I have the following problem to solve. I have to build a graph viewer to view a massive data set.
We have some files in a particular format that has millions of records representing the result of an experiment. Each record represents a sample point on a large graph plot. The biggest file I have seen has 43.7 Million records.
An average file contains 10 Million records. Each record is small (76 Bytes + optional 12 Bytes each). The complete data cannot be loaded in to the main memory as it is too large. I have build a new file format that compresses the data to 48 bytes per record and organises the data in to chunks that are associated to each other. I want to "view" the data by displaying the records in a 2D/3D plot. As the data is very dense, I would like to progressively increase the level of detail by loading more data and removing data that is not shown in the view from the main memory.
I would also like to access group of associated records in real time and pre-load similar records in order to keep the loading time to bare minimum. This will give the user a smooth control to view the data instead of an experience similar to viewing a video on YouTube with a very slow internet connection. the user cannot randomly and has to use the controls to navigate and I would like to use this info to load the relevant records into the main memory.
The data has to be loaded progressively from the disc based on what is currently in the main memory. Records in the main memory that are not required in the current context can be removed and if required re loaded.
How to I access data from a disc at high speeds based on some hash number
How do I manage main memory if the data to be viewed in the current context is too large. If your answer is level of detail, then how do I build it for a large data set and should this data be part of the file ?
I have been working on this for the last two weeks and I seem to get stuck due to IO speed.
I am working in native C++ and I cannot use work under GPL. If you need any more info, let me know.
Ram
Under most modern file systems (Linux, Unixes, Windows) you can map a file into memory.
Which means you can access the content of the file as if it was entirely in memory (eg you can use data[i++], strchr(data,..), etc) and it's the operating system that does the mapping between used memory and file. When you want to read some data that is not already in memory, the o/s will fetch it from the file.
You should read this question's answer: Mmap() an entire large file
I think you are looking for organization similar to what's used to store level geometry in games, just that you maybe (depending on how your program works and what data you need to show) need just one dimension. See Quadtree and similar methods (bottom of that article).
Related
I have a working integration of FatFS in my C++ application running on a Cortex M4-based platform.
My application consists of logging data to a data format called MDF.
On the implementation side, I log data (to a given file) in batches of buffers; The number of buffers depends on how fast I acquire the data: log batch of one buffer . . . do other stuff . . . log batch of five buffer . . . do other stuff . . . etc.
There is also a header which is 24 bytes and contains the number of bytes of data. On a PC, I would just save the header at the end of the measurement but this is an embedded product which could be de-powered at any point in time. If I don't save the header periodically, the file becomes "corrupted".
Therefore, in order to maintain coherency I need to re-save the header after saving every batch of data and that's where my issue is.
This means that I have to call f_lseek before writing the header and then before I write the batch of data.
I am using f_cache_fptr so f_lseek is not painfully slow but I'd like to avoid needing to call f_lseek so frequently.
QUESTION
Is it possible to somehow have 2 seek locations so that I don't need to call f_seek to ping-pong between header-location and data-location?
I am open to modifying FatFS.
The problem, at the low-level, is simpler because the header only shares one 512 byte sector with the data: 24 bytes of header followed by 488 bytes of data.
Is it possible to somehow have 2 seek locations so that I don't need to call f_seek to ping-pong between header-location and data-location?
Not as far as I can tell, no, and it doesn't really seem to make sense. A FIL has only one current position, indicating where the next data written to it will go. What would it even mean for there to be two? How would the system know where to write? It certainly wouldn't be correct to write to both places.
Note in particular that with some operating systems and file systems, it is possible to open the same file more than once, but FatFS supports duplicate file opens only when all openings involved are for read-only mode.
I guess it would be possible to modify FatFS to give it the ability to store one file position when you seek to another, and then later to return to the first. So that would mean adding at least one member to the FIL structure, and adding at least one new function.
But why muck with the innards of FatFS? That's going to be at least a little risky. As long as you have to add a function anyway, how about just implementing a FRESULT my_f_write_at_beginning(FIL* fp, const void* buff, UINT btw, UINT* bw) on top of the existing functions? It can store the current position, seek to the beginning of the file, perform the write (maybe ensuring that the full number of bytes specified is written), and then seek back to the original position.
But fundamentally, no, there is no escaping ping-ponging back and forth, because doing so is part of the requirement you laid out.
On a PC, I would just save the header at the end of the measurement but this is an embedded product which could be de-powered at any point in time. If I don't save the header periodically, the file becomes "corrupted".
Therefore, in order to maintain coherency I need to re-save the header after saving every batch of data and that's where my issue is.
More correctly; you need to save the buffer and the header (footer?), and update the directory entry to reflect the new file size, and update the file allocation table to account for sectors allocated; and you need to write to at least 3 completely separate sectors "atomically" so that everything is consistent if there's a power failure at the wrong time.
This isn't entirely possible on most hardware.
However, there is a way to do it "somewhat safely". Specifically:
pre-allocate enough clusters for a completely new copy of the file (including the new data to append to the end) and update the file allocation table accordingly. If there's a power failure while doing this (or immediately after this point) the risk is lost clusters, which is an "ignore-able" problem that will waste some space but can be fixed easily with a typical "check disk" utility.
create a whole new copy of the file's data in the pre-allocated clusters (copy the old data, then append the new data and header). If there's a power failure in the middle of doing this (or immediately after this point), then the risk is the same as before - just some lost clusters (ignore-able).
atomically update the directory entry; changing both the file size and the "starting cluster number" with the same atomic (single sector) write. If there's a power failure after this point the risk is the same lost clusters (where the old version of the file's data was instead of where the new version of the file data is).
free the clusters that the old version of the file used by doing writes to the file allocation table. After this point you've completed successfully, so a power failure is fine.
To make this less awful for performance you can have two "cluster chains" and alternate between them; such that one chain of clusters is for the current version of the file and the other will become the next version of the file. This avoids the need to copy a lot of older data from one place to another (if you know the old data is still in previously used clusters). It could also avoid the need to allocate and free most clusters in the file allocation table, but only with a significant increase in the risk of lost clusters.
Of course for any of this to work you'd need a guarantee that single-sector writes are atomic; and you can't be using FAT12 (where an entry in the file allocation table can be split by a sector boundary).
I have a file as follows:
The file consists of 2 parts: header and data.
The data part is separated into equally sized pages. Each page holds data for a specific metric. Multiple pages (needs not to be consecutive) might be needed to hold data for a single metric. Each page consists of a page header and a page body. A page header has a field called "Next page" that is the index of the next page that holds data for the same metric. A page body holds real data. All pages have the same & fixed size (20 bytes for header and 800 bytes for body (if data amount is less than 800 bytes, 0 will be filled)).
The header part consists of 20,000 elements, each element has information about a specific metric (point 1 -> point 20000). An element has a field called "first page" that is actually index of the first page holding data for the metric.
The file can be up to 10 GB.
Requirement: Re-order data of the file in the shortest time, that is, pages holding data for a single metric must be consecutive, and from metric 1 to metric 20000 according to alphabet order (header part must be updated accordingly).
An apparent approach: For each metric, read all data for the metric (page by page), write data to new file. But this takes much time, especially when reading data from the file.
Is there any efficient ways?
One possible solution is to create an index from the file, containing the page number and the page metric that you need to sort on. Create this index as an array, so that the first entry (index 0) corresponds to the first page, the second entry (index 1) the second page, etc.
Then you sort the index using the metric specified.
When sorted, you end up with a new array which contains a new first, second etc. entries, and you read the input file writing to the output file in the order of the sorted index.
An apparent approach: For each metric, read all data for the metric (page by page), write data to new file. But this takes much time, especially when reading data from the file.
Is there any efficient ways?
Yes. After you get a working solution, measure it's efficiency, then decide which parts you wish to optimize. What and how you optimize will depend greatly on what results you get here (what are your bottlenecks).
A few generic things to consider:
if you have one set of steps that read data for a single metric and move it to the output, you should be able to parallelize that (have 20 sets of steps instead of one).
a 10Gb file will take a bit to process regardless of what hardware you run your code on (concievably, you could run it on a supercomputer but I am ignoring that case). You / your client may accept a slower solution if it displays it's progress / shows a progress bar.
do not use string comparisons for sorting;
Edit (addressing comment)
Consider performing the read as follows:
create a list of block offset for the blocks you want to read
create a list of worker threads, of fixed size (for example, 10 workers)
each idle worker will receive the file name and a block offset, then create a std::ifstream instance on the file, read the block, and return it to a receiving object (and then, request another block number, if any are left).
read pages should be passed to a central structure that manages/stores pages.
Also consider managing the memory for the blocks separately (for example, allocate chunks of multiple blocks preemptively, when you know the number of blocks to be read).
I first read header part, then sort metrics in alphabetic order. For each metric in the sorted list I read all data from the input file and write to the output file. To remove bottlenecks at reading data step, I used memory mapping. The results showed that when using memory mapping the execution time for an input file of 5 GB was reduced 5 ~ 6 times compared with when not using memory mapping. This way temporarily solve my problems. However, I will also consider suggestions of #utnapistim.
I'm trying to find a solution to store a binary file in it's smallest size on disk. I'm reading vehicles VIN and plate number from a database that is 30 Bytes and when I put it in a txt file and save it, its size is 30B, but its size on disk is 4KB, which means if I save 100000 files or more, it would kill storage space.
So my question is that how can I write this 30B to an individual binary file to its smallest size on disk, and what is the smallest possible size of 30B on disk including other info such as file name and permissions?
Note: I do not want to save those text in database, just I want to make separate binary files.
the smallest size of a file is always the cluster size of your disk, which is typically 4k. for data like this, having many records in a single file is really the only reasonable solution.
although another possibility would be to store those files in an archive, a zip file for example. under windows you can even access the zip contents pretty similar to ordinary files in explorer.
another creative possibility: store all the data in the filename only. a zero byte file takes only 1024 bytes in the MFT. (assuming NTFS)
edit: reading up on resident files, i found that on the newer 4k sector drives, the MFT entry is actually 4k, too. so it doesn't get smaller than this, whether the data size is 0 or not.
another edit: huge directories, with tens or hundreds of thousands of entries, will become quite unwieldy. don't try to open one in explorer, or be prepared to go drink a coffee while it loads.
Most file systems allocate disk space to files in chunks. It is not possible to take less than one chunk, except for possibly a zero-length file.
Google 'Cluster size'
You should consider using some indexed file library like gdbm: it is associating to arbitrary key some arbitrary data. You won't spend a file for each association (only a single file for all of them).
You should reconsider your opposition to "databases". Sqlite is a library giving you SQL and database abilities. And there are noSQL databases like mongodb
Of course, all this is horribly operating system and file system specific (but gdbm and sqlite should work on many systems).
AFAIU, you can configure and use both gdbm and sqlite to be able to store millions of entries of a few dozen bytes each quite efficienty.
on filesystems you have the same problem. the smallest allocate size is one data-node and also a i-node. For example in IBM JFS2 is the smallest blocksize 4k and you have a inode to allocate. The second problem is you will write many file in short time. It makes a performance problems, to write in short time many inodes.
Every write operation must jornaled and commit. Or you us a old not jornaled filesystem.
A Idear is, grep many of your data recorders put a separator between them and write 200-1000 in one file.
for example:
0102030400506070809101112131415;;0102030400506070809101112131415;;...
you can index dem with the file name. Sequence numbers or so ....
My current approach:
I have one domain class - Application
Each application in my system is stored in "applications" bucket under APPLICATION_KEY key
Apart from application metadata stored in this bucket, each application has its own bucket called "time_metrics/APPLICATION_KEY" where I store time series in a way:
KEY - timestamp / VALUE - some attributes
My concern is efficiency of queries made over specific time window for given application. Currently to get time series from some specific time window and eventually make some reductions I have to make map/reduce over whole "time_metric/APPLICATION_KEY" bucket, which what I have found is not the recommended use case for Riak Map/Reduce.
My question: what would be the best db structure for this kind of a system and how efficiently query it.
Adding onto #macintux's answer.
Basho has had a few customers that have used riak for time series metrics.
Boundary has a nice tech talk about how they use Riak with their network monitoring software. They rollup data into different chunks of time (1m, 5m, 15m) for analysis.
They also have a series of blog posts about lessons learned while implementing this system.
Kivra also has a good slide deck about how they use timeseries data with riak.
You could roll up your data into some sort of arbitrary time length, then read the range you need by issuing regular K/V gets, and then reconstruct the larger picture / reduce in your application.
If you have spare computing power and you know in advance what keys you need, you certainly can use Riak's MapReduce, but often retrieving the keys and running your processing on the client will be as fast (and won't strain your cluster).
Some general ideas:
Roll up your data into larger blocks
If you're concerned about losing data if your client crashes while buffering it, you can always store the data as it arrives
Similar idea: store the data as it arrives, then retrieve it and roll it up at certain intervals
You can automatically expire data once you're confident it is being reliably stored in larger blocks, using either the Bitcask or Memory backends
Memory backend is quite useful (RAM permitting) for any data that only needs to be stored for a limited period of time
Related: don't be afraid to store multiple copies of your data to make reading/reporting easier later
Multiple chunks of time (5- and 15-minute blocks, for example)
Multiple report formats
Having said all that, if you're doing straight key/value requests (it's ideal to always be able to compute the keys you need, rather than doing indexing or searching), Riak can support very heavy traffic loads, so I wouldn't recommend spending too much time creating alternative storage mechanisms unless you know you're going to face latency problems.
I'm working on a Qt GUI for visualizing 'live' data which is received via a TCP/IP connection. The issue is that the data is arriving rather quickly (a few dozen MB per second) - it's coming in faster than I'm able to visualize it even though I don't do any fancy visualization - I just show the data in a QTableView object.
As if that's not enough, the GUI also allows pressing a 'Freeze' button which will suspend updating the GUI (but it will keep receiving data in the background). As soon as the Freeze option was disabled, the data which has been accumulated in the background should be visualized.
What I'm wondering is: since the data is coming in so quickly, I can't possibly hold all of it in the memory. The customer might even keep the GUI running over night, so gigabytes of data will accumulate. What's a good data storage system for writing this data to disk? It should have the following properties:
It shouldn't be too much work to use it on a desktop system
It should be fast at appending new data at the end. I never need to touch previously written data anymore, so writing into anywhere but the end is not needed.
It should be possible to randomly access records in the data. This is because scrolling around in my GUI will make it necessary to quickly display the N to N+20 (or whatever the height of my table is) entries in the data stream.
The data which is coming in can be separated into records, but unfortunately the records don't have a fixed size. I'd rather not impose a maximum size on them (at least not if it's possible to get good performance without doing so).
Maybe some SQL database, or something like CouchDB? It would be great if somebody could share his experience with such scenarios.
I think that sqlite might do the trick. It seems to be fast. Unfortunately, I have no data flow like yours, but it works well as a backend for a log recorder. I have a GUI where you can view the n, n+k logs.
You can also try SOCI as a C++ database access API, it seems to work fine with sqlite (I have not used it for now but plan to).
my2c
I would recommend a simple file based solution.
If you can use fixed size records: If the you get the data continuously with constant sample rate, random access to data is easy and very fast when you know the time stamp of first data point and the sample rate. If the sample rate varies, then write time stamp with each data point. Now random access requires binary search, but it is still fast enough.
If you have variable size records: Write the variable size data to one file and to other file write indexes (which are fixed size) to the data file. And if the sample rate varies, write time stamps too. Now you can do the random access fast using the index file.
If you are using Qt to implement this kind of solution, you need two sets of QFile and QDataStream instances, one for writing and one for reading.
And a note about performance: don't flush the file after every data point write. But remember to flush the file before doing any random access to it.