Hi Ioi, Now that the HOTSPOTMKSHOME env var is pointing at my cygwin directory...
On 8/28/13 9:32 PM, Ioi Lam wrote: > Pete, > > By the way, for the purposes of testing your build speed, instead of > spending a long time building JDK8, maybe you can just build hotspot. > It's not completely proportional but would give you some ideas whether > you should consider getting another laptop :-) > > Start Menu -> All Programs -> Microsoft Visual Studio 2010 -> Visual > Studio Tools -> Visual Studio Command Prompt (2010). > > This will open a DOS window with the x86 version of the VS2010 > tool-chain selected. > > cd hotspot\make\windows > PATH %PATH%;c:\cygwin\bin > create d:\re\jdk8-b102\windows-i586 > > When it's done, open the project file > hotspot\build\vs-i486\jvm.vcxproj in VS2010 IDE. > > Tools -> Options -> Project and Solutions -> VC++ Project Settings -> > Build Timing-> Yes > > Choose the"compiler1_debug" + "Win32" build combination > > Build the solution 112446 ms = ~1.9 min > Clean the solution > Build the solutionfor the second time 97939 ms = ~1.6 min > > If you have 8GB RAM, the second build should have all the files > cached, so you should have very little disk reads. It will still have > writes because you're creating the obj files. > > On my laptop, even before the SSD swap, the second build time is below > 1 minute, and you can see CPU utilization constantly at 100% (all 4 > "CPUs" in Windows Task Manager are at 100%). I have 8 GB RAM and 2 cores. While often at or close to 100% both cores were not constantly at that level; often CPU usage was in the 70s. (I used Process Explorer.) CL was about 40% of the cpu usage. IO was about 80% read and 20% write, mostly by cl.exe and some by mspdbsrv.exe. Process explorer shows IO values of 10 MB at the peaks, and I'll guess a 5 MB average. I don't know what kind of values these are; I'm guessing the size of a block of a single transfer. > > Switching to an SSD actually doesn't help with this scenario (becuase > it is CPU bound), but it does make the system 100 times more > responsive. Switching between apps is instantaneous. I can opening up > a few copies of different versions of Visual Studio IDE and they come > up just in a few seconds. With the HDD it would take ages. > > Good luck! > > - Ioi > > > > > > On 08/28/2013 07:14 PM, Ioi Lam wrote: >> On 08/28/2013 11:47 AM, Pete Brunet wrote: >>> Thanks Tim, >>> >>> On 8/27/13 6:44 PM, Tim Bell wrote: >>>> Good advice on the virus scanning front, and of course every bit >>>> helps. >>>> >>>> Also keep in mind that laptop components often sacrifice performance >>>> in favor of saving space, power, and battery life. A laptop is not a >>>> workstation. >>>> >>>> If your laptop drive is a sad little unit like mine, you pay and pay. >>>> I use my laptop as a gateway to get access to more powerful hardware. >>>> >>>> Not sure what version of Windows you have, but as a quick measure you >>>> could try running WinSAT if you have it, like this: >>>> >>>> >>>> C:/Windows/System32/WinSAT.exe disk >>> Since winset ran in a separate window and it closed before I could use >>> the info I needed this advice: >>> http://superuser.com/questions/93826/winsat-command-line-closes-too-fast >>> >>> Running as admin solved it. >>>> >>>> I'm sure your mileage will vary, but on my laptop I measured numbers >>>> like: >>>> >>>> 56.536 ms for "Average Read Time with Sequential Writes" >>>> 15.061 ms for "Average Read Time with Random Writes" >>>> 153.103 ms for "Latency: 95th Percentile" >>> Here are my numbers: 23.629, 30.545, 113.723 which are about 10x worse >>> than your numbers below. It would be interesting to see someone's >>> numbers for a laptop SSD. >> >> Here are my numbers on a standard issue Dell laptop (Win7, i5-2520M >> 2.5Ghz, 8GB RAM). I can't tell the model number since I am using >> remote log-on now. >> >> I had swapped out the standard HDD with a Samsung 840 PRO SSD 256GB >> (MZ-7PD256BW). It cost about $230 on amazon and is rated as "best" by >> some web site I last checked :-) >> >> > Disk Sequential 64.0 Read 530.91 MB/s 7.9 >> > Disk Random 16.0 Read 449.02 MB/s 7.9 >> > Responsiveness: Average IO Rate 0.21 ms/IO 7.9 >> > Responsiveness: Grouped IOs 6.77 units 7.7 >> > Responsiveness: Long IOs 0.45 units 7.9 >> > Responsiveness: Overall 3.02 units 7.9 >> > Responsiveness: PenaltyFactor 0.0 >> > Disk Sequential 64.0 Write 471.76 MB/s 7.9 >> > Average Read Time with Sequential Writes 0.270 ms 7.9 >> > Latency: 95th Percentile 1.463 ms 7.9 >> > Latency: Maximum 5.702 ms 7.9 >> > Average Read Time with Random Writes 0.267 ms 7.9 >> > Total Run Time 00:01:05.79 >> >> I'll pull the latest jdk8 and post my build times. >> >> Ioi >> >>>> Whereas on a VM located somewhere out in a datacenter I have never >>>> seen, I measured: >>>> >>>> 2.573 ms for "Average Read Time with Sequential Writes" >>>> 3.574 ms for "Average Read Time with Random Writes " >>>> 13.008 ms for "Latency: 95th Percentile" >>>> >>>> In both cases I was running Windows 7 64-bit. Virus scanning and >>>> other S/W are installed as per corporate policy. Also, I am only >>>> picking on the VM because it was convenient. Try the same tests on >>>> some bare metal workstation hardware if you have a chance. >>> What do I need to do to use a datacenter VM? Would the latency from >>> Austin negate the gain in disk performance? >>>> Hope this helps. >>>> >>>> Tim >>>> >>>> On 08/27/13 03:51 PM, Kelly O'Hair wrote: >>>>> Peter, >>>>> >>>>> It might be important for you to make sure the boot jdk image, and >>>>> all tools used to build (like VS2010) also in the ignore list. Maybe >>>>> CYGWIN too. >>>>> The boot jdk rt.jar file in particular can be slow to virus scan. >>>>> It kind of depends on how smart (or not) the virus scanner is on >>>>> re-scanning already scanned files. >>>>> >>>>> -kto >>>>> >>>>> On Aug 27, 2013, at 2:14 PM, Pete Brunet wrote: >>>>> >>>>>> Here are my release build times with my repo tree in the Norton 360 >>>>>> real >>>>>> time virus checking ignore list: >>>>>> >>>>>> 00:03:34 corba >>>>>> 00:12:51 demos >>>>>> 00:15:10 docs >>>>>> 00:18:42 hotspot >>>>>> 00:15:55 images >>>>>> 00:01:42 jaxp >>>>>> 00:07:32 jaxws >>>>>> 00:43:04 jdk >>>>>> 00:04:17 langtools >>>>>> 00:02:31 nashorn >>>>>> 02:05:18 TOTAL >>>>>> >>>>>> Here are the prior times: >>>>>> >>>>>> 00:05:55 corba >>>>>> 00:09:46 demos >>>>>> 00:15:00 docs >>>>>> 00:18:37 hotspot >>>>>> 00:17:13 images >>>>>> 00:03:32 jaxp >>>>>> 00:11:41 jaxws >>>>>> 01:05:35 jdk >>>>>> 00:06:47 langtools >>>>>> 00:02:26 nashorn >>>>>> 02:37:45 TOTAL >>>>>> >>>>>> A half hour improvement is nice, but looks like I need to switch to >>>>>> an SSD. >>>>>> >>>>>> Pete >>>>>> >>>>>> On 8/9/13 11:25 AM, Pete Brunet wrote: >>>>>>> Any suggestions for SSDs? My T500 takes 2.5" SATA II. >>>>>>> >>>>>>> On 8/9/13 10:44 AM, Andreas Rieber wrote: >>>>>>>> Hi Peter, >>>>>>>> >>>>>>>> try perfmon.exe to see where the bottleneck is. Most likely it is >>>>>>>> memory or io. If you have the option, go for a SSD drive (you will >>>>>>>> get >>>>>>>> build times ~10 minutes). >>>>>>>> >>>>>>>> Andreas >>>>>>>> >>>>>>>> >>>>>>>> On 09.08.13 16:44, Pete Brunet wrote: >>>>>>>>> My product is Norton 360. To turn it off I right click on the >>>>>>>>> Norton >>>>>>>>> 360 icon in the system tray and choose Disable Antivirus >>>>>>>>> Auto-Protect. >>>>>>>>> If anyone knows if this is not sufficient and what else needs to >>>>>>>>> be done >>>>>>>>> please let me know. >>>>>>>>> >>>>>>>>> Pete >>>>>>>>> >>>>>>>>> On 8/9/13 1:58 AM, Erik Joelsson wrote: >>>>>>>>>> I wonder if anti virus is really turned off here. In my >>>>>>>>>> experience, >>>>>>>>>> it's hard to make it stop completely. I have a slightly older >>>>>>>>>> laptop, >>>>>>>>>> same brand, that builds this in around 40-45 minutes. >>>>>>>>>> >>>>>>>>>> /Erik >>>>>>>>>> >>>>>>>>>> On 2013-08-09 03:32, Pete Brunet wrote: >>>>>>>>>>> Haven't rebuilt a release version yet for a true comparison but >>>>>>>>>>> here are >>>>>>>>>>> my fastdebug times with anti-virus turned off; about the >>>>>>>>>>> same as >>>>>>>>>>> release >>>>>>>>>>> with anti-virus on. >>>>>>>>>>> >>>>>>>>>>> 00:04:30 corba >>>>>>>>>>> 00:07:25 demos >>>>>>>>>>> 00:16:28 docs >>>>>>>>>>> 00:16:15 hotspot >>>>>>>>>>> 00:23:13 images >>>>>>>>>>> 00:04:18 jaxp >>>>>>>>>>> 00:07:29 jaxws >>>>>>>>>>> 00:59:41 jdk >>>>>>>>>>> 00:05:11 langtools >>>>>>>>>>> 00:01:42 nashorn >>>>>>>>>>> 02:26:32 TOTAL >>>>>>>>>>> >>>>>>>>>>> Pete >>>>>>>>>>> >>>>>>>>>>> On 8/6/13 12:55 PM, Kelly O'Hair wrote: >>>>>>>>>>>> 4. Using a debug jdk as your boot jdk >>>>>>>>>>>> >>>>>>>>>>>> :) >>>>>>>>>>>> >>>>>>>>>>>> Langtools taking 6 minutes is crazy long. Since there is no >>>>>>>>>>>> native >>>>>>>>>>>> code there, #1 would be my first guess. >>>>>>>>>>>> >>>>>>>>>>>> -kto >>>>>>>>>>>> >>>>>>>>>>>> On Aug 5, 2013, at 5:54 PM, Weijun Wang wrote: >>>>>>>>>>>> >>>>>>>>>>>>> From my experiences there are 3 reasons a compile could be >>>>>>>>>>>>> slow: >>>>>>>>>>>>> >>>>>>>>>>>>> 1. Anti-Virus programs >>>>>>>>>>>>> 2. Virtual Machines >>>>>>>>>>>>> 3. Network folders >>>>>>>>>>>>> >>>>>>>>>>>>> --Max >>>>>>>>>>>>> >>>>>>>>>>>>> On 8/6/13 12:48 AM, Pete Brunet wrote: >>>>>>>>>>>>>> Hi Erik, >>>>>>>>>>>>>> >>>>>>>>>>>>>> 00:05:55 corba >>>>>>>>>>>>>> 00:09:46 demos >>>>>>>>>>>>>> 00:15:00 docs >>>>>>>>>>>>>> 00:18:37 hotspot >>>>>>>>>>>>>> 00:17:13 images >>>>>>>>>>>>>> 00:03:32 jaxp >>>>>>>>>>>>>> 00:11:41 jaxws >>>>>>>>>>>>>> 01:05:35 jdk >>>>>>>>>>>>>> 00:06:47 langtools >>>>>>>>>>>>>> 00:02:26 nashorn >>>>>>>>>>>>>> 02:37:45 TOTAL >>>>>>>>>>>>>> >>>>>>>>>>>>>> Pete >>>>>>>>>>>>>> >>>>>>>>>>>>>> On 8/5/13 6:55 AM, Erik Joelsson wrote: >>>>>>>>>>>>>>> That was disappointingly slow for sure. Do you have the >>>>>>>>>>>>>>> build time >>>>>>>>>>>>>>> summary listing? >>>>>>>>>>>>>>> >>>>>>>>>>>>>>> /Erik >>>>>>>>>>>>>>> >>>>>>>>>>>>>>> On 2013-08-02 22:26, Pete Brunet wrote: >>>>>>>>>>>>>>>> FWIW I just built 32 bit jdk8 release on 64 bit win 7 with >>>>>>>>>>>>>>>> cygwin 1.7.21 >>>>>>>>>>>>>>>> on a dual core 8 GB 2.53 GHz Lenovo T500 and the build >>>>>>>>>>>>>>>> time was >>>>>>>>>>>>>>>> 2:37:45. >> >