Problem Statement:
I have a very high bandwidth data link that is UDP based. The source of this data is not configurable, and sends on UDP a stream of datagrams. We have code that uses the standard methods for receiving data on the UDP socket that works adequately. I wanted to know if
Does there exist a command interface to extract multiple UDP datagrams at a time? to improve efficiency?
If one doesn't exist, does it make sense to create a kernel module to provide the capability?
I am a novice, and i wanted to understand what thought process has to happen when writing your own kernel module seems appropriate. I know that such a surgical procedure isn't meant to done lightly, but there must be a set of criteria where that action is prudent. Maybe not in my case, but in general.
HW / Kernel Module Perspective
A typical network adapter these days would be capable of distributing received packets across multiple hardware Rx queues thus letting the host run multiple software Rx queues bound to different CPU cores reading out packets in parallel. From a single HW/SW queue perspective, the host may poll it for new packets (see Linux NAPI), with each poll ideally yielding a batch of packets, and, alternatively, the host may still use interrupt-driven approach for Rx signalling with interrupt coalescing turned on for improved efficiency.
Existing NIC drivers in Linux kernel strive to stick with the most performant techniques, and the kernel itself should be able to leverage all of that properly.
Userland / Application Perspective
There's PACKET_MMAP interface provided by Linux kernel for improved Rx/Tx efficiency on the application side. Long story short, an application can set up a memory buffer shared between the kernel- and userspace and read out incoming packets from it, ideally in batches, or blocks, thus avoiding costly kernel-to-userspace copies and context switches so customary when using regular methods.
For added efficiency, the application may have multiple sockets bound to the NIC in separate threads / processes and demand that packet reception be load balanced across these sockets (see AF_PACKET fanout mode description).
DPDK Perspective
Kernel bypass framework that allows an application to seize full control of a network adapter by means of a vendor-specific poll-mode driver, or PMD, effectively running in userspace as part of the application and by its very nature not needing any kernel-to-userspace copies, context switches and, most likely, locking. Multi-queue receive operation, load balancing (round robin, RSS, you name it) and more cutting edge offloads are likely to be available, too (it's vendor specific).
Summary
The short of it, given the fact that multiple network acceleration techniques already exist, one need never write their own kernel module to solve the problem in question. By the looks of it, your application, which, as you say, uses standard methods, is not aware of PACKET_MMAP technique. So I'd be tempted to suggest looking at this one closely. DPDK approach might require that the application be effectively re-implemented from scratch, so I would first go for PACKET_MMAP approach as a low-hanging fruit.
Related
You know, when an application opens a file and write to it, the system chooses in which cluster will be stored. I want to choose myself ! Let me tell you what I really want to do... In fact, I don't necessarily want to write anything. I have a HDD with a BAD range of clusters in the middle and I want to mark that space as it is occupied by a file, and eventually set it as a hidden-unmoveable-system one (like page file in windows) so that it won't be accessed anymore. Any ideas on how to do that ?
Later Edit:
I think THIS is my last hope. I just found it, but I need to investigate... Maybe a file could be created anywhere and then relocated to the desired cluster. But that requires writing, and the function may fail if that cluster is bad.
I believe the answer to your specific question: "Can I write a file to a specific cluster location" is, in general, "No".
The reason for that is that the architecture of modern operating systems is layered so that the underlying disk store is accessed at a lower level than you can access, and of course disks can be formatted in different ways so there will be different kernel mode drivers that support different formats. Even so, an intelligent disk controller can remap the addresses used by the kernel mode driver anyway. In short there are too many levels of possible redirection for you to be sure that your intervention is happening at the correct level.
If you are talking about Windows - which you haven't stated but which appears to assumed - then you need to be looking at storage drivers in the kernel (see https://learn.microsoft.com/en-us/windows-hardware/drivers/storage/). I think the closest you could reasonably come would be to write your own Installable File System driver (see https://learn.microsoft.com/en-us/windows-hardware/drivers/ddi/_ifsk/). This is really a 'filter' as it sits in the IO request chain and can intercept and change IO Request Packets (IRPs). Of course this would run in the kernel, not in userspace, and normally this would be written in C and I note your question is tagged for Delphi.
Your IFS Driver can sit at differnt levels in the request chain. I have used this technique to intercept calls to specific file system locations (paths / file names) and alter the IRP so as to virtualise the request - even calling back to user space from the kernel to resolve how the request should be handled. Using the provided examples implementing basic functionality with an IFS driver is not too involved because it's a filter and not a complete storgae system.
However the very nature of this approach means that another filter can also alter what you are doing in your driver.
You could look at replacing the file system driver that interfaces to the hardware, but I think that's likely to be an excessive task under the circumstances ... and as pointed out already by #fpiette the disk controller hardware can remap your request anyway.
In the days of MSDOS the access to the hardware was simpler and provided by the BIOS which could be hooked to allow the requests to be intercepted. Modern environments aren't that simple anymore. The IFS approach does allow IO to be hooked, but it does not provide the level of control you need.
EDIT regarding suggestion by the OP of using FSCTL_MOVE_FILE
For simple environment this may well do what you want, it is designed to support a defragmentation process.
However I still think there's no guarantee that this actually will do what you want.
You will note from the page you have linked to it states that it is moving one or more virtual clusters of a file from one logical cluster to another within the same volume
This is a code that's passed to the underlying storage drivers which I have referred to above. What the storage layer does is up to the storage layer and will depend on the underlying technology. With more advanced storage there's no guarantee this actually addresses the physical locations which I believe your question is asking about.
However that's entirely dependent on the underlying storage system. For some types of storage relocation by the OS may not be honoured in the same way. As an example consider an enterprise storage array that has a built in data-tiering function. Without the awareness of the OS data will be relocated within the storage based on the tiering algorithms. Also consider that there are technologies which allow data to be directly accessed (like NVMe) and that you are working with 'virtual' and 'logical' clusters, not physical locations.
However, you may well find that in a simple case, with support in the underlying drivers and no remapping done outside the OS and kernel, this does what you need.
Since you problem is to mark bad cluster, you don't need to write any program. Use the command line utility CHKDSK that Windows provides.
I an elevated command prompt (Run as administrator), run the command:
chkdsk /r c:
The check will be done on the next reboot.
Don't forget to read the documentation.
I have a distributed program which communicates with ZeroMQ that runs on HPC clusters.
ZeroMQ uses TCP sockets, so by default on HPC clusters the communications will use the admin network, so I have introduced an environment variable read by my code to force communication on a particular network interface.
With Infiniband (IB), usually it is ib0. But there are cases where another IB interface is used for the parallel file system, or on Cray systems the interface is ipogif, on some non-HPC systems it can be eth1, eno1, p4p2, em2, enp96s0f0, or whatever...
The problem is that I need to ask the administrator of the cluster the name of the network interface to use, while codes using MPI don't need to because MPI "knows" which network to use.
What is the most portable way to discover the name of the high-performance network interface on a linux HPC cluster? (I don't mind writing a small MPI program for this if there is no simple way)
There is no simple way and I doubt a complete solution exists. For example, Open MPI comes with an extensive set of ranked network communication modules and tries to instantiate all of them, selecting in the end the one that has the highest rank. The idea is that ranks somehow reflect the speed of the underlying network and that if a given network type is not present, its module will fail to instantiate, so faced with a system that has both Ethernet and InfiniBand, it will pick InfiniBand as its module has higher precedence. This is why larger Open MPI jobs start relatively slowly and is definitely not fool proof - in some cases one has to intervene and manually select the right modules, especially if the node has several network interfaces of InfiniBand HCAs and not all of them provide node-to-node connectivity. This is usually configured system-wide by the system administrator or the vendor and is why MPI "just works" (pro tip: in not-so-small number of cases it actually doesn't).
You may copy the approach taken by Open MPI and develop a set of detection modules for your program. For TCP, spawn two or more copies on different nodes, list their active network interfaces and the corresponding IP addresses, match the network addresses and bind on all interfaces on one node, then try to connect to it from the other node(s). Upon successful connection, run something like the TCP version of NetPIPE to measure the network speed and latency and pick the fastest network. Once you've gotten this information from the initial small set of nodes, it is very likely that the same interface is used on all other nodes too, since most HPC systems are as homogeneous as possible when it comes to their nodes' network configuration.
If there is a working MPI implementation installed, you can use it to launch the test program. You may also enable debug logging in the MPI library and parse the output, but this will require that the target system has an MPI implementation supported by your log parser. Also, most MPI libraries use native InfiniBand or whatever high-speed network API there is and will not tell you which is the IP-over-whatever interface, because they won't use it at all (unless configured otherwise by the system administrator).
Q : What is the most portable way to discover the name of the high-performance network interface on a linux HPC cluster?
This seems to be in a gray-zone - trying to solve a multi-faceted problem among site-specific hardware (technical) interface naming and theirs non-technical, weakly administratively maintained, preferred ways of use.
As-is State :
ZeroMQ can (as per RFC 37/ZMTP v3.0+) specify <hardware(interface)>:<port>/<service> details :
zmq_bind (server_socket, "tcp://eth0:6000/system/name-service/test");
And:
zmq_connect (client_socket, "tcp://192.168.55.212:6000/system/name-service/test");
yet has no means, to my knowledge, to reverse-engineer the primary use of such an interface, in the holistic context of the HPC-site and it's hardware configuration.
Seems to me, your idea of pre-testing the administrative mappings via MPI-tool first and letting ZeroMQ deployment use these externally detected (if indeed auto-detectable, as you assumed above) configuration details for a proper (preferred) interface usage.
The Safe Way to Go :
Asking the HPC-infrastructure Support Team ( who is responsible for knowing all of the above and trained to help Scientific Teams to use the HPC in the most productive manner ) would be my preferred way to go.
Disclaimer :
Sorry in case this did not help your will to read & auto-detect all the needed configuration details ( a universal BlackBox-HPC-ecosystem detection and auto-configuration strategy would hardly be a trivial one-liner, I guess, wouldn't it? )
How does one actually use kqueue() for doing simple async r/w's?
It's inception seems to be as a replacement for epoll(), and select(), and thus the problem it is trying to solve is scaling to listening on large number of file descriptors for changes.
However, if I want to do something like: read data from descriptor X, let me know when the data is ready - how does the API support that? Unless there is a complimentary API for kicking-off non-blocking r/w requests, I don't see a way other than managing a thread pool myself, which defeats the purpose.
Is this simply the wrong tool for the job? Stick with aio?
Aside: I'm not savvy with how modern BSD-based OS internals work - but is kqueue() built on aio or visa-versa? I would imagine it would depend on whether the OS io subsystem system is fundamentally interrupt-driven or polling.
None of the APIs you mention, aside from aio itself, has anything to do with asynchronous IO, as such.
None of select(), poll(), epoll(), or kqueue() are helpful for reading from file systems (or "vnodes"). File descriptors for file system items are always "ready", even if the file system is network-mounted and there is network latency such that a read would actually block for a significant time. Your only choice there to avoid blocking is aio or, on a platform with GCD, dispatch IO.
The use of kqueue() and the like is for other kinds of file descriptors such as sockets, pipes, etc. where the kernel maintains buffers and there's some "event" (like the arrival of a packet or a write to a pipe) that changes when data is available. Of course, kqueue() can also monitor a variety of other input sources, like Mach ports, processes, etc.
(You can use kqueue() for reads of vnodes, but then it only tells you when the file position is not at the end of the file. So, you might use it to be informed when a file has been extended or truncated. It doesn't mean that a read would not block.)
I don't think either kqueue() or aio is built on the other. Why would you think they were?
I used kqueues to adapt a Linux proxy server (based on epoll) to BSD. I set up separate GCD async queues, each using a kqueue to listen on a set of sockets. GCD manages the threads for you.
Good evening.
I'm looking for a method to share data from my application system-wide, so that other applications could read that data and then do whatever they want with it (e.g. format it for display, use it for logging, etc). The data needs to be updated dynamically in the method itself.
WMI came to mind first, but then you've got the issue of applications pausing while reading from WMI. Additionally, i've no real idea how to setup my own namespace or classes if that's even possible in Delphi.
Using files is another idea, but that could get disk heavy, and it's a real awful method to use for realtime data.
Using a driver would probably be the best option, but that's a little too intrusive on the users end for my liking, and i've no idea on where to even start with it.
WM_COPYDATA would be great, but i'm not sure if that's dynamic enough, and whether it'll be heavy on resources or not.
Using TCP/IP would be the best choice for over the network, but obviously is of little use when run on a single system with no networking requirement.
As you can see, i'm struggling to figure out where to go with this. I don't want to go into one method only to find that it's not gonna work out in the end. Essentially, something like a service, or background process, to record data and then allow other applications to read that data. I'm just unsure on methods. I'd prefer to NOT need elevation/UAC to do this, but if needs be, i'll settle for it.
I'm running in Delphi 2010 for this exercise.
Any ideas?
You want to create some Client-Server architecture, which is also called IPC.
Using WM_COPYDATA is a very good idea. I found out it is very fast, lightweight, and efficient on a local machine. And it can be broadcasted over the system, to all applications at once (to be used with care if some application does not handle it correctly).
You can also share some memory, using memory mapped files. This is may be the fastest IPC option around for huge amount of data, but synchronization is a bit complex (if you want to share more than one buffer at once).
Named pipes are a good candidates for local. They tend to be difficult to implement/configure over a network, due to security issues on modern Windows versions (and are using TCP/IP for network communication - so you should better use directly TCP/IP instead).
My personal advice is that you shall implement your data sharing with abstract classes, able to provide several implementations. You may use WM_COPYDATA first, then switch to named pipes, TCP/IP or HTTP in order to spread your application over a network.
For our Open Source Client-Server ORM, we implemented several protocols, including WM_COPY_DATA, named pipe, HTTP, or direct in-process access. You can take a look at the source code provided for implementation patterns. Here are some benchmarks, to give you data from real implementations:
Client server access:
- Http client keep alive: 3001 assertions passed
first in 7.87ms, done in 153.37ms i.e. 6520/s, average 153us
- Http client multi connect: 3001 assertions passed
first in 151us, done in 305.98ms i.e. 3268/s, average 305us
- Named pipe access: 3003 assertions passed
first in 78.67ms, done in 187.15ms i.e. 5343/s, average 187us
- Local window messages: 3002 assertions passed
first in 148us, done in 112.90ms i.e. 8857/s, average 112us
- Direct in process access: 3001 assertions passed
first in 44us, done in 41.69ms i.e. 23981/s, average 41us
Total failed: 0 / 15014 - Client server access PASSED
As you can see, fastest is direct access, then WM_COPY_DATA, then named pipes, then HTTP (i.e. TCP/IP). Message was around 5 KB of JSON data containing 113 rows, retrieved from server, then parsed on the client 100 times (yes, our framework is fast :) ). For huge blocks of data (like 4 MB), WM_COPY_DATA is slower than named pipes or HTTP-TCP/IP.
Where are several IPC (inter-process communication) methods in Windows. Your question is rather general, I can suggest memory-mapped files to store your shared data and message broadcasting via PostMessage to inform other application that the shared data changed.
If you don't mind running another process, you could use one of the NoSQL databases.
I'm pretty sure that a lot of them won't have Delphi drivers, but some of them have REST drivers and hence can be driven from pretty much anything.
Memcached is an easy way to share data between applications. Memcached is an in-memory key-value store for small chunks of arbitrary data (strings, objects).
A Delphi 2010 client for Memcached can be found on google code:
http://code.google.com/p/delphimemcache/
related question:
Are there any Caching Frameworks for Delphi?
Googling for 'delphi interprocess communication' will give you lots of pointers.
I suggest you take a look at http://madshi.net/, especially MadCodeHook (http://help.madshi.net/madCodeHook.htm)
I have good experience with the product.
We have an application server developed with Delphi 2010 and Indy 10. This server receives more than 50 requests per second and it works well. But in some cases, it seems to me that Indy is very obscure. Their components are good, but sometimes I found myself digging into the source code only to understand a simple thing. Indy lacks on good documentation and good support.
The last thing that i came across was a big problem for me: I must detect when a client disconnects non gracefully (When the the client crashes or shutdown, for instance. Not telling the server that it will disconnect) and indy was not able to do that. If I want that, I will have to develop a algorithm like heartbeat, pooling or TCP keep-alive. I do not want to spend more time doing a, at least I think, component job. After a few study, I found out that this is not Indy's fault, but this is an issue of all blocking sockets components.
Now I am really thinking of changing the core of the Server to another good suite. I must admit I am tending to use a non-blocking socket. Based on that, I have some questions:
What do a benefit from changing from blocking to non-blocking sockets?
Will I be able to detect client disconnects (non gracefully)?
What component suite has the best product? By best product I mean: fast, good support, good tools and easy to implement.
I know this must be a subjective question, but I really want to hear that from you. My first question is the one I care most. I do not care if I have to pay 100, 500, 1000, 10000 dollars, but I want a complete solution. For now, I am thinking about Ip*works .
EDIT
I think some guys are not understand what I want. I don't want to create my own socket. I have been working with sockets for a long time and I am getting tired of it. Really.
And non-blocking sockets CAN detect client disconnects. That is a fact and it has good documentation all over the internet. A non-blocking socket checks the socket state for new incoming data all the time, and it makes possible to detect that the socket is not valid. This is not a heartbeat algorithm. A heartbeat algorithm is used on client side and it sends periodically packets (aka keep-alive) to the server to tells it is still alive.
EDIT
I am not make myself clear. Maybe because English is not my main language. I am not saying that it is possible to detect a dropped connection without trying to send or receiving data from a socket. What I am saying is that every non-blocking socket is able to do that because they constantly tries to read from the socket for new incoming data. Why is that so hard to understand? If you guys download and run ip*works demos, in special, the echoserver and echoclient ones (both use TCP) you can test by yourselves. I already tested it, and it works like I expected to do. Even if you use the old TCPSocketServer and TCPSocketClient in a non-blocking mode you will see what I meant.
"What do a benefit from changing from blocking to non-blocking sockets? Will I be able to detect client disconnects (non gracefully)?"
Just my two cents to get the ball rolling on this question - I'm not a socket EXPERT, but I do have a good deal of experience with them. If I'm mistaken, I'm sure someone will correct me... :-)
I assume that since you're running a server using blocking sockets with 50 connections per second, you have a threading mechanism in place to handle client requests. If so, you don't really stand to gain anything from non-blocking sockets. On the contrary - you will have to change your server logic to be event driven- based on events fired in your main thread from the non-blocking sockets, or use constant polling to know what your sockets are up to.
Non-blocking sockets can't detect clients disconnecting without notification any more than blocking sockets can - they don't have telepathic powers... The nature of the TCP/IP 'conversation' between client and server is the same - blocking and non-blocking is only with respect to your application's interaction with the socket connection conducting the 'conversation'.
If you need to purge dead connections, you need to implement a heartbeat or timeout mechanism on your socket (I've never seen a modern socket implementation that didn't support timeouts).
What do a benefit from changing from blocking to non-blocking sockets?
Increased speed, availability, and throughput (from my experience). I had an IndySockets client that was getting about 15 requests per second and when I went directly to asynchronous sockets the throughput increased to about 90 requests per second (on the same machine). In a separate benchmark test on a server at a data-center with a 30 Mbit connection I was able to get more than 300 requests per second.
Will I be able to detect client disconnects (non gracefully)?
That's one thing I haven't had to try yet, since all of my code has been on the client side.
What component suite has the best product? By best product I mean: fast, good support, good tools and easy to implement.
You can build your own socket client in a couple of days and it can be very robust and fast... much faster than most of the stuff I've seen "off the shelf". Feel free to take a look at my asynchronous socket client: http://codesprout.blogspot.com/2011/04/asynchronous-http-client.html
Update:
(Per Mikey's comments)
I'm asking you for a generic, technical explanation of how NBS increase throughput as opposed to a properly designed BS server.
Let's take a high load server as an example: say your server is supposed to handle 1000 connections at any given time, with blocking sockets you would have to create 1000 threads and even if they're mostly idle, the CPU will still spend a lot of time context switching. As the number of clients increases you will have to increase the number of threads in order to keep up and the CPU will inevitably increase the context switching. For every connection you establish with a blocking socket, you will incur the overhead of spawning of a new thread and you eventually you will incur the overhead of cleaning up after the thread. Of course, the first thing that comes to mind is: why not use the ThreadPool, you can reuse the threads and reduce the overhead of creating/cleaning-up of threads.
Here is how this is handled on Windows (hence the .NET connection): sure you could, but the first thing you'll notice with the .NET ThreadPool is that it has two types of threads and it's not a coincidence: user threads and I/O completion port threads. Asynchronous sockets use the IO completion ports which "allows a single thread to perform simultaneous I/O operations on different handles, or even simultaneous read and write operations on the same handle."(1) The I/O completion port threads are specifically designed to handle I/O in a much more efficient way than you would ever be able to achieve if you used the user threads in ThreadPool, unless you wrote your own kernel-mode driver.
"The completion port uses some special voodoo to make sure only a specific number of threads can run at once — if one thread blocks in kernel-mode, it will automatically start up another one."(2)
There are other advantages also: "in addition to the nonblocking advantage of the overlapped socket I/O, the other advantage is better performance because you save a buffer copy between the TCP stack buffer and the user buffer for each I/O call." (3)
I am using Indy and Synapse TCP libraries with good results for some years now, and did not find any showstoppers in them. I use the libraries in threads - client and server side, stability and performance was not a problem. (Six thousand request and response messages per second and more with the server running on the same system are typical.)
Blocking sockets are very useful if the protocol is more advanced than a simple 'send a string / receive a string'. Non-blocking sockets cause a higher coupling of message protocol handlers with the socket read / write logic, so I quickly moved away from non-blocking code.
No library can overcome the limitations of the TCP/IP protocol regarding detection of connection loss. Only trying to read or send data can tell wether the connection is still present.
In Windows, there is a third option which is overlapped I/O. Non-blocking sockets are essential a model using Windows messages developed to avoid single-threaded GUI apps to become "blocked" while waiting for data. A modern application IMHO would be better designed using threads and overlapped I/O.
See for example http://support.microsoft.com/kb/181611
Aahhrrgghh - the myth of being able to always detect "dropped" connections. If you pull the power on a machine with a client connection then the server cannot tell, without sending data, that the connection is "dead". The is through the design of the TCP protocol. Don't take my word for it - read this article (Detection of Half-Open (Dropped) TCP/IP Socket Connections).
This article explains the main differences between blocking and non-blocking:
Introduction to Indy, by Chad Z. Hower
Pros of Blocking
Easy to program - Blocking is very easy to program. All user code can
exist in one place, and in a
sequential order.
Easy to port to Unix - Since Unix uses blocking sockets, portable code
can be written easily. Indy uses this
fact to achieve its single source
solution.
Work well in threads - Since blocking sockets are sequential they
are inherently encapsulated and
therefore very easily used in threads.
Cons of Blocking
User Interface "Freeze" with clients - Blocking socket calls do not
return until they have accomplished
their task. When such calls are made
in the main thread of an application,
the application cannot process the
user interface messages. This causes
the User Interface to "freeze" because
the update, repaint and other messages
cannot be processed until the blocking
socket calls return control to the
applications message processing loop.
He also wrote:
Blocking is NOT Evil
Blocking sockets have been repeatedly
attacked with out warrant. Contrary to
popular belief, blocking sockets are
not evil.
It is not is an issue of all blocking sockets components that they are unable to detect a client disconnect. There is no technical advantage on the side of non-blocking components in this area.