WEBVTT 00:00.000 --> 00:10.920 Hello, warm. I'm Luca Basi. I'm a computer science student at University of 00:10.920 --> 00:18.040 Bolonia. I'm doing a scholarship funded by Garra, ATI and F and Knaf in 00:18.040 --> 00:23.520 Bolonia. Today we'll present a site acts that is initiative for network 00:23.520 --> 00:29.320 traffic tagging for scientific computer. Let me introduce a little bit 00:29.320 --> 00:34.120 the environment. So we're talking about scientific communities that 00:34.120 --> 00:40.160 move very large amounts of data between small and large data 00:40.160 --> 00:45.240 center, in particular using the connectivity provided by national 00:45.240 --> 00:50.520 research and location networks. One of these is the Italian one that is 00:50.520 --> 00:56.640 called Garra. That is, for example, provide a whopping 1.6-terabit per second 00:56.640 --> 01:03.200 connections between the Knaf data center in Bolonia. That is the National 01:03.200 --> 01:07.360 Military Center of Italy, National Institute for Nuclear Physics and 01:07.360 --> 01:14.120 Bechern. That's probably, you know, in Geneva. And all of these with a 01:14.120 --> 01:19.800 latency for whole year, nine and half milliseconds thanks to multidermaine 01:19.800 --> 01:27.920 shared spectrum. So we need to bear with this amount of data transfer better 01:27.920 --> 01:36.240 understand and understand of the network traffic and how to optimize it to ensure 01:36.240 --> 01:47.360 to use the network as effective as possible. The worldwide LAC computing grid is 01:47.360 --> 01:53.000 distributed computing infrastructure that supports the storage and distribution 01:53.000 --> 02:00.120 and the analysis of the data produced by the large 02:00.120 --> 02:09.320 other on colliders experiments in Geneva. And we have like one more 170 sites 02:09.320 --> 02:17.640 around the world. There is the Tier 1, the Tier 0, that is one in Geneva. 14 Tier 02:17.640 --> 02:26.080 1s and one of these is the one in Bolonia managed by Knaf. And 100 and 02:26.080 --> 02:34.080 half Tier 2 around the globe. So the is quite complex to manage that 02:34.080 --> 02:41.480 transfer between all these sites and provide data with a lot of 02:41.480 --> 02:48.800 latency to all physics around the world. So the site access initiative was 02:48.800 --> 02:55.040 burned to provide the intensification of science domain and their 02:55.040 --> 03:01.440 high level activities of an actor level and this can be done with the 03:01.440 --> 03:06.640 rising packet and flow market. In particular, this is important for 03:06.640 --> 03:12.640 network providers to identify with the traffic owner and the purpose of 03:12.640 --> 03:17.920 these traffic for the experiments to understand out there that are flows 03:17.920 --> 03:25.000 performed along the complex network path and to sites so to data center to 03:25.000 --> 03:33.120 game visibility into our very different data flows performs around the network. 03:33.120 --> 03:40.960 In particular, stacks provides two ways to mark a network traffic flow marking 03:40.960 --> 03:48.240 with the UDP for flies and packet market, packet marking using the IPvsa flow 03:48.240 --> 03:55.360 label, either field invader. Each flow can be identified with two 03:55.360 --> 04:04.400 information, the experiment and the activity of the experiment. For example, 04:04.400 --> 04:21.040 if we take the address experiment, that is one of a major experiment at 04:21.040 --> 04:27.680 Sharon that has experimented the two, that is doing data consolidation activity 04:27.680 --> 04:36.080 that has ID4, the error will be shot at 100, 30. All the these ID are 04:36.080 --> 04:42.880 statically mapped at the moment in a flow registry. That is just a JSON file with 04:42.880 --> 04:48.240 this information. So let's talk about the two way to mark traffic and let's 04:48.240 --> 04:54.680 start with flow label with UDP for flies. Fresh flies are just simple UDP 04:54.680 --> 05:04.200 packets in a six log format with a defined JSON body. Packets are intended to be sent to 05:04.200 --> 05:11.240 the same host of destination of a traffic to a specific port and and this packet 05:11.240 --> 05:17.240 that is intended to be work-radable because we want that all the arrays and network 05:17.240 --> 05:24.200 provider that receive this traffic, cool intercept is UDP fly flies and every formation 05:24.200 --> 05:31.880 about the network traffic. Also, the packets could also be sent to a specific 05:31.880 --> 05:40.680 regional global collectors if the sites prefer. They use of CSLog format was because 05:40.680 --> 05:47.640 it's easy to reuse a radius tablish tool, for example log session and these type of 05:47.640 --> 05:58.760 marking was for IPv4 and IPv6. Content is not limited but is limited as long as fit a 05:58.760 --> 06:09.880 single frame. The other way to mark traffic is with IPv5 label invader. So I 06:09.880 --> 06:16.360 BFF traffic control program is to insert these information, the site-axing 06:16.360 --> 06:27.960 information in the IPv6 flow label. Some is a 20 bits long field, some are used for entropy. 06:27.960 --> 06:34.360 So I random, some from the communities of experiments that produce the activity and 06:34.600 --> 06:42.760 some other bits are used for the activity. Actually, we are proposing these two AETF 06:43.720 --> 06:51.400 with IRFC and is available to link invaders if you're curious to a specific 06:51.800 --> 06:59.480 to a specification of this part. So now let's see implementation of 06:59.480 --> 07:06.920 standard web data of site-axing software used in WICG. So let me introduce storm, 07:06.920 --> 07:13.320 storm stands for storage resource manager. Is a grid storage solution for this base storage 07:13.320 --> 07:22.120 system with a POSIX file system, typically is a distributed file system such as 07:22.120 --> 07:33.640 GPSS. In particular in WICG, storm is adopted by the tier 1 of set-out knaff and by some 07:33.640 --> 07:41.480 other tier 2. In particular storm is a suite of software and today we will talk about specifically 07:41.480 --> 07:49.000 about some web data. That is one that provides connectivity, provides functionality 07:49.000 --> 07:55.480 through web data protocol. The web data protocol is an extension of the HTTP protocol that 07:55.480 --> 08:05.240 allow to manage files on a remote server. For example, we added some new HTTP verbs, for example, 08:05.240 --> 08:13.560 MK call to create a directory or copy to copy the file remotely. And in particular, it's 08:13.560 --> 08:23.000 on WebDub, it implements an extension of the copy verb of WebDub. So an extension of an extension 08:23.000 --> 08:30.920 that is called third-party copies. And these are a really useful functionality because 08:30.920 --> 08:38.200 permits to transfer data between data center without actually doing some local copies. So you can 08:38.200 --> 08:45.240 say to a remote server copy the file that you have locally to these other remote endpoint or 08:45.240 --> 08:53.800 get to these files from these remote endpoint and copy to your remote storage. So that is the 08:53.800 --> 09:02.760 most used functionality through transfer data between the various data center in WICG. 09:02.760 --> 09:11.400 The latest release of Storm WebDub has obtained support for Site-Eyes Ailer. 09:11.400 --> 09:20.200 In particular, it relies on Flodim. Flodim is a demon that has different plugins to retrieve 09:20.200 --> 09:26.760 the flow information and a set of commands to mark the traffic. In particular, Storm WebDub 09:26.760 --> 09:36.200 used the NPAPI plugin to retrieve the one to receive identifiers through an impipe. And 09:36.200 --> 09:45.560 sends the UDPFrive flies using Flodim. So Storm WebDub writes a line to a specific file in this 09:45.560 --> 09:54.360 form, where a state is start-to-end of the flow. The protocol, so TCP or RDP, 09:54.360 --> 10:03.320 the source IP import, the mediation IP import, and the information extracted from the Site-Eyes Ailer. 10:03.320 --> 10:11.160 So we experimented in the activity ID. So let's see how the third-party copy in push mode. 10:11.160 --> 10:19.960 Also, copy this file that you have, log it to this remote server, worse using Site-Eyes Ailer. 10:19.960 --> 10:27.960 The client to HTTP copy request specifying a remote destination. In this case, we are sending 10:27.960 --> 10:36.920 this request to server-racing, copy this file to destination B. And in the request, it is included 10:36.920 --> 10:47.480 the Site-Eyes Ailer. In this case, 65 or random. Storm WebDub received a request. Start 10:47.480 --> 10:55.640 doing the HTTP put request to server B as extracted by the HTTP copy request. And at the same 10:55.640 --> 11:04.120 time, send the information about activity ID experiment ID to the flow, the Demon, that send 11:04.120 --> 11:12.120 the UDP file files. The transfer continues, and when the transfer ends, Storm WebDub communicated 11:12.120 --> 11:21.160 the end of the network flow to flow D, and flow D send the UDP file files containing the information 11:21.160 --> 11:31.480 that the transfer of the file is ended. So to sum up, Site-Eyes permits to mark packets 11:31.480 --> 11:37.160 and network flow with information about experiments and activity responsible for a network 11:37.160 --> 11:44.520 traffic. The latest release of Storm WebDub has obtained support to these two parts of 11:44.520 --> 11:53.080 this information. And at the moment, is in development, a parting of the moment is written 11:53.080 --> 12:06.040 in Python, and we are also exploring to use other fields in the IPv6 Ailer instead of a 12:06.040 --> 12:13.080 flow label. We are trying with the destination option, a copy option as alternative. So thank 12:13.080 --> 12:19.080 you for your attention. You refine the website of Site-Eyes Initiative. There is a mail list. 12:19.080 --> 12:25.800 We are open to contribution for other type of scientific experiments. At the moment, we are 12:25.800 --> 12:30.840 mainly eye-energy physics because this initiative was born and served, but we are open to 12:32.120 --> 12:39.000 other type of scientific experiment, the source code of the webDub, and all the Site-Eyes 12:39.000 --> 12:45.720 contributors. So thank you. And if there is question.