So，a simple test can show AES is much faster inside KVM because of poor _x86_64_AES_encrypt_compact
###inside### [r...@xen-727057^10.227 ~]# time ssh 10.12.10.192 'time dd if=/dev/zero bs=1M count=300 | openssl aes-128-cbc -k 1...@2323 > /dev/null' 300+0 records in 300+0 records out 314572800 bytes (315 MB) copied, 3.38921 seconds, 92.8 MB/s real 0m3.399s user 0m2.415s sys 0m1.701s real 0m3.591s user 0m0.000s sys 0m0.000s ###outside### [r...@xen-727057^10.227 ~]# time ssh 10.12.10.48 'time dd if=/dev/zero bs=1M count=300 | openssl aes-128-cbc -k 1...@2323 > /dev/null' 300+0 records in 300+0 records out 314572800 bytes (315 MB) copied, 5.22114 seconds, 60.2 MB/s real 0m5.224s user 0m4.972s sys 0m0.655s real 0m5.324s user 0m0.000s sys 0m0.000s Regards, Kirby Zhou from SOHU-RD +86-10-6272-8261 -----Original Message----- From: Kirby Zhou [mailto:[email protected]] Sent: Thursday, August 12, 2010 5:07 PM To: 'Red Hat Enterprise Linux 6 (Santiago) Beta releases discussion mailing-list' Subject: RE: [rhelv6-beta-list] A simple program runs under KVM faster than under nake PC. With oprofile and gdb, I found that, openssl takes different code path inside and outside KVM. Outside the KVM, openssl calls _x86_64_AES_encrypt_compact, it seems slow than_x86_64_AES_encrypt. #### outside KVM #### [...@djt_10_48 ~]# opcontrol --start; opcontrol --reset; ./a.out 1 10000; opcontrol --dump; opreport -l Profiler running. Signalling daemon... done crc:a4580383 warning: /no-vmlinux could not be found. CPU: CPU with timer interrupt, speed 0 MHz (estimated) Profiling through timer interrupt samples % app name symbol name 20694 78.9847 no-vmlinux /no-vmlinux 4797 18.3092 libcrypto.so.0.9.8e _x86_64_AES_encrypt_compact 455 1.7366 libz.so.1.2.3 crc32 167 0.6374 libcrypto.so.0.9.8e AES_cbc_encrypt 61 0.2328 a.out io_routine(void*) 7 0.0267 bash /bin/bash 4 0.0153 GBK.so gconv 3 0.0115 oprofiled /usr/bin/oprofiled 2 0.0076 libc-2.5.so mbrtowc 2 0.0076 libcrypto.so.0.9.8e _x86_64_AES_set_encrypt_key 1 0.0038 ld-2.5.so _dl_allocate_tls_init 1 0.0038 ld-2.5.so _dl_sysdep_start 1 0.0038 ld-2.5.so do_lookup_x 1 0.0038 libc-2.5.so __close_nocancel 1 0.0038 libc-2.5.so _dl_addr 1 0.0038 libc-2.5.so _int_malloc 1 0.0038 libc-2.5.so malloc 1 0.0038 libc-2.5.so memset #### inside KVM #### [...@djt_10_192 ~]# opcontrol --start; opcontrol --reset; ./a.out 1 10000; opcontrol --dump; opreport -l Profiler running. Signalling daemon... done crc:a4580383 warning: /no-vmlinux could not be found. CPU: CPU with timer interrupt, speed 0 MHz (estimated) Profiling through timer interrupt samples % app name symbol name 13659 81.7415 no-vmlinux /no-vmlinux 2338 13.9916 libcrypto.so.0.9.8e _x86_64_AES_encrypt 481 2.8785 libz.so.1.2.3 crc32 72 0.4309 libcrypto.so.0.9.8e AES_cbc_encrypt 54 0.3232 a.out io_routine(void*) 14 0.0838 bash /bin/bash 6 0.0359 libc-2.5.so _int_malloc 5 0.0299 ld-2.5.so _dl_relocate_object 5 0.0299 ld-2.5.so do_lookup_x 5 0.0299 libc-2.5.so _nl_intern_locale_data 5 0.0299 libcrypto.so.0.9.8e _x86_64_AES_set_encrypt_key 4 0.0239 ld-2.5.so _dl_lookup_symbol_x 3 0.0180 GBK.so gconv 3 0.0180 ld-2.5.so _dl_map_object_deps 3 0.0180 ld-2.5.so dl_main 3 0.0180 ld-2.5.so strcmp 3 0.0180 libc-2.5.so _dl_addr 3 0.0180 libc-2.5.so mbrtowc 3 0.0180 libc-2.5.so memset 2 0.0120 gawk /bin/gawk 2 0.0120 ld-2.5.so _dl_load_cache_lookup 2 0.0120 ld-2.5.so _dl_setup_hash 2 0.0120 libc-2.5.so __register_atfork 2 0.0120 libc-2.5.so _int_free 2 0.0120 libc-2.5.so malloc 2 0.0120 oprofiled /usr/bin/oprofiled 1 0.0060 grep /bin/grep 1 0.0060 ld-2.5.so .text 1 0.0060 ld-2.5.so _dl_check_map_versions 1 0.0060 ld-2.5.so _dl_map_object_from_fd 1 0.0060 ld-2.5.so _dl_new_object 1 0.0060 ld-2.5.so _dl_sysdep_start 1 0.0060 ld-2.5.so check_match.8509 1 0.0060 ld-2.5.so memcpy 1 0.0060 ld-2.5.so memset 1 0.0060 libc-2.5.so _IO_file_finish@@GLIBC_2.2.5 1 0.0060 libc-2.5.so __close_nocancel 1 0.0060 libc-2.5.so __default_morecore 1 0.0060 libc-2.5.so __init_misc 1 0.0060 libc-2.5.so _int_realloc 1 0.0060 libc-2.5.so _nl_load_locale_from_archive 1 0.0060 libc-2.5.so create_cd_newstate 1 0.0060 libc-2.5.so free 1 0.0060 libc-2.5.so new_composite_name 1 0.0060 libc-2.5.so parse_reg_exp 1 0.0060 libc-2.5.so re_string_reconstruct 1 0.0060 libc-2.5.so setlocale 1 0.0060 libc-2.5.so sigprocmask 1 0.0060 libc-2.5.so strrchr 1 0.0060 libc-2.5.so strstr 1 0.0060 libc-2.5.so sysconf 1 0.0060 libc-2.5.so wctype_l 1 0.0060 mkdir /bin/mkdir Regards, Kirby Zhou from SOHU-RD +86-10-6272-8261 -----Original Message----- From: [email protected] [mailto:[email protected]] On Behalf Of Kirby Zhou Sent: Monday, August 09, 2010 2:43 PM To: 'Red Hat Enterprise Linux 6 (Santiago) Beta releases discussion mailing-list' Subject: Re: [rhelv6-beta-list] A simple program runs under KVM faster than under nake PC. Please check my source code. This is a PURE CPU program. No read, No Write. So I do not think cache is the point. Regards, Kirby Zhou from SOHU-RD +86-10-6272-8261 -----Original Message----- From: [email protected] [mailto:[email protected]] On Behalf Of John Summerfield Sent: Monday, August 09, 2010 8:01 AM To: Red Hat Enterprise Linux 6 (Santiago) Beta releases discussion mailing-list Subject: Re: [rhelv6-beta-list] A simple program runs under KVM faster than under nake PC. Kirby Zhou wrote: > Both RHEL5/RHEL6 encounter the same situation. > > Is there any potential bug? > > > > ]# g++ testaes.cpp -lcrypto -lz If the program is I/O bound, I would expect the virtual box to run faster _for smallish files_. hdparm can report I/O speeds an order of magnitude faster than on real machines. The reason is that the host caches the real disk. If you overflow the cache, results will be different. Time this, I find it's quire a bit slower in any VM I've tried. #!/usr/bin/perl #use integer; $i = 0; while ($i < 10000) { $j = 0; while ($j < 10000) { ++$j; } ++$i; } 07:59 [sum...@penguin ~]$ -- Cheers John -- spambait [email protected] [email protected] -- Advice http://webfoot.com/advice/email.top.php http://www.catb.org/~esr/faqs/smart-questions.html http://support.microsoft.com/kb/555375 You cannot reply off-list:-) _______________________________________________ rhelv6-beta-list mailing list [email protected] https://www.redhat.com/mailman/listinfo/rhelv6-beta-list _______________________________________________ rhelv6-beta-list mailing list [email protected] https://www.redhat.com/mailman/listinfo/rhelv6-beta-list _______________________________________________ rhelv5-list mailing list [email protected] https://www.redhat.com/mailman/listinfo/rhelv5-list
