Download presentation
Presentation is loading. Please wait.
Published byBeverley Nash Modified over 9 years ago
1
(Preliminary) Results of Evaluation of the CCT SB110 Peter Chochula and Svetozár Kapusta 1 1 Comenius University, Bratislava
2
P. Chochula and S. Kapusta ALICE TB, December 25, 2003 Credits The measurements were possible thanks to extensive help of Markus Joos (EP/ESS)
3
P. Chochula and S. Kapusta ALICE TB, December 25, 2003 Motivation of the tests Motivation of tests: If CCT SB110 should be adopted in ALICE, we should think about interfacing with online systems DIM is the communication standard adopted in ALICE online systems. So far no tests were performed using DIM and CCT VP110 Why DCS team has been involved in testing: DCS team participates in software development for the SPD project, which is using VME Some other detectors mention in URD’s the possibility of monitoring their equipment using VME
4
P. Chochula and S. Kapusta ALICE TB, December 25, 2003 Test Methods A real hardware (SPD JTAG controller) has been used We tested continuous reading of JTAG memory, measuring transfer speed for different sizes of transferred blocks The same tests were performed using DIM RPC call transmitting the command to acquire data and returning the array containing the content of JTAG’s buffer Performance of CCT VP110 has been compared with National Instruments PCI to VME bridge (MXI-II)
5
P. Chochula and S. Kapusta ALICE TB, December 25, 2003 Comment on test methods We wanted to see the VP110 operating in almost realistic conditions We were not interested in peak performance Software was restricted to standard functions using only information available in existing documentation Hardware has been used in the default (factory preset) configuration
6
P. Chochula and S. Kapusta ALICE TB, December 25, 2003 Test Setup.. P-IV Win XP P-IV Win XP MXI VP110 DIM Server DIM Client DIM Server LAN
7
P. Chochula and S. Kapusta ALICE TB, December 25, 2003 Measurements of sustained transfer rate Time needed to read a block of a given size from JTAG’s memory has been measured VISA Library has been used to communicate with MXI-II ATLAS driver library has been used for CCT VP110 Data was read in a programmatic loop, transferring a 32-bit word at a time Corresponding functions are VI_IN for MXI and ReadSafe for VP100. Results were compared with “pseudo block transfer” function VI_MoveIN for MXI (this is using the hardware to loop over read commands) In addition tests were performed using LabView and C++ languages
8
P. Chochula and S. Kapusta ALICE TB, December 25, 2003 Sustained transfer rate as a function of the size of transferred block MXI VI_MoveIn (C++) MXI VI_MoveIn (LabView) MXI VI_In (C++) MXI VI_In (LabView) CCT ReadSafe (C++)
9
P. Chochula and S. Kapusta ALICE TB, December 25, 2003 Comparison of VI_In and ATLAS driver functions MXI VI_In (C++) MXI VI_In (LabView) CCT ReadSafe (C++)
10
P. Chochula and S. Kapusta ALICE TB, December 25, 2003 Partial Conclusions VI_MoveIn significantly outperforms programmatic approach MXI profits from faster CPU No significant difference between LabView and C++ (however this is not very relevant information)
11
P. Chochula and S. Kapusta ALICE TB, December 25, 2003 Tests with DIM protocol Compilation of DIM package on VP110 went very smoothly We “suffered” a bit by the fact that AFS was not available (this could be easily implemented if needed)
12
P. Chochula and S. Kapusta ALICE TB, December 25, 2003 Performance tests involving DIM (Very PRELIMINARY!!) MXI VI_In (C++) CCT ReadSafe (C++) Please note: presented numbers are very preliminary. Fair tests should be performed maybe on separate network or at least outside working hours)
13
P. Chochula and S. Kapusta ALICE TB, December 25, 2003 Impact of performance differences between VP110 and MXI For monitoring (a few kB read in about every 1s) there will be no visible difference in performance However factor ~2 difference in speed would mean that we will need 20 instead of 10 minutes for typical SPD calibration SPD therefore prefers to keep existing MXI setups and use them also in the future, profiting from fast PCs SPD has some ~10 working installations with MXI, the same setups are used by LHC-B and Na60 and no problems were experienced over ~4 years
14
P. Chochula and S. Kapusta ALICE TB, December 25, 2003 Remarks (I) We experienced problems due to the fact that support for AMO used by the JTAG controller was not implemented in ATLAS library The problem has been promptly solved by Markus Recommendation: all perspective users should tests their hardware with VP110 well in advance as some features might not be available and it could take some time to get the system working
15
P. Chochula and S. Kapusta ALICE TB, December 25, 2003 Remarks (II) VP110 requires crate with J0 connectors (MXI-2 does not implement this feature) Additional backplane adapter is needed in case that one aims for using the PCI, external IDE disks etc.
16
P. Chochula and S. Kapusta ALICE TB, December 25, 2003 Remarks (III) The price difference between MXI-II and VP110 is about 1 kCHF, MXI needs in addition an extra PC However: BOOTP and NFS server should be provided for diskless VP110 based systems ( of course this will be shared by all controllers) At least a serial monitor (FALCO) is needed for VP110 (will they still exists or they will be replaced by a cheap PC?)
17
P. Chochula and S. Kapusta ALICE TB, December 25, 2003 Remarks (IV) There are ways to reduce price of MXI based systems profiting from daisy- chaining and grouping features:....
18
P. Chochula and S. Kapusta ALICE TB, December 25, 2003 Remarks (V) There is typically a need for extra PC in control system acting as OPC server, or controlling CANbus chain etc The same computer can control MXI This PC can be upgraded if needed, VP110 will stay the same (PIII 800MHz with possible upgrade in 2004 to PIII 1.2GHz) We plan tests with 2U rack-mounted PC hosting MXI, CAN controller and running OPC over Ethernet
19
P. Chochula and S. Kapusta ALICE TB, December 25, 2003 Future plans Finish DIM performance tests Perform long-term stability tests Test the setup using a rack-mounted PC running some parallel tasks (CAEN OPC server etc.)
20
P. Chochula and S. Kapusta ALICE TB, December 25, 2003 Conclusions DIM has been successfully tested on VP110 (long-term stability still has to be checked) The performance of implemented CPU could be a limiting factor for some applications We would like to recommend to consider MXI as alternative to VP110 where applicable Hardware should be always tested for software compatibility
Similar presentations
© 2025 SlidePlayer.com. Inc.
All rights reserved.