WAN Test-bed with Objectivity 5.2 in a multi-server configuration
proposal
5 January
2000
Monarc Italian test-beds Working Group
A.Brunengo,
A.Ghiselli, L.Luminari, L.Perini, S.Resconi, M.Sgaravatto, C.Vistoli
1 Introduction................................................................................ 3
2 Test
Model................................................................................. 3
2.1 Test
objectives....................................................................... 4
2.2 Test
characteristics................................................................ 4
3 QoS
network.............................................................................. 4
4 Testbed...................................................................................... 5
4.1 Test-bed
links........................................................................ 6
4.2 Hardware
Configuration........................................................... 6
5 Conclusions................................................................................ 6
6 Bibliografy................................................................................. 7
The test results carry out with Objectivity 5.1 on different network configurations [1] enhanced several interesting points; some of them are related to the software release like limited server capacity to deal with an high number of links (>30) and high CPU consuming with high network speed (100Mbps). In order to investigate the improvements of the new release 5.2 of Objectivity some tests will be repeated over LAN and WAN scenarios. The WAN tests brought out the minimum network speed representing a strong bottleneck for the job execution time: with network speed less than 8Mbps the job execution time is ~ 1000sec, that is unacceptable. LAN tests at 10Mbps demonstrated to be a reasonable scenario regarding CPU usage, number of Jobs and network utilization ( 10Mbps: 1 job=200, 6 jobs = 600, CPU/server < 20%, CPU/client ~50% , throughput 8.5Mbps). Therefore the value of 10Mbps has been considered as lower threshold for network speed through WAN connections in order to investigate multi-server configuration and to evaluate the comparison between LAN and WAN behaviors. The comparison between LAN and WAN at the same speed is very interesting to investigate the influence of WAN latency on the system performance.
The logical layout of the multi-server test-bed is represented in the following figure:
There will be three data servers, at least, equipped with multiprocessor systems and multigigabytes storage systems. Several client machines will connect the data servers through LAN and WAN links. This will provide a direct comparison between LAN and WAN behavior and to see the network impact to the application behavior and efficiency.
The most important specific objectives are:
§ Check AMS and Lock server behavior and performance in a multi-server configuration.
§ Stress test by running several analysis jobs accessing to the Data Base and performance measure.
§ Locate system bottlenecks.
§ Collect ‘response time’ measures to give input to model simulation.
§ Understand network impact to the application behavior and efficiency.
§ Understand the impact of different data models on the multi-server configuration.
The general test scenario is very simple regarding to database characteristics and structure. This choice has been done in order evaluate the system performance in the simplest configuration.
§ Atlfast++ program is used to populate the database following the Tag/Event data model proposed by the LHC++ project to read data from the database
§ 3 AMS servers
§ 1 single federation Objectivity database containing about 50.000 events (~ 2Gbytes).
§ Application program performs read/write access to the database
The procedure followed to perform these tests consists in submitting an increasing number of concurrent jobs from each client and then monitoring CPU utilization, network throughput and job execution time(wall clock time).
Network Layout will be configured with QoS mechanisms based on Differentiated Services (DS) allowing different priority traffic flows. The aim is to perform a careful evaluation of TCP performance and Application performance and draw conclusions about how to configure DS to provide a premium service in this situation.
The work will be performed in the context of the Globus Architecture for Reservation and Allocation (GARA) which builds on these mechanisms to deliver per-flow, advance reservation, end-to-end Quality of service.
The first layout is described in the figure below.
There will be 3 data server: one in Milano, one in CNAF and one at CERN. These servers will be interconnected with dedicated links at 10Mbps. At each of these servers will be linked different client sites at 10Mbps: Genova to Milano, Padova, Roma and Bologna will connect CNAF. However each client can access every data servers in the test-bed; for ex. A client in Roma can access server at CERN or in Milano.
10Mbps Links through GARR-B:
§ Milano-CERN (via TEN-155)
§ CNAF-Milano
§ Padova-CNAF
§ Roma-CNAF
§ Genova-Milano
Dedicated network equipment will manage the dedicated links for the IP test-bed. Since the managed bandwidth offered by GARR-B and TEN-155 is via ATM/VPs the necessary network hardware is:
1. More flexible solution
§ If an ATM switch is connected to GARR-B for production traffic then a Dedicated Router connected to the ATM switch and with a LAN interface is necessary.
2.
Or less flexible
solution
§ If a production router with an ATM interface to GARR-B ATM switch is used, then a dedicated LAN interface is necessary. In this case QoS can not be tested.
Dedicated Workstation
Multi-process CPU with memory size greater than 256MB and storage size equal/greater than 20Gbytes
Testing Objectivity 5.2 has the highest priority; since it is supposed to be ‘multithread’ and therefore able to use multiprocessor systems in efficient way, we should be able to actually test the scenario with 4 SUN connected at 100Mbps and then at 1000Mbps.
The most important test will be to configure multi-servers and several clients on WAN based on 10Mbps links in order to compare the results with those obtained with the same speeds on LAN. It will be important to check if delays at that speed will affect the results. Later on configuring DS on the test-bed will provide the opportunity to evaluate the feasibility to give a premium service to this applications together with background traffic. GARA tools will be used for advance reservation of premium service.
1 - “Network Tests with Objectivity 5.1 using a single federated database” A.Brunengo, A.Ghiselli, L.Luminari, L.Perini, S.Resconi, M.Sgaravatto, C.Vistoli, MONARC Internal Note, Nov. 99 (Http://www.cern.ch/MONARC/docs/monarc-docs.html)