Torrent performance testing?
Long story cut short: I wanna test the performance of several "hybrid" clients in a closed environment, both on Linux and Windows. We all know there are several clients that work on both OS'es. We also know there are several "flavors" of OS but we already decided on Server 2008 vs. RedHat 5.
Anyhow, bandwidth is not the issue since it's a closed environment. Also, there is a very basic tracker online as well that will server the swarm. The server will be getting a 100Mbit uplink within the environment and the "clients" will all get a Gbit downstream.
But the main focus is: how can I simulate 10, 50, 100 or maybe even 250 connections to the server with only 5 or 10 clients? Is there some kind of proven load-testing for torrents (served by that one server)? Or does anyone, with decent tech-knowledge, have a decent idea where to start from?
Please: no discussions or flaming on which OS or which client to use!
@Mods: if this topic is in the wrong place, please move it around...:happy:
Re: Torrent performance testing?
Need more information.
A "hybrid" client? What do you mean by that?
A closed environment, you mean testing with your own lan?
What exactly is your goal here? Are you trying to see how to spread the file the fastest with 100mbps upload? Are you trying to gauge CPU/RAM/IO usage of the client or server? Are you trying to develop better swarming techniques?
The bittorrent protocol isn't very resource heavy, nearly all modern bittorrent clients can max out a 100mbps port on even the slowest cpu's and a 7200rpm hard disk. To give you an example, the basic OVH kimsufi package (1.2ghz celery cpu, 2gb ram, 250gb hard disk) is the platform for most seedboxes you'll see on this site.
The speed limit is whatever the slowest common denominator is. Most of the time it's the port/connection speed of your internet. On connections at 1gbps (uncommon) the limit if usually the speed of the hard disk, assuming it's a conventional magnetic one.
If you want my opinion, uTorrent is the fastest out of all the clients. Why? It gives you the most control. You'll only notice the difference on connections beyond 100mbps though.
I'd of course start with the Bittorrent SDK. It is designed precisely for the testing conditions you describe.
Quote:
The BitTorrent SDK can be used to power a wide variety of content delivery scenarios off the PC.
http://www.bittorrent.com/devices/sdk-for-devices
If it's a throughput scenario your testing, you should remember that torrenting isn't much unlike lots of other every day traffic on a network. There's a gambit of network benchmarking, testing, and burn-in software out there designed to answer what-if scenarios on usage.
Re: Torrent performance testing?
More information...
The hybrid client is a client that is/was developed to run on both OS and maybe with a layer to emulate between it. For uTorrent, that would be Wine on Linux but for instance Azureus was made for Windows and Linux in native format.
The goal is to see what client can distribute the fastest/best without straining the underlying hardware-layer too hard. As you point out, a lot of seedboxes are basic hardware with magnetic drives; I want to determine at what point one OS is better over the other in terms of HDD-I/O, busspeeds, CPU-load and so on for several clients (although clients will not be tested at the same time but in seperate batches ofcourse).
One of the other goals to reach is to check if the breakevenpoint for one OS justified the fact to stay with that OS and develop, for instance, scripts for that client of switch to another OS and maybe another client to get some scripting done. Eventually, the test and the outcome should/could lead to some development where a seedbox can be totally independent for several trackers including downloading and uploading (both data and torrent-files).
Testing, as stated, will be in a closed environment thus a LAN to eliminated the overhead or strain on "the internet" where a lot of factors are not under my own control like switching, packet-shaping on a certain hop and so on. But I can also reverse the testing and give the 1st seeder a Gbit-uplink and have 25 clients connect at 100Mbit each.
I will look into that SDK for some more info after I'm done building the server and clients on a seperate VLAN from my own LAN.
Re: Torrent performance testing?
Yeah that's a good idea running the seedbox on 1gbps. Only after 100mbps will you start to meet some limitations of torrenting. Personally I don't think you'll find much difference in performance between windows/linux, I would just go with whatever is easier to work with overall.
I still think uTorrent will be the fastest out of all of them, again because of the features and flexibility it has. It has the best disk caching and management out of all the other clients, so if your trying to squeeze the most performance out of an overworked 7200rpm drive, it performs the best. Modern cpu's are way faster then you'll ever need, your next bottleneck after network bandwidth is hard disk speed.
I would suspect you'll be able to max out a standard magnetic drive at around 500mbps. If your looking to bump it up to a full 1gbps I would recommend using an SSD, or if you can't afford one setup a ramdisk. I used to use ramdisks for network benchmarking. I can link you to a torrent I created with some of the old ramdisk software I used to use, but it's mostly Windows Xp/Windows 2000/Windows 2003.
On the other hand if your making your own scripts and features, ease of development would probably be the most important. I'd rather have a client that's easy to program for and work with then fast. It sounds to me from your last post that your goal is to add usability rather then speed.
I forgot to add, it was recently confirmed that the uTorrent dev team is in progress creating a native linux version of uTorrent. I would probably expect some alpha builds out by this summer.
