desktop computers. With this in mind, it would seem reasonable to assume that any such application would be programmed to not exceed 2GB memory usage under normal conditions, and those few that do would have been built with this magic flag already enabled for the benefit of 64-bit users.
So, have I made some incorrect assumptions? If not, how does this tool help in practice? I don't see how it could, yet I see quite a few people around the internet claiming it works (for some definition of works).
Your troublesome assumptions are these ones:
Some applications do check if allocations are successful, but even then can at best quit gracefully on failure. I've never in my (short) life come across an application that could fail a memory allocation and still keep going with no loss of functionality or impact on correctness, and I have a feeling that such applications are from extremely rare to essentially non-existent in the realm of desktop computers.
There do exist applications that do better than "quit gracefully" on failure. Yes, functionality will be impacted (after all, there wasn't enough memory to continue with the requested operation), but many apps will at least be able to stay running - so, for example, you may not be able to add any more text to your enormous document, but you can at least save the document in its current state (or make it smaller, etc.)
With this in mind, it would seem reasonable to assume that any such application would be programmed to not exceed 2GB memory usage under normal conditions, and those few that do would have been built with this magic flag already enabled for the benefit of 64-bit users.
The trouble with this assumption is that, in general, an application's memory usage is determined by what you do with it. So, as over the past years storage sizes have grown, and memory sizes have grown, the sizes of files that people want to operate on have also grown - so an application that worked fine when 1GB files were unheard of may struggle now that (for example) high definition video can be taken by many consumer cameras.
Putting that another way: applications that used to fit comfortably within 2GB of memory no longer do, because people want do do more with them now.
I do think the following extract from your link of 4 GB Patch pretty much explains the reason of how and why it works.
Why things are this way on x64 is easy to explain. On x86 applications have 2GB of virtual memory out of 4GB (the other 2GB are reserved for the system). On x64 these two other GB can now be accessed by 32bit applications. In order to achieve this, a flag has to be set in the file's internal format. This is, of course, very easy for insiders who do it every day with the CFF Explorer. This tool was written because not everybody is an insider, and most probably a lot of people don't even know that this can be achieved. Even I wouldn't have written this tool if someone didn't explicitly ask me to.
And to expand on CFF,
The CFF Explorer was designed to make PE editing as easy as possible,
but without losing sight on the portable executable's internal
structure. This application includes a series of tools which might
help not only reverse engineers but also programmers. It offers a
multi-file environment and a switchable interface.
And to quote a Microsoft insider, Larry Miller of Microsoft MCSA on a blog post about patching games using the tool,
Under 32 bit windows an application has access to 2GB of VIRTUAL
memory space. 64 bit Windows makes 4GB available to applications.
Without the change mentioned an application will only be able to
access 2GB.
This was not an arbitrary restriction. Most 32 bit applications simply
can not cope with a larger than 2GB address space. The switch
mentioned indicates to the system that it is able to cope. If this
switch is manually set most 32 bit applications will crash in 64 bit
environment.
In some cases the switch may be useful. But don't be surprised if it
crashes.
And finally to add from MSDN - Migrating 32-bit Managed Code to 64-bit,
There is also information in the PE that tells the Windows loader if
the assembly is targeted for a specific architecture. This additional
information ensures that assemblies targeted for a particular
architecture are not loaded in a different one. The C#, Visual Basic
.NET, and C++ Whidbey compilers let you set the appropriate flags in
the PE header. For example, C# and THIRD have a /platform:{anycpu,
x86, Itanium, x64} compiler option.
Note: While it is technically possible to modify the flags in the PE header of an assembly after it has been compiled, Microsoft does not recommend doing this.
Finally to answer your question - how does this tool help in practice?
Since you have malloc in your tags, I believe you are working on unmanaged memory. This patch would mostly result in invalid pointers as they become twice the size now, and almost all other primitive datatypes would be scaled by a factor of 2X.
But for managed code since all these are handled by the CLR in .NET, this would mean really helpful and would not have much problems unless you are dealing with any of the following :
Invoking platform APIs via p/invoke
Invoking COM objects
Making use of unsafe code
Using marshaling as a mechanism for sharing information
Using serialization as a way of persisting state
To summarize, being a programmer I would not use the tool to convert my application and rather would migrate it myself by changing build targets. being said that if I have a exe that can do well like games with more RAM, then this is worth a try.
I have an image processing routine that I believe could be made very parallel very quickly. Each pixel needs to have roughly 2k operations done on it in a way that doesn't depend on the operations done on neighbors, so splitting the work up into different units is fairly straightforward.
My question is, what's the best way to approach this change such that I get the quickest speedup bang-for-the-buck?
Ideally, the library/approach I'm looking for should meet these criteria:
Still be around in 5 years. Something like CUDA or ATI's variant may get replaced with a less hardware-specific solution in the not-too-distant future, so I'd like something a bit more robust to time. If my impression of CUDA is wrong, I welcome the correction.
Be fast to implement. I've already written this code and it works in a serial mode, albeit very slowly. Ideally, I'd just take my code and recompile it to be parallel, but I think that that might be a fantasy. If I just rewrite it using a different paradigm (ie, as shaders or something), then that would be fine too.
Not require too much knowledge of the hardware. I'd like to be able to not have to specify the number of threads or operational units, but rather to have something automatically figure all of that out for me based on the machine being used.
Be runnable on cheap hardware. That may mean a $150 graphics card, or whatever.
Be runnable on Windows. Something like GCD might be the right call, but the customer base I'm targeting won't switch to Mac or Linux any time soon. Note that this does make the response to the question a bit different than to this other question.
What libraries/approaches/languages should I be looking at? I've looked at things like OpenMP, CUDA, GCD, and so forth, but I'm wondering if there are other things I'm missing.
I'm leaning right now to something like shaders and opengl 2.0, but that may not be the right call, since I'm not sure how many memory accesses I can get that way-- those 2k operations require accessing all the neighboring pixels in a lot of ways.
Easiest way is probably to divide your picture into the number of parts that you can process in parallel (4, 8, 16, depending on cores). Then just run a different process for each part.
In terms of doing this specifically, take a look at OpenCL. It will hopefully be around for longer since it's not vendor specific and both NVidia and ATI want to support it.
In general, since you don't need to share too much data, the process if really pretty straightforward.
I would also recommend Threading Building Blocks. We use this with the IntelĀ® Integrated Performance Primitives for the image analysis at the company I work for.
Threading Building Blocks(TBB) is similar to both OpenMP and Cilk. And it uses OpenMP to do the multithreading, it is just wrapped in a simpler interface. With it you don't have to worry about how many threads to make, you just define tasks. It will split the tasks, if it can, to keep everything busy and it does the load balancing for you.
Intel Integrated Performance Primitives(Ipp) has optimized libraries for vision. Most of which are multithreaded. For the functions we need that aren't in the IPP we thread them using TBB.
Using these, we obtain the best result when we use the IPP method for creating the images. What it does is it pads each row so that any given cache line is entirely contained in one row. Then we don't divvy up a row in the image across threads. That way we don't have false sharing from two threads trying to write to the same cache line.
Have you seen Intel's (Open Source) Threading Building Blocks?
I haven't used it, but take a look at Cilk. One of the big wigs on their team is Charles E. Leiserson; he is the "L" in CLRS, the most widely/respected used Algorithms book on the planet.
I think it caters well to your requirements.
From my brief readings, all you have to do is "tag" your existing code and then run it thru their compiler which will automatically/seamlessly parallelize the code. This is their big selling point, so you dont need to start from scratch with parallelism in mind, unlike other options (like OpenMP).
If you already have a working serial code in one of C, C++ or Fortran, you should give serious consideration to OpenMP. One of its big advantages over a lot of other parallelisation libraries / languages / systems / whatever, is that you can parallelise a loop at a time which means that you can get useful speed-up without having to re-write or, worse, re-design, your program.
In terms of your requirements:
OpenMP is much used in high-performance computing, there's a lot of 'weight' behind it and an active development community -- www.openmp.org.
Fast enough to implement if you're lucky enough to have chosen C, C++ or Fortran.
OpenMP implements a shared-memory approach to parallel computing, so a big plus in the 'don't need to understand hardware' argument. You can leave the program to figure out how many processors it has at run time, then distribute the computation across whatever is available, another plus.
Runs on the hardware you already have, no need for expensive, or cheap, additional graphics cards.
Yep, there are implementations for Windows systems.
Of course, if you were unwise enough to have not chosen C, C++ or Fortran in the beginning a lot of this advice will only apply after you have re-written it into one of those languages !
Regards
Mark
We are planning to develop a datamining package for windows. The program core / calculation engine will be developed in F# with GUI stuff / DB bindings etc done in C# and F#.
However, we have not yet decided on the model implementations. Since we need high performance, we probably can't use managed code here (any objections here?). The question is, is it reasonable to develop the models in FORTRAN or should we stick to C (or maybe C++). We are looking into using OpenCL at some point for suitable models - it feels funny having to go from managed code -> FORTRAN -> C -> OpenCL invocation for these situations.
Any recommendations?
F# compiles to the CLR, which has a just-in-time compiler. It's a dialect of ML, which is strongly typed, allowing all of the nice optimisations that go with that type of architecture; this means you will probably get reasonable performance from F#. For comparison, you could also try porting your code to OCaml (IIRC this compiles to native code) and see if that makes a material difference.
If it really is too slow then see how far that scaling hardware will get you. With the performance available through a modern PC or server it seems unlikely that you would need to go to anything exotic unless you are working with truly brobdinagian data sets. Users with smaller data sets may well be OK on an ordinary PC.
Workstations give you perhaps an order of magnitude more capacity than a standard dekstop PC. A high-end workstation like a HP Z800 or XW9400 (similar kit is available from several other manufacturers) can take two 4 or 6 core CPU chips, tens of gigabytes of RAM (up to 192GB in some cases) and has various options for high-speed I/O like SAS disks, external disk arrays or SSDs. This type of hardware is expensive but may be cheaper than a large body of programmer time. Your existing desktop support infrastructure shouldn be able to this sort of kit. The most likely problem is compatibility issues running 32 bit software on a 64-bit O/S. In this case you have various options like VMs or KVM switches to work around the compatibility issues.
The next step up is a 4 or 8 socket server. Fairly ordinary wintel servers go up to 8 sockets (32-48 cores) and perhaps 512GB of RAM - without having to move off the Wintel platform. This gives you fairly wide range of options within your platform of choice before you have to go to anything exotic1.
Finally, if you can't make it run quickly in F#, validate the F# prototype and build a C implementation using the F# prototype as a control. If that's still not fast enough you've got problems.
If your application can be structured in a way that suits the platform then you could look at a more exotic platform. Depending on what will work with your application, you might be able to host it on a cluster, cloud provider or build the core engine on a GPU, Cell processor or FPGA. However, in doing this you're getting into (quite substantial) additional costs and exotic dependencies that might cause support issues. You will probably also have to bring a third-party consultant who knows how to program the platform.
After all that, the best advice is: suck it and see. If you're comfortable with F# you should be able to prototype your application fairly quickly. See how fast it runs and don't worry too much about performance until you have some clear indication that it really will be an issue. Remember, Knuth said that premature optimisation is the root of all evil about 97% of the time. Keep a weather eye out for issues and re-evaluate your strategy if you think performance really will cause trouble.
Edit: If you want to make a packaged application then you will probably be more performance-sensitive than otherwise. In this case performance will probably become an issue sooner than it would with a bespoke system. However, this doesn't affect the basic 'suck it and see' principle.
For example, at the risk of starting a game of buzzword bingo, if your application can be parallelized and made to work on a shared-nothing architecture you might see if one of the cloud server providers [ducks] could be induced to host it. An appropriate front-end could be built to run locally or through a browser. However, on this type of architecture the internet connection to the data source becomes a bottleneck. If you have large data sets then uploading these to the service provider becomes a problem. It may be quicker to process a large dataset locally than to upload it through an internet connection.
I would advise not to bother with optimizations yet. First try to get a working prototype, then find out where computation time is spent. You can probably move the biggest bottlenecks out into C or Fortran when and if needed -- then see how much difference it makes.
As they say, often 90% of the computation is spent in 10% of the code.
Hypothetically, if I were to build the same app using a few popular/similar frameworks, say PHP(cakePHP|Zend), Django, and Rails, should the memory consumption of each be roughly the same?
Also, I'm sure many have evaluated or used each and would be interested in which you settled on and why?
Code with whatever framework you like best. Then pray your app is popular enough to cause memory problems. We should all be so lucky.
No, it will absolutely vary wildly from one framework to another.
That said, in most cases the memory footprint of the framework is not the determining factor in site performance nor in selection of a framework. It's usually more a matter of using the right tool for the job, since each framework has its own strengths and weaknesses.
It is hard to efficiently say, I would say that PHP frameworks will have mostly a similar footprint, which is typically less than other frameworks such as Rails and Django. But it depends what you include as rails, such as mongrel (rails server proxy). Overall it depends on your code as well however PHP will most of the time give an easier time on the server. (Without any language Bias, I use both PHP and Rails)
Just for getting some perspective let me report a real case memory consumption using a Smalltalk web framework AIDA/Web.
For running 40+ websites on a single Smalltalk image on a single server it currently consumes 330MB of memory.
The only one of those frameworks I have used is CakePHP. I found that it's not to bad footprint wise however it is a lot more heavy that normal PHP without using a framework obviously but can be a good trade off.
A good comparison of some of the most popular PHP frameworks can be found at http://www.avnetlabs.com/php/php-framework-comparison-benchmarks.
Memory is cheap these days. Go with what will make your development easiest (which is usually what your team knows best).
But... In my experience, Django isn't terribly memory hungry. I've run it on my shared host with less than 100 MB of RAM. But my experience is sheerly anecdotal. YMMV. If you go with Django, here are some tips to keep memory usage down.
EDIT: And don't go with zope if memory footprint is important to you.