MyClass is all about providing access to a single file. It must CheckHeader(), ReadSomeData(), UpdateHeader(WithInfo), etc.
But since the file that this class represents is very complex, it requires special design considerations.
That file contains a potentially huge folder-like tree structure with various node types and is block/cell based to handle fragmentation better. Size is usually smaller than 20 MB. It is not of my design.
How would you design such a class?
Read a ~20MB stream into memory?
Put a copy on temp dir and keep its path as property?
Keep a copy of big things on memory and expose them as read-only properties?
GetThings() from the file with exception-throwing code?
This class(es) will be used only by me at first, but if it ends good enough I might open-source it.
(This is a question on design, but platform is .NET and class is about offline registry access for XP)
It depends what you need to do with this data. If you only need to process it linearly one time, then it might be faster to just take the performance hit of a large file in memory.
If however you need to do various things with the file beyond a single, linear parsing, I would parse the data into a lightweight database such as SQLite and then operate on that. This way all of your file's structure is preserved and all subsequent operations on the file will be faster.
Registry access is quite complex. You are basically reading a large binary tree. The class design should rely heavily on the stored data structures. Only then you can choose an appropriate class design. To stay flexible you should model the primitives such as REG_SZ, REG_EXPAND_SZ, DWORD, SubKey, .... Don Syme has in his book Expert F# a nice section about binary parsing with binary combinators. The basic idea is that your objects know by themself how to deserialize from a binary representation. When you have a stream of bytes which is structured like this
<Header>
<Node1/>
<Node2>
<Directory1>
</Node2>
</Header>
you start with a BinaryReader to read the binary objects byte by byte. Since you know that the first thing must be the header you can pass it to the Header object
public class Header
{
static Header Deserialize(BinaryReader reader)
{
Header header = new Header();
int magic = reader.ReadByte();
if( magic == 0xf4 ) // we have a node entry
header.Insert(Node.Read( reader );
else if( magic == 0xf3 ) // directory entry
header.Insert(DirectoryEntry.Read(reader))
else
throw NotSupportedException("Invalid data");
return header;
}
}
To stay performant you can e.g. delay parsing the data up to a later time when specific properties of this or that instance are actually accessed.
Since the registry in Windows can get quite big it is not possible to read it completely into memory at once. You will need to chunk it. One solution that Windows applies is that the whole file is allocated in paged pool memory which can span several gigabytes but only the actually accessed parts are swapped out from disk into memory. That allows Windows to deal with a very large registry file in an efficient manner. You will need something similar for your reader as well. Lazy parsing is one aspect and the ability to jump around in the file without the need to read the data in between is cruical to stay performant.
More infos about paged pool and the registry can be found there:
http://blogs.technet.com/b/markrussinovich/archive/2009/03/26/3211216.aspx
Your Api design will depend on how you read the data to stay efficient (e.g. use a memory mapped file and read from different mapped regions). With .NET 4 a Memory Mapped file implementation has arrived that is quite good now but wrappers around the OS APIs exist as well.
Yours,
Alois Kraus
To support delayed loading from a memory mapped file it would make sense not to read the byte array into the object and parse it later but go one step furhter and store only the offset and length of the memory chunk from the memory mapped file. Later when the object is actually accessed you can read and deserialize the data. This way you can traverse the whole file and build a tree of objects which contain only the offsets and the reference to the memory mapped file. That should save huge amounts of memory.
Related
I am working on physics simulation research. I have a large fixed grid in one of my projects that does not vary with time. The fields on the grid, on the other hand, vary with time in the simulation. I need to use VTK to record the field data in each step for visualization (Paraview).
The method I am using is to write a separate *.vtu file to disk at each time step. This basically serves the purpose, but actually writes a lot of duplicate data (re-recording the geometry of the mesh at each step), which not only consumes more disk space, but also wastes time on encoding and parsing.
I would like to have a way to write the mesh information only once, and the rest of the time only new field data is written, while being able to guarantee the same visualization. Please let me know if VTK and Paraview provide such an interface and how to implement it.
Using .pvtu and refer to the same .vtu as Piece for each step should do the trick.
See this similar post on the ParaView discourse, and the pvtu doc
EDIT
This seems to be a side effect of the format, this is not supported by the writer.
The correct solution is to use another file format ...
Let me provide my own research findings for reference.
As Nico said, with the combination of pvtu/vtu files, we could theoretically implement a geometry structure stored in a separate vtu file, referenced by a pvtu file. Setting the NumberOfPieces attribute of the ptvu file to 1 would enable the construction of only one separate vtu file.
However, the VTK library does not expose a dedicated operation interface to control the writing process of vtu files. No matter how it is set, as long as the writer's input contains geometry structures, the writer will write geometry information to disk, and this process cannot be skipped through the exposed interface.
However, it is indeed possible to make multiple pvtu files point to the same vtu file by manually editing the piece node in the ptvu file, and paraview can recognize and visualize such a file group properly.
I did not proceed to try adding arrays to the unstructured grid and using pvtu output.
So, I think the conclusion is.
if you don't want to dive into VTK's library code and XML implementation, then this approach doesn't make sense.
if you are willing to write a series of files, delete most of them from the vtu file, and then point all the pvtu's piece nodes to the only surviving vtu file by editing the pvtu file, you can save a lot of disk space, but will not shorten the write, read, and parse times.
If you implement an XML writer by yourself, you can achieve all the requirements in theory, but it requires a lot of coding work.
I have to process XML, now using NSInputStream breaks my code as I have to rewrite lots of things.
Will dataWithContentsOfFile entire file into memory, or only read contents requested for getBytes method?
I am using NSData as input parameter to NSXMLParser, I wonder is there any documentation regarding this?
There is no documentation on apple's doc regarding internals of NSData's dataWithContentsOfFile or its implementation.
When you allocating NSData for NSXMALParser ,it means creating data buffer for that object and every object occupy memory (RAM) ,because iPhone IOS know very well how to use virtual memory. When you reading entire data from the file, it occupying object data and if the data size is more than a few(depended upon OS algo) memory pages, the object uses virtual memory management. A data object can also wrap preexisting data, regardless of how the data was allocated. The object contains no information about the data itself (such as its type); the responsibility for deciding how to use the data lies with the client. In particular, it will not handle byte-order swapping when distributed between big-endian and little-endian machines.
I recommend you to read again this link
https://developer.apple.com/library/mac/#documentation/Cocoa/Conceptual/BinaryData/BinaryData.html#//apple_ref/doc/uid/10000037i its related to iOS. But yes one thing concern with you ….in IOS there is type of owner of object. One is user and another one is IOS object. if you creating NSData that means you allocating memory buffer and assigning data pointer to this veriable, but at that moment entire data is resides inside memory. Its our assumption, but during this period IOS know how to handle this scenario. IOS uses vertual memory technique to handle data pages.
I am developing one application. In it I get 1000 results from xml parsing. Every result has different attributes. So I create one class for the attributes and create one object for every result. I save the results in one array. My doubt is due to the fact that this is a lot of results and I may face memory problems. If this is a problem, how do I handle that? Please tell me how to do this.
If you're parsing an exceptionally large document, use NSXMLParser and a delegate object to parse the document. Rather than creating an enormous tree of objects to represent the XML, the parser will call your delegate each time it encounters a new attribute, element, etc. This way you can build up your data objects directly, without wasting memory on an intermediate XML parse tree representation.
Once you are doing this, you can save the objects as you create them, or in batches.
If you're very memory conscious, you can actually use NSXMLParser to parse the input stream as it is downloading, so you never even need to have the full XML text in memory. (To avoid interruptions you could also download to a disk file, then parse from the file.)
Memory management in Objective-C is very well explained in below discussion..
http://www.iphonedevsdk.com/forum/iphone-sdk-tutorials/7295-getters-setters-properties-newbie.html
hope this will answer your question...
For those who have seen my other questions: I am making progress but I haven't yet wrapped my head around this aspect. I've been pouring over stackoverflow answers and sites like Cocoa With Love but I haven't found an app layout that fits (why such a lack of scientific or business app examples? recipe and book examples are too simplistic).
I have a data analysis app that is laid out like this:
Communication Manager (singleton, manages the hardware)
DataController (tells Comm.mgr what to do, and checks raw data it receives)
Model (receives data from datacontroller, cleans, analyzes and stores it)
MainViewController (skeleton right now, listens to comm.mgr to present views and alerts)
Now, never will my data be directly shown on a view (like a simple table of entities and attributes), I'll probably use core plot to plot the analyzed results (once I figure that out). The raw data saved will be huge (10,000's of points), and I am using a c++ vector wrapped in an ObjC++ class to access it. The vector class also has the encodeWithCoder and initWithCoder functions which use NSData as a transport for the vector. I'm trying to follow proper design practices, but I'm lost on how to get persistent storage into my app (which will be needed to store and review old data sets).
I've read several sources that say the "business logic" should go into the model class. This is how I have it right now, I send it the raw data, and it parses, cleans and analyzes the results and then saves those into ivar arrays (of the vector class). However, I haven't seen a Core Data example yet that has a Managed Object that is anything but a simple storage of very basic attributes (strings, dates) and they never have any business logic. So I wonder, how can I meld these two aspects? Should all of my analysis go into the data controller and have it manage the object context? If so, where is my model? (seems to break the MVC architecture if my data is stored in my controller - read: since these are vector arrays, I can't be constantly encoding and decoding them into NSData streams, they need a place to exist before I save them to disk with Core Data, and they need a place to exist after I retrieve them from storage and decode them for review).
Any suggestions would be helpful (even on the layout I've already started). I just drew some of the communication between objects to give you an idea. Also, I don't have any of the connections between the model and view/view controllers yet (using NSLog for now).
While vector<> is great for handling your data that you are sampling (because of its support for dynamically resizing underlying storage), you may find that straight C arrays are sufficient (even better) for data that is already stored. This does add a level of complexity but it avoids a copy for data arrays that are already of a known and static size.
NSData's -bytes returns a pointer to the raw data within an NSData object. Core Data supports NSData as one its attribute types. If you know the size of each item in data, then you can use -length to calculate the number of elements, etc.
On the sampling side, I would suggest using vector<> as you collect data and, intermittently, copy data to an NSData attribute and save. Note: I ran into a bit of problem with this approach (Truncated Core Data NSData objects) that I attribute to Core Data not recognizing changes made to NSData attribute when it is backed by an NSMutableData object and that mutable object's data is changed.
As for MVC question. I would suggest that data (model) is managed in by Model. Views and Controllers can ask Model for data (or subsets of data) in order to display. But ownership of data is with the Model. In my case, which may be similar to yours, there were times when the Model returns abridged data sets (using Douglas-Peucker algorithm). The views and controllers were none the wiser that points were being dropped - even though their requests to the Model may have played in a role in that (graph scaling factors, etc.).
Update
Here is a snippet of code from my Data class which extends NSManagedObject. For a filesystem solution, NSFileHandle's -writeData: and methods for monitoring file offset might allow similar (better) management controls.
// Exposed interface for adding data point to stored data
- (void) addDatum:(double_t)datum
{
[self addToCache:datum];
}
- (void) addToCache:(double_t)datum
{
if (cache == nil)
{
// This is temporary. Ideally, cache is separate from main store, but
// is appended to main store periodically - and then cleared for reuse.
cache = [NSMutableData dataWithData:[self dataSet]];
[cache retain];
}
[cache appendBytes:&datum length:sizeof(double_t)];
// Periodic copying of cache to dataSet could happen here...
}
// Called at end of sampling.
- (void) wrapup
{
[self setDataSet:[NSData dataWithData:cache]]; // force a copy to alert Core Data of change
[cache release];
cache = nil;
}
I think you may be reading into Core Data a bit too much. I'm not that experienced with it, so I speak as a non-expert, but there are basically two categories of data storage systems.
First is the basic database, like SQLite, PostgreSQL, or any number of solutions. These are meant to store data and retrieve it; there's no logic so you have to figure out how to manage the tables, etc. It's a bit tricky at times but they're very efficient. They're best at managing lots of data in raw form; if you want objects with that data you have to create and manage them yourself.
Then you have something like Core Data, which shouldn't be considered a database as much as an "object persistence" framework. With Core Data, you create objects, store them, and then retrieve them as objects. It's great for applications where you have large numbers of objects that each contain several pieces of data and relationships with other objects.
From my limited knowledge of your situation, I would venture a guess that a true database may be better suited to your needs, but you'll have to make the decision there (like I said, I don't know much about your situation).
As for MVC, my view is that the view should only contain display code, the model should only contain code for managing the data itself and its storage, and the controller should contain the logic that processes the data. In your case it sounds like you're gathering raw data and processing it before storing it, in which case you'd want to have another object to process the data before storing it in the model, then a separate controller to sort, manage, and otherwise prepare the data before the view receives it. Again, this may not be the best explanation (or the best methods for your situation) so take it with a grain of salt.
EDIT: Also, if you're looking on getting into Core Data more, I like this book. It explains the whole object-persistence-vs-database concept a lot better than I can.
I have created a package using Moose and I would like to nstore some large instances. The resulting binary files are very large (500+MB) so I would like to compress them.
What is the best way for doing that?
Should I open a filehandle with bzip etc. then store using fd_nstore?
With MooseX::Storage, most of this is already done for you -- you just need to specify your serialization and I/O formats.
While compression is certainly a viable option, you might also want consider simply serializing less.
Could it be that your objects contain a lot of data that could easily be rebuilt from other data they also contain? For example, if you have attributes that are lazily build from other attributes (e.g. using Moose's lazy + builder or lazy_build), there is not much point in storing the values of those attributes at all unless the recomputation is incredibly expensive. And even then it might be worth considering, as reading lots of data off disk isn't the fastest thing either.
If you find that you want to serialize only parts of your objects, and still want to use Storable, you can define custom STORABLE_freeze and STORABLE_thaw hooks, as described in the Storable documentation.
However, there's also alternative serializers available. MooseX::Storage is one of them, and happens to support many serialization backends and formats, and can also be told easily about which attributes to serialize and which to skip for that purpose.
Have a look at Data::Serializer. It optionally uses zlib (via Compress::Zlib) or PPMd (via Compress::PPMd) to compress your serialized data.