Hi everyone, I am using the gem5 standard library to simulate a cluster architecture with Ruby MESI_THREE_LEVEL cache protocol. The Script I am using is configs/example/gem5_library/x86-parsec-benchmark.py. I create a new Topology file in src/python/gem5/components/cachehierarchies/ruby/topologies and use cluster_id to route cache message between corresponding l2 and l3 caches. It seems there must happen dead lock, so that router cannot work correctly. The benchmark keep running, which should be end in 10mins. And it can work right using P2P provided in original repo. Have anyone seen this situation? What can I do to debug and solve this bug? Follow is my architecture: — Regards, Congwu Zhang

Hi everyone, I ran parsec.blackscholes benchmark with test(benchmark size) dataset, and it could get results in about 1 min. However `simsmall` dataset keeps running about 30 mins without an end. Is this normal situation? Or does it mean my gem5 garnet network runs into a dead lock or something else? How can I debug this? Any help would be appreciated! — Regards, Congwu Zhang > On Jul 10, 2023, at 19:18, zhangcongwu via gem5-users <gem5-users@gem5.org> wrote: > > Hi everyone, > > I am using the gem5 standard library to simulate a cluster architecture with Ruby MESI_THREE_LEVEL cache protocol. The Script I am using is configs/example/gem5_library/x86-parsec-benchmark.py. I create a new Topology file in src/python/gem5/components/cachehierarchies/ruby/topologies and use cluster_id to route cache message between corresponding l2 and l3 caches. > > It seems there must happen dead lock, so that router cannot work correctly. The benchmark keep running, which should be end in 10mins. > > And it can work right using P2P provided in original repo. > > Have anyone seen this situation? What can I do to debug and solve this bug? > > Follow is my architecture: > <config.board.cache_hierarchy.ruby_system.dot.pdf> > > — > Regards, > Congwu Zhang_______________________________________________ > gem5-users mailing list -- gem5-users@gem5.org > To unsubscribe send an email to gem5-users-leave@gem5.org

Hi Congwu, Were you running a full-system simulation of PARSEC? This process usually takes a substantial amount of time. Here's a benchmark from my experience for reference: On a Linux Ubuntu local machine (i7 12th gen), a full-system simulation of PARSEC on Garnet's 16-node Mesh NoC with 16 L2 caches, and without any other caches It took approximately 15 minutes for setup and about 1.2 hours for the region of interest (ROI) of Streamcluster Simsmall. If you're using a virtual machine to run gem5, expect the ROI time for the same case to be around 5 times longer than on a local machine. I hope this information is helpful. Best regards, Henry -----Original Message----- From: zhangcongwu via gem5-users <gem5-users@gem5.org> Sent: 11 July 2023 02:00 To: The gem5 Users mailing list <gem5-users@gem5.org> Cc: zhangcongwu <zhangcongwu@ict.ac.cn> Subject: [gem5-users] Update [Garnet Network Dead Lock]: Running parsec benchmark stuck using mesi_three_level cluster_id Hi everyone, I ran parsec.blackscholes benchmark with test(benchmark size) dataset, and it could get results in about 1 min. However `simsmall` dataset keeps running about 30 mins without an end. Is this normal situation? Or does it mean my gem5 garnet network runs into a dead lock or something else? How can I debug this? Any help would be appreciated! — Regards, Congwu Zhang > On Jul 10, 2023, at 19:18, zhangcongwu via gem5-users <gem5-users@gem5.org> wrote: > > Hi everyone, > > I am using the gem5 standard library to simulate a cluster architecture with Ruby MESI_THREE_LEVEL cache protocol. The Script I am using is configs/example/gem5_library/x86-parsec-benchmark.py. I create a new Topology file in src/python/gem5/components/cachehierarchies/ruby/topologies and use cluster_id to route cache message between corresponding l2 and l3 caches. > > It seems there must happen dead lock, so that router cannot work correctly. The benchmark keep running, which should be end in 10mins. > > And it can work right using P2P provided in original repo. > > Have anyone seen this situation? What can I do to debug and solve this bug? > > Follow is my architecture: > <config.board.cache_hierarchy.ruby_system.dot.pdf> > > — > Regards, > Congwu Zhang_______________________________________________ > gem5-users mailing list -- gem5-users@gem5.org To unsubscribe send an > email to gem5-users-leave@gem5.org _______________________________________________ gem5-users mailing list -- gem5-users@gem5.org To unsubscribe send an email to gem5-users-leave@gem5.org