Re: [Xen-devel] [PATCH 1/5] xentrace_format: print timestamps in nanoseconds
Hello George, On 11.09.18 18:54, George Dunlap wrote: FYI, I never use xentrace_format; as far as I'm concerned it's been made obsolete by xenalyze, and if it were up to me I'd remove it from the tree. Lots of people seem to find it useful, so I review patches. But I really care very little about what its functionality ends up being. Anyway, thank you for your review. -- *Andrii Anisov* ___ Xen-devel mailing list Xen-devel@lists.xenproject.org https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [Xen-devel] [PATCH 1/5] xentrace_format: print timestamps in nanoseconds
Hello Dario, On 12.09.18 10:42, Dario Faggioli wrote: Sorry, I'm not sure I'm getting this properly. When you say "with the current formula", do you mean before or after this series? I did mean the formula existing before this series. IAC, changing the default format file that we ship so that it prints time in the format seconds.nanoseconds (if CPU speed is specified), would be nice IMO. On the second though I realized it is quite inconvenient. From the formats file we can't distinguish if tsc contains float sec.ns or integer cycles. For cycles it will look like following: CPU0 1090699305.0 (+ 336) rtds:tickle [ cpu = 0 ] As said already, the one reason why I think it still could be useful, and we should keep it, is that the way a trace will look like can be changed or tweaked rather easily, and without having to recompile anything, via the formats file, which is rather cool. :-) +1 -- *Andrii Anisov* ___ Xen-devel mailing list Xen-devel@lists.xenproject.org https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [Xen-devel] [PATCH 1/5] xentrace_format: print timestamps in nanoseconds
On Tue, 2018-09-11 at 16:54 +0100, George Dunlap wrote: > On 09/11/2018 04:19 PM, Andrii Anisov wrote: > > > > On 11.09.18 13:44, George Dunlap wrote: > > > What I do in xenalyze is to have the timestamps in seconds, but > > > always > > > print down to the nanosecond. (For this I actually break cpu > > > cycles > > > into s and ns separately, and then print "%u.%09u".) > > > > Here, we can have the same. With the 0current formula in > > xentrace_format, but changing `%(tsc)d` to `%(tsc).9f` in formats. > > Sorry, I'm not sure I'm getting this properly. When you say "with the current formula", do you mean before or after this series? IAC, changing the default format file that we ship so that it prints time in the format seconds.nanoseconds (if CPU speed is specified), would be nice IMO. > FYI, I never use xentrace_format; as far as I'm concerned it's been > made > obsolete by xenalyze, and if it were up to me I'd remove it from the > tree. Lots of people seem to find it useful, so I review patches. > I almost entirely concur. As said already, the one reason why I think it still could be useful, and we should keep it, is that the way a trace will look like can be changed or tweaked rather easily, and without having to recompile anything, via the formats file, which is rather cool. :-) Regards, Dario -- <> (Raistlin Majere) - Dario Faggioli, Ph.D, http://about.me/dario.faggioli Software Engineer @ SUSE https://www.suse.com/ signature.asc Description: This is a digitally signed message part ___ Xen-devel mailing list Xen-devel@lists.xenproject.org https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [Xen-devel] [PATCH 1/5] xentrace_format: print timestamps in nanoseconds
On 09/11/2018 04:19 PM, Andrii Anisov wrote: > > On 11.09.18 13:44, George Dunlap wrote: >> What I do in xenalyze is to have the timestamps in seconds, but always >> print down to the nanosecond. (For this I actually break cpu cycles >> into s and ns separately, and then print "%u.%09u".) > Here, we can have the same. With the 0current formula in > xentrace_format, but changing `%(tsc)d` to `%(tsc).9f` in formats. > BTW, I've just noticed, that reltsc is allways in cycles. And it seems > odd, as well. FYI, I never use xentrace_format; as far as I'm concerned it's been made obsolete by xenalyze, and if it were up to me I'd remove it from the tree. Lots of people seem to find it useful, so I review patches. But I really care very little about what its functionality ends up being. So feel free to propose whatever patches you want; I have very little opinion, as long as it keeps working. :-) -George ___ Xen-devel mailing list Xen-devel@lists.xenproject.org https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [Xen-devel] [PATCH 1/5] xentrace_format: print timestamps in nanoseconds
On 11.09.18 13:44, George Dunlap wrote: What I do in xenalyze is to have the timestamps in seconds, but always print down to the nanosecond. (For this I actually break cpu cycles into s and ns separately, and then print "%u.%09u".) Here, we can have the same. With the 0current formula in xentrace_format, but changing `%(tsc)d` to `%(tsc).9f` in formats. BTW, I've just noticed, that reltsc is allways in cycles. And it seems odd, as well. -- *Andrii Anisov* ___ Xen-devel mailing list Xen-devel@lists.xenproject.org https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [Xen-devel] [PATCH 1/5] xentrace_format: print timestamps in nanoseconds
On 09/11/2018 11:44 AM, George Dunlap wrote: > On 09/11/2018 11:32 AM, Andrii Anisov wrote: >> Hello George, >> >> >> On 11.09.18 13:15, George Dunlap wrote: if mhz: - tsc = tsc / (mhz*100.0) + tsc = tsc * 1000.0 / mhz >>> Why do you prefer this? >> I'm playing with scheduling from one hand, so time stamps in seconds >> does not give understanding about what's going on. >> From other hand I'm quite confused about how useful timestamps in >> seconds could be for traces. As per my understanding, tracer should be >> useful for debugging some rapidly changing processes. > > Oh, sorry -- I missed the point of this patch. > > --- > xentrace_format: print timestamps in nanoseconds > > ...rather than seconds. This should rather say, "..rather than seconds, when the clock speed is specified." -George ___ Xen-devel mailing list Xen-devel@lists.xenproject.org https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [Xen-devel] [PATCH 1/5] xentrace_format: print timestamps in nanoseconds
On 09/11/2018 11:32 AM, Andrii Anisov wrote: > Hello George, > > > On 11.09.18 13:15, George Dunlap wrote: >>> if mhz: >>> - tsc = tsc / (mhz*100.0) >>> + tsc = tsc * 1000.0 / mhz >> Why do you prefer this? > I'm playing with scheduling from one hand, so time stamps in seconds > does not give understanding about what's going on. > From other hand I'm quite confused about how useful timestamps in > seconds could be for traces. As per my understanding, tracer should be > useful for debugging some rapidly changing processes. Oh, sorry -- I missed the point of this patch. --- xentrace_format: print timestamps in nanoseconds ...rather than seconds. Having timestamps for rapidly-occurring events in nanoseconds makes it easier to understand what's going on. While here, document the -c option. --- What I do in xenalyze is to have the timestamps in seconds, but always print down to the nanosecond. (For this I actually break cpu cycles into s and ns separately, and then print "%u.%09u".) But this is also fine with me. -George ___ Xen-devel mailing list Xen-devel@lists.xenproject.org https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [Xen-devel] [PATCH 1/5] xentrace_format: print timestamps in nanoseconds
Hello George, On 11.09.18 13:15, George Dunlap wrote: if mhz: -tsc = tsc / (mhz*100.0) +tsc = tsc * 1000.0 / mhz Why do you prefer this? I'm playing with scheduling from one hand, so time stamps in seconds does not give understanding about what's going on. From other hand I'm quite confused about how useful timestamps in seconds could be for traces. As per my understanding, tracer should be useful for debugging some rapidly changing processes. -- *Andrii Anisov* ___ Xen-devel mailing list Xen-devel@lists.xenproject.org https://lists.xenproject.org/mailman/listinfo/xen-devel