In Elixir (and Erlang), we are encouraged to use processes to divide the work and have them communicate with small messages. Somehow though, I also need to handle not-so-small data which might not only be useful in a single process, and I'm unsure how. Here's my use case:
I've designed a simple card game which allows multiple players to join the same game through their browser, but also to create new ones. Basically, I'm keeping the card game in a process (so I create a new process whenever a player asks to create a new game). I would also like my processes to somehow save the card game on disk (or whatever storage is available). My first reaction was to avoid doing this in the game process itself, so it wouldn't "slow down" my game too much, since while the disk is accessed to write the game down (and serialization has to occur), messages sent by players to the game will be delayed. So I thought I'd create "save" processes whose job was simple: to handle the card game for a given game and store it on disk. These processes would be servers reacting to casts, so that the game process could just hand over the data whenever an action occurred (here: that's my card game, save it and well, I'm off). And now arises another problem: the card game has to be sent over the network (which might be a bit long, if on a different node). This might slow down process communication. In fact, it might also slow down the heartbeat of individual processes.
My games aren't that large. At current testing, they weigh about 4k. And yet, 4k of data might be a lot on a slow network. (Don't take network speed for granted.) I don't think I really have to worry, in my situation (I could actually save the game directly in the game process and save the trouble, it won't slow down my game that much) but I'm interested in solutions and I'm coming up blank.
The advantage of "save" processes was that they could live on another node: if the game process crashed, one would be recreated dynamically and ask save processes if anyone had the copy of game ID 121. If the save process crashed, the game processes could send their updated copy to another process/node. It seemed like a good way to keep things in good state. Of course, having a game process and save process crash at the same time would ruin some data, but there's so much one can do in a distributed environment (or any environment, for that matter). Plus, in this scenario, communication between the node(s) hosting the games (it can be spread on several nodes) and the node(s) saving data wouldn't have to be particularly fast, since the only communication would be one-sided and incremental (unless an error occurred, as described).
This is more a theoretical question. Elixir (or Erlang) isn't the only way to create distributed system, though the large message system and heartbeat might be different. Still, I would like to hear thoughts on ways to improve my system to handle data saving.
Thanks for your answers,
I think the main issue here is how to save large data without blocking and without causing a backlog.
Blocking can happen if the main process also does the saving, but if it hands it off to a separate process, that can cause a backlog and possible data loss in case of crash.
The best forward for this I can think of is to not save the whole state every time, but save each mutation to the game state as an individual event, and have some logic to recreate the state from individual events when trying to restore state.
To optimize this further, the "save process" can also periodically dump the whole state, so that the max number of entries to roll up on recovery is limited.
What I described here is a very basic version of how many databases write transactions first to an append-only log file, and roll it up in batches later.
Related
Vulkan is intended to be thin and explicit to user, but queues are a big exception to this rule: queues may be multiplexed by driver and it's not always obvious if using multiple queues from a family will improve performance or not.
After one of driver updates, I've got 2 transfer-only queues instead of one, but I'm pretty sure that there will be no benefit in using them in parallel for data streaming compared to just using one of them (will be happy to be proved wrong)
So why not just say "we have N separate hardware queues and if you want to use some of them in parallel, just mutex it yourself"? Now it looks like there's no way to know, how independent queues in family really are.
GPUs these days have to contend with a multi-processed world. Different programs can access the same hardware, and GPUs have to be able to deal with that. As such, having parallel input streams for a single piece of actual hardware is no different from being able to create more CPU threads than you have actual CPU cores.
That is, a queue from a family is probably not "mutexing" access to the actual hardware. At least, not in a CPU way. If multiple queues from a family are different paths to execute stuff on the same hardware, then the way that hardware gets populated from these multiple queues probably happens at the GPU level. That is, it's an actual hardware feature.
And you could never get performance equivalent to that hardware feature by "mutexing it yourself". For example:
I've got 2 transfer-only queues instead of one, but I'm pretty sure that there will be no benefit in using them in parallel for data streaming compared to just using one of them
Let's assume that there really is only one hardware DMA channel with a fixed bandwidth behind that transfer queue. This means that, at any one time, only one thing can be DMA'd from CPU memory to GPU memory at one time.
Now, let's say you have some DMA work to do. You want to upload a bunch of stuff. But every now and then, you need to download some rendering product. And that download needs to complete ASAP, because you need to reuse the image that stores those bytes.
With prioritized queues, you can give the download transfer queue much higher priority than the upload queue. If the hardware permits it, then it can interrupt the upload to perform the download, then get back to the upload.
With your way, you'd have to upload each item one at a time at regular intervals. A process that will have to be able to be interrupted by a possible download. To do that, you'd basically have to have a recurring tasks that shows up to perform and submit a single upload to the transfer queue.
It'd be much more efficient to just throw the work at the GPU and let its priority system take care of it. Even if there is no priority system, then it'll probably perform operations round-robin, jumping back and forth between the input transfer queue operations rather than waiting for one queue to run dry before trying another.
But of course, this is all hypothetical. You'd need to do profiling work to make sure that these things pan out.
The main issue with queues within families is that they sometimes represent distinct hardware with their own dedicated resources and sometimes they don't. AMD's hardware for example offers two transfer queues, but these actually use separate DMA channels. Granted, they probably still share the same overall bandwidth, but it's not a simple case of one queue having to wait to execute work until the other queue has executed a transfer command.
I have a system that runs windows via a USB stick (it's a proprietary machine). This type of machine is commonly powered off by 'pulling the plug'. There is no way around it, that is how it is operated.
We occasionally have drive corruption on the USB stick, or at least corruption in the directory that we write things into. Is there really any software solution to get around this problem other than 'write as little/infrequently as possible'?
It's a windows machine and the applications that write are typically written in Java/C# if that is useful to anyone. The corruption typically shows up as a write directory or the parent of a write directory that can no longer be accessed due to the corruption. The only way to deal with it is to delete it via command line and start over.
Is there any way to programmatically deal with such a scenario, to perhaps restore a previous state of the memory as opposed to deleting and starting anew?
I don't feel as though there is any way to prevent this type of thing from happening given our current design. If you do enough writes and keep pulling the plug you are eventually going to get a corruption and that's just facts. Especially in this design. Even if the backup batteries are charged, if the software doesn't shutdown gracefully within the battery's discharge time, the corruptions could still occur. Not to mention as gravitymixes said above its going to damage hardware eventually which we have seen before.
A system redesign needs to considered for this project as a whole. Some type of networked solution comes to mind immediately where data is sent off the volatile machine to be logged on a machine with a more reliable power source over a reliable network connection with writing to the disk on the actual volatile machine as a last ditch effort if network comms are not reliable at a given point in time (backfill). I feel like this would increase hardware life as well. Of course the problem of network reliability then becomes your problem.
I realize this number will change based on many factors, but in general, when I write data to a hard-drive (e.g. copy a file), how long does it take for that data to actually be written to the platter after Windows says the copy is done?
Could anyone point me in the right direction to discover more on this topic?
If you are looking for a hard number, that is pretty much unknowable. Generally it is the order of a tens to a few hundred milliseconds for the data to start reaching the disk platters, but can be as high as several seconds in a large server disk array with RAID and de-duplication.
The flow of events goes something like this.
The application calls a function like fwrite().
This call is handled by the filesystem layer in your Operating System, which has to figure out what specific disk sectors are to be manipulated.
The SATA/IDE driver in your OS will talk to the hard drive controller hardware. On a modern PC, it typically uses DMA to feed the data to the disk.
The data sits in a write cache inside the hard disk (RAM).
When the physical platters and heads have made it into position, it will begin to transfer the contents of cache onto the platters.
Steps 3-6 may repeat several times depending on how much data is to be written, where on the disk it is to be written. Additionally, there is usually filesystem metadata that must be updated (e.g. free space counters), which will trigger more writes to the disk.
The time it takes from steps 1-3 can be unpredictable in a general purpose OS like Windows due to task scheduling, background threads, and your disk write is probably queued up with a few dozen other processes. I'd say it is usually on the order of 10-100msec on a typical PC. If you go to the Windows Resource Monitor and click the Disk tab, you can get an idea of the average disk queue length. You can use the Performance Monitor to produce more finely-controlled graphs.
Steps 3-4 are largely controlled by the disk controller and disk interface (SATA, SAS, etc). In the server world, you can be talking about a SAN with FC or iSCSI network switches, which impose their own latencies.
Step 5 will be controlled by they physical performance of the disk. Many consumer-grade HDD manufacturers do not post average seek times anymore, but 10-20msec is common.
Interesting detail about Step 5: Some HDDs lie about flushing their write cache to get better benchmark scores.
Step 6 will depend on your filesystem and how much data you are writing.
You are right that there can be a delay between Windows indicating that data writing is finished and the last data actually written. Things to consider are:
Device Manager, Disk Drive, Properties, Policies - Options for disabling Write Caching.
You might be better off using Direct I/O so that Windows does not save it temporarily in File Cache.
If your program writes the data, you can log what has been copied.
If you are sending the data over a network, you are likely to have no control of when the remote system has finished.
To see what is happening, you can set up Perfmon logging. One of my examples of monitoring:
http://www.roylongbottom.org.uk/monitor1.htm#anchor2
I have a long running (5-10 hours) Mac app that processes 5000 items. Each item is processed by performing a number of transforms (using Saxon), running a bunch of scripts (in Python and Racket), collecting data, and serializing it as a set of XML files, a SQLite database, and a CoreData database. Each item is completely independent from every other item.
In summary, it does a lot, takes a long time, and appears to be highly parallelizable.
After loading up all the items that need processing it, the app uses GCD to parallelize the work, using dispatch_apply:
dispatch_apply(numberOfItems, dispatch_get_global_queue(DISPATCH_QUEUE_PRIORITY_HIGH, 0), ^(size_t i) {
#autoreleasepool {
...
}
});
I'm running the app on a Mac Pro with 12 cores (24 virtual). So I would expect to have 24 items being processed at all times. However, I found through logging that the number of items being processed varies between 8 and 24. This is literally adding hours to the run time (assuming it could work on 24 items at a time).
On the one hand, perhaps GCD is really, really smart and it is already giving me the maximum throughput. But I'm worried that, because much of the work happens in scripts that are spawned by this app, maybe GCD is reasoning from incomplete information and isn't making the best decisions.
Any ideas how to improve performance? After correctness, the number one desired attribute is shortening how long it takes this app to run. I don't care about power consumption, hogging the Mac Pro, or anything else.
UPDATE: In fact, this looks alarming in the docs: "The actual number of tasks executed by a concurrent queue at any given moment is variable and can change dynamically as conditions in your application change. Many factors affect the number of tasks executed by the concurrent queues, including the number of available cores, the amount of work being done by other processes, and the number and priority of tasks in other serial dispatch queues." (emphasis added) It looks like having other processes doing work will adversely affect scheduling in the app.
It'd be nice to be able to just say "run these blocks concurrently, one per core, don't try to do anything smarter".
If you are bound and determined, you can explicitly spawn 24 threads using the NSThread API, and have each of those threads pull from a synchronized queue of work items. I would bet money that performance would get noticeably worse.
GCD works at its most efficient when the work items submitted to it never block. That said, the workload you're describing is rather complex and rife with opportunities for your threads to block. For starters, you're spawning a bunch of other processes. Right here, this means that you're already relying on the OS to divvy up time/resources between your master task and these slave tasks. Other than setting the OS priority of each subprocess, the OS scheduler has no way to know which processes are more important than others, and by default, your subprocesses are going to have the same priority as their parent. That said, it doesn't sound like you have anything to gain by tweaking process priorities. I'm assuming you're blocking the master task thread that's waiting for the slave tasks to complete. That is effectively parking that thread -- it can do no useful work. But like I said, I don't think there's much to be gained by tweaking the OS priorities of your slave tasks, because this really sounds like it's an I/O bound workflow...
You go on to describe three I/O-heavy operations ("serializing it as a set of XML files, a SQLite database, and a CoreData database.") So now you have all these different threads and processes vying for what is presumably a shared bulk storage device. (i.e. unless you're writing to 24 different databases, on 24 separate hard drives, one for each core, your process is ultimately going to be serialized at the disk accesses.) Even if you had 24 different hard drives, writing to a hard drive (even an SSD) is comparatively slow. Your threads are going to be taken off of the CPU they were running on (so that another thread that's waiting can run) for virtually any blocking disk write.
If you wanted to maximize the performance you're getting out of GCD, you would probably want to rewrite all the stuff you're doing in subtasks in C/C++/Objective-C, bringing them in-process, and then conducting all the associated I/O using dispatch_io primitives. For API where you don't control the low-level reads and writes, you would want to carefully manage and tune your workload to optimize it for the hardware you have. For instance, if you have a bunch of stuff to write to a single, shared SQLite database, there's no point in ever having more than one thread trying to write to that database at once. You'd be better off making one thread (or a serial GCD queue) to write to SQLite and submitting tasks to that after pre-processing is done.
I could go on for quite a while here, but the bottom line is that you've got a complex, seemingly I/O bound workflow here. At the highest-level, CPU utilization or "number of running threads" is going to be a particularly poor measure of performance for such a task. By using sub-processes (i.e. scripts), you're putting a lot of control into the hands of the OS, which knows effectively nothing about your workload a priori, and therefore can do nothing except use its general scheduler to divvy up resources. GCD's opaque thread pool management is really the least of your problems.
On a practical level, if you want to speed things up, go buy multiple, faster (i.e. SSD) hard drives, and rework your task/workflow to utilize them separately and in parallel. I suspect that would yield the biggest bang for your buck (for some equivalence relation of time == money == hardware.)
Is their a Grand Unified Theory of logging? Shall we develop one? Question (just to show this is not a discussion :), how can I improve on the following? (note that I live mainly in the embedded world, but non-embedded suggestions are also welcome)
How do you log, when do you log, what do you log, what do you do with log files?
How do you log - I generally have macros, #ifdef TESTING, sort of thing. They write to RAM and a low priority process writes them out when the system is idle (using UDP, since I do embedded systems)
When do you log - same as voting, early and often. At every (in)significant program event, I log at varying levels. Events received, transaction succeed/fail, data updated, etc
What do you log - Fatal/Error/Warning/Info/Debug/Trace is covered in When to use the different log levels?
What do you do with log files - 1) keep them (in CVS), both pass and fail 2) capture everything and filter later in case I can't repeat a problem. I have tools to filter the log by "level" (Fatal/Error/etc), process, file, etc. And to draw message sequence charts, dump data structures, draw histograms of memory usage - what am I missing?
Hmmm, binary or ascii log file format? Ascii is bulkier, but binary requires more processing. I have done both, currently I use ascii
Question - did I miss anything, and how can I improve on this?
You could "instrument" your code in many different ways, everything from start-up/shut-down events to individual machine instruction execution (using a processor emulator). Of all the possibilities, what's worth doing? Don't just do it for the sake of completeness; have a specific goal in mind. A business case if you like, with a benefit you expect to receive. E.g.:
Insight into CPU task execution times/patterns to enable optimisation (if you need to improve performance).
Insight into other systems to resolve system integration issues (e.g. what messages is your VoIP box sending and receiving when it connects to a particular peer?)
Insight into the nature of errors (for field diagnostics)
Aid in development
Aid in validation testing
I imagine that there's no grand unified theory of logging, because what you do would depend on many details:
Quantity of data
Type of data
Events
Streamed audio/video
Available storage
Storage speed
Storage capacity
Available channels to extract data
Bandwidth
Cost
Availability
Internet connected 24×7
Site visit required
Need to unlock a rusty gate, climb a ladder onto a roof, to plug in a cable, after filling out OHS documentation
Need to wait until the Antarctic winter is over and the ice sheets thaw
Random access vs linear access (e.g. if you compress it, do you need to read from the start to decompress and access some random point?)
Need to survive error conditions
Watchdog reboots
Possible data corruption
Due to failing power supply
Due to unreliable storage media
Need to survive a plane crash
As for ASCII vs binary, I usually prefer to keep the logging simple, and put any nice presentation in a PC application that decodes the data. It's usually easier to create a user-friendly presentation in PC software (written in e.g. Python) rather than in the embedded system itself.
did I miss anything, and how can I
improve on this?
Asynchronous logging.
Using multiple log files for the same process for different logging abstractions. e.g. the process' activities are logged in a normal log file. And the process' stats (periodic statistics that you might be interested in) are logged in a separate stats log file.
Hmmm, binary or ascii log file format?
Ascii is bulkier, but binary requires
more processing. I have done both,
currently I use ascii
ASCII is good. More often than not, logs are meant to be used for debugging purposes. A human readable form eases and speeds this up.
However, if your logs are used mostly to record information which is used later on for analysis and generation of reports (e.g. stats or latencies etc.) a binary format would be preferred. You can go one step ahead and use a custom format along with a db service which does index based sorting, where the index can be a tuple of time with the event type.
--
One thing which may be helpful is to have a "maybeLogger" object which will accept log records for an operation which may or may not succeed, and then either ditch those records if the operation succeeds or fails in an uninteresting way, or log them if it does something interesting. This is relatively easy to do in something like .net. In an embedded system, it can only be done really easily if the amount of stuff to be logged is small enough to fit in free RAM, but one could probably use a garbage-collection-based approach to hold stuff in flash (have one 'stream' of data in flash for new log entries, and another for ones that are confirmed to be interesting; periodically move data which is known to be good from the first stream to the second).
Here's my $0.02.
I only log when I'm having a problem and need to track down the source. Usually this has to do with a customer's environment, so I can't just attach the debugger. My solution is to enable the Telnet port and use that to print out statements as to where the program is and values of variables.
I do ASCII only because it's over telnet.
Another aspect of telnet is that it is pretty simple. It's a TCP port with text being thrown out. Very little processing other than the normal TCP headaches.
The log files are dumped as soon as I get them because I have not tried to capture and save a telnet session. I guess I could with WireShark, but I don't need a history of that session. I just need to find the problem and verify a fix.