Re: [EXTERNAL] Re: GC Tuning https://thelastpickle.com/blog/2018/04/11/gc-tuning.html
Thanks Jon! I used that tool and I did a test to compare LCS and STCS and it works great. However, I was referring to the JVM flags that you use since there are a lot of flags that I found as default and I would like to exclude the unused or wrong ones from the current configuration. I have also another thread opened where I am trying to figure out Kernel Settings for TCP https://lists.apache.org/thread.html/7708c22a1d95882598cbcc29bc34fa54c01fcb33c40bb616dcd3956d@%3Cuser.cassandra.apache.org%3E Do you have anything to add to that? Thanks, Sergio Il giorno lun 21 ott 2019 alle ore 15:09 Jon Haddad ha scritto: > tlp-stress comes with workloads pre-baked, so there's not much > configuration to do. The main flags you'll want are going to be: > > -d : duration, I highly recommend running your test for a few days > --compaction > --compression > -p: number of partitions > -r: % of reads, 0-1 > > For example, you might run: > > tlp-stress run KeyValue -d 24h --compaction lcs -p 10m -r .9 > > for a basic key value table, running for 24 hours, using LCS, 10 million > partitions, 90% reads. > > There's a lot of options. I won't list them all here, it's why I wrote the > manual :) > > Jon > > > On Mon, Oct 21, 2019 at 1:16 PM Sergio wrote: > >> Thanks, guys! >> I just copied and paste what I found on our test machines but I can >> confirm that we have the same settings except for 8GB in production. >> I didn't select these settings and I need to verify why these settings >> are there. >> If any of you want to share your flags for a read-heavy workload it would >> be appreciated, so I would replace and test those flags with TLP-STRESS. >> I am thinking about different approaches (G1GC vs ParNew + CMS) >> How many GB for RAM do you dedicate to the OS in percentage or in an >> exact number? >> Can you share the flags for ParNew + CMS that I can play with it and >> perform a test? >> >> Best, >> Sergio >> >> >> Il giorno lun 21 ott 2019 alle ore 09:27 Reid Pinchback < >> rpinchb...@tripadvisor.com> ha scritto: >> >>> Since the instance size is < 32gb, hopefully swap isn’t being used, so >>> it should be moot. >>> >>> >>> >>> Sergio, also be aware that -XX:+CMSClassUnloadingEnabled probably >>> doesn’t do anything for you. I believe that only applies to CMS, not >>> G1GC. I also wouldn’t take it as gospel truth that -XX:+UseNUMA is a good >>> thing on AWS (or anything virtualized), you’d have to run your own tests >>> and find out. >>> >>> >>> >>> R >>> >>> *From: *Jon Haddad >>> *Reply-To: *"user@cassandra.apache.org" >>> *Date: *Monday, October 21, 2019 at 12:06 PM >>> *To: *"user@cassandra.apache.org" >>> *Subject: *Re: [EXTERNAL] Re: GC Tuning >>> https://thelastpickle.com/blog/2018/04/11/gc-tuning.html >>> >>> >>> >>> *Message from External Sender* >>> >>> One thing to note, if you're going to use a big heap, cap it at 31GB, >>> not 32. Once you go to 32GB, you don't get to use compressed pointers [1], >>> so you get less addressable space than at 31GB. >>> >>> >>> >>> [1] >>> https://blog.codecentric.de/en/2014/02/35gb-heap-less-32gb-java-jvm-memory-oddities/ >>> <https://urldefense.proofpoint.com/v2/url?u=https-3A__blog.codecentric.de_en_2014_02_35gb-2Dheap-2Dless-2D32gb-2Djava-2Djvm-2Dmemory-2Doddities_=DwMFaQ=9Hv6XPedRSA-5PSECC38X80c1h60_XWA4z1k_R1pROA=OIgB3poYhzp3_A7WgD7iBCnsJaYmspOa2okNpf6uqWc=e9Ahs5XXRBicgUhMZQaboxsqb6jXpjvo48kEojUWaQc=Q7jI4ZEqVMFZIMPoSXTvMebG5fWOUJ6lhDOgWGxiHg8=> >>> >>> >>> >>> On Mon, Oct 21, 2019 at 11:39 AM Durity, Sean R < >>> sean_r_dur...@homedepot.com> wrote: >>> >>> I don’t disagree with Jon, who has all kinds of performance tuning >>> experience. But for ease of operation, we only use G1GC (on Java 8), >>> because the tuning of ParNew+CMS requires a high degree of knowledge and >>> very repeatable testing harnesses. It isn’t worth our time. As a previous >>> writer mentioned, there is usually better return on our time tuning the >>> schema (aka helping developers understand Cassandra’s strengths). >>> >>> >>> >>> We use 16 – 32 GB heaps, nothing smaller than that. >>> >>> >>> >>> Sean Durity >>> >>> >>> >>> *From:* Jon Haddad >>> *Sent:* Monday, October 21, 2019 10:43 AM >>> *To:* user@
Re: [EXTERNAL] Re: GC Tuning https://thelastpickle.com/blog/2018/04/11/gc-tuning.html
tlp-stress comes with workloads pre-baked, so there's not much configuration to do. The main flags you'll want are going to be: -d : duration, I highly recommend running your test for a few days --compaction --compression -p: number of partitions -r: % of reads, 0-1 For example, you might run: tlp-stress run KeyValue -d 24h --compaction lcs -p 10m -r .9 for a basic key value table, running for 24 hours, using LCS, 10 million partitions, 90% reads. There's a lot of options. I won't list them all here, it's why I wrote the manual :) Jon On Mon, Oct 21, 2019 at 1:16 PM Sergio wrote: > Thanks, guys! > I just copied and paste what I found on our test machines but I can > confirm that we have the same settings except for 8GB in production. > I didn't select these settings and I need to verify why these settings are > there. > If any of you want to share your flags for a read-heavy workload it would > be appreciated, so I would replace and test those flags with TLP-STRESS. > I am thinking about different approaches (G1GC vs ParNew + CMS) > How many GB for RAM do you dedicate to the OS in percentage or in an exact > number? > Can you share the flags for ParNew + CMS that I can play with it and > perform a test? > > Best, > Sergio > > > Il giorno lun 21 ott 2019 alle ore 09:27 Reid Pinchback < > rpinchb...@tripadvisor.com> ha scritto: > >> Since the instance size is < 32gb, hopefully swap isn’t being used, so it >> should be moot. >> >> >> >> Sergio, also be aware that -XX:+CMSClassUnloadingEnabled probably >> doesn’t do anything for you. I believe that only applies to CMS, not >> G1GC. I also wouldn’t take it as gospel truth that -XX:+UseNUMA is a good >> thing on AWS (or anything virtualized), you’d have to run your own tests >> and find out. >> >> >> >> R >> >> *From: *Jon Haddad >> *Reply-To: *"user@cassandra.apache.org" >> *Date: *Monday, October 21, 2019 at 12:06 PM >> *To: *"user@cassandra.apache.org" >> *Subject: *Re: [EXTERNAL] Re: GC Tuning >> https://thelastpickle.com/blog/2018/04/11/gc-tuning.html >> >> >> >> *Message from External Sender* >> >> One thing to note, if you're going to use a big heap, cap it at 31GB, not >> 32. Once you go to 32GB, you don't get to use compressed pointers [1], so >> you get less addressable space than at 31GB. >> >> >> >> [1] >> https://blog.codecentric.de/en/2014/02/35gb-heap-less-32gb-java-jvm-memory-oddities/ >> <https://urldefense.proofpoint.com/v2/url?u=https-3A__blog.codecentric.de_en_2014_02_35gb-2Dheap-2Dless-2D32gb-2Djava-2Djvm-2Dmemory-2Doddities_=DwMFaQ=9Hv6XPedRSA-5PSECC38X80c1h60_XWA4z1k_R1pROA=OIgB3poYhzp3_A7WgD7iBCnsJaYmspOa2okNpf6uqWc=e9Ahs5XXRBicgUhMZQaboxsqb6jXpjvo48kEojUWaQc=Q7jI4ZEqVMFZIMPoSXTvMebG5fWOUJ6lhDOgWGxiHg8=> >> >> >> >> On Mon, Oct 21, 2019 at 11:39 AM Durity, Sean R < >> sean_r_dur...@homedepot.com> wrote: >> >> I don’t disagree with Jon, who has all kinds of performance tuning >> experience. But for ease of operation, we only use G1GC (on Java 8), >> because the tuning of ParNew+CMS requires a high degree of knowledge and >> very repeatable testing harnesses. It isn’t worth our time. As a previous >> writer mentioned, there is usually better return on our time tuning the >> schema (aka helping developers understand Cassandra’s strengths). >> >> >> >> We use 16 – 32 GB heaps, nothing smaller than that. >> >> >> >> Sean Durity >> >> >> >> *From:* Jon Haddad >> *Sent:* Monday, October 21, 2019 10:43 AM >> *To:* user@cassandra.apache.org >> *Subject:* [EXTERNAL] Re: GC Tuning >> https://thelastpickle.com/blog/2018/04/11/gc-tuning.html >> <https://urldefense.proofpoint.com/v2/url?u=https-3A__thelastpickle.com_blog_2018_04_11_gc-2Dtuning.html=DwMFaQ=9Hv6XPedRSA-5PSECC38X80c1h60_XWA4z1k_R1pROA=OIgB3poYhzp3_A7WgD7iBCnsJaYmspOa2okNpf6uqWc=e9Ahs5XXRBicgUhMZQaboxsqb6jXpjvo48kEojUWaQc=YFRUQ6Rdb5mcFf6GqguRYCsrcAcP6KzjozIgYp56riE=> >> >> >> >> I still use ParNew + CMS over G1GC with Java 8. I haven't done a >> comparison with JDK 11 yet, so I'm not sure if it's any better. I've heard >> it is, but I like to verify first. The pause times with ParNew + CMS are >> generally lower than G1 when tuned right, but as Chris said it can be >> tricky. If you aren't willing to spend the time understanding how it works >> and why each setting matters, G1 is a better option. >> >> >> >> I wouldn't run Cassandra in production on less than 8GB of heap - I >> consider it the absol
Re: [EXTERNAL] Re: GC Tuning https://thelastpickle.com/blog/2018/04/11/gc-tuning.html
Think of GB to OS as something intended to support file caching. As such the amount is whatever suits your usage. If your use is almost exclusively reading, then file cache memory doesn’t matter that much if you’re operating with your storage as those nvme ssd drives that the i3’s come with. There is already a chunk cache that you should be tuning in C* instead, and feeding fast from the O/S file cache, assuming compressed SSTables, maybe turns out to be less of a concern. If you have moderate write activity then your situation changes because then that same file cache is how your dirty background pages turn into eventual flushes to disk, and so you have to watch the impact of read stalls when the I/O fills with write requests. You might not see this so obviously on nvme drives, but that could depend a lot on the distro and kernels and how the filesystem is mounted. My super strong advice on issues like this is to not cargo-cult other people’s tunings. Look at them for ideas, sure. But learn how to do your own investigations, and budget the time for it into your project. Budget a LOT of time for it if your measure of “good performance” is based on latency; when “good” is defined in terms of throughput your life is easier. Also, everything is always a little different in virtualization, and lord knows you can have screwball things appear in AWS. The good news is you don’t need a perfect configuration out of the gate; you need a configuration you understand and can refine; understanding comes from knowing how to do your own performance monitoring. From: Sergio Reply-To: "user@cassandra.apache.org" Date: Monday, October 21, 2019 at 1:16 PM To: "user@cassandra.apache.org" Subject: Re: [EXTERNAL] Re: GC Tuning https://thelastpickle.com/blog/2018/04/11/gc-tuning.html Message from External Sender Thanks, guys! I just copied and paste what I found on our test machines but I can confirm that we have the same settings except for 8GB in production. I didn't select these settings and I need to verify why these settings are there. If any of you want to share your flags for a read-heavy workload it would be appreciated, so I would replace and test those flags with TLP-STRESS. I am thinking about different approaches (G1GC vs ParNew + CMS) How many GB for RAM do you dedicate to the OS in percentage or in an exact number? Can you share the flags for ParNew + CMS that I can play with it and perform a test? Best, Sergio Il giorno lun 21 ott 2019 alle ore 09:27 Reid Pinchback mailto:rpinchb...@tripadvisor.com>> ha scritto: Since the instance size is < 32gb, hopefully swap isn’t being used, so it should be moot. Sergio, also be aware that -XX:+CMSClassUnloadingEnabled probably doesn’t do anything for you. I believe that only applies to CMS, not G1GC. I also wouldn’t take it as gospel truth that -XX:+UseNUMA is a good thing on AWS (or anything virtualized), you’d have to run your own tests and find out. R From: Jon Haddad mailto:j...@jonhaddad.com>> Reply-To: "user@cassandra.apache.org<mailto:user@cassandra.apache.org>" mailto:user@cassandra.apache.org>> Date: Monday, October 21, 2019 at 12:06 PM To: "user@cassandra.apache.org<mailto:user@cassandra.apache.org>" mailto:user@cassandra.apache.org>> Subject: Re: [EXTERNAL] Re: GC Tuning https://thelastpickle.com/blog/2018/04/11/gc-tuning.html<https://urldefense.proofpoint.com/v2/url?u=https-3A__thelastpickle.com_blog_2018_04_11_gc-2Dtuning.html=DwMFaQ=9Hv6XPedRSA-5PSECC38X80c1h60_XWA4z1k_R1pROA=OIgB3poYhzp3_A7WgD7iBCnsJaYmspOa2okNpf6uqWc=UvvIpm6RP7FYRQH6S5EPXTsxAMsezbm6QzHNB0zmMG0=jmk5lyXeQ6gwlVWF86TKWUbIhy57G5tOnlLEps8-DQw=> Message from External Sender One thing to note, if you're going to use a big heap, cap it at 31GB, not 32. Once you go to 32GB, you don't get to use compressed pointers [1], so you get less addressable space than at 31GB. [1] https://blog.codecentric.de/en/2014/02/35gb-heap-less-32gb-java-jvm-memory-oddities/<https://urldefense.proofpoint.com/v2/url?u=https-3A__blog.codecentric.de_en_2014_02_35gb-2Dheap-2Dless-2D32gb-2Djava-2Djvm-2Dmemory-2Doddities_=DwMFaQ=9Hv6XPedRSA-5PSECC38X80c1h60_XWA4z1k_R1pROA=OIgB3poYhzp3_A7WgD7iBCnsJaYmspOa2okNpf6uqWc=e9Ahs5XXRBicgUhMZQaboxsqb6jXpjvo48kEojUWaQc=Q7jI4ZEqVMFZIMPoSXTvMebG5fWOUJ6lhDOgWGxiHg8=> On Mon, Oct 21, 2019 at 11:39 AM Durity, Sean R mailto:sean_r_dur...@homedepot.com>> wrote: I don’t disagree with Jon, who has all kinds of performance tuning experience. But for ease of operation, we only use G1GC (on Java 8), because the tuning of ParNew+CMS requires a high degree of knowledge and very repeatable testing harnesses. It isn’t worth our time. As a previous writer mentioned, there is usually better return on our time tuning the schema (aka helping developers understand Cassandra’s strengths). We use 16 – 32 GB heaps, nothing smaller than that.
Re: [EXTERNAL] Re: GC Tuning https://thelastpickle.com/blog/2018/04/11/gc-tuning.html
Thanks, guys! I just copied and paste what I found on our test machines but I can confirm that we have the same settings except for 8GB in production. I didn't select these settings and I need to verify why these settings are there. If any of you want to share your flags for a read-heavy workload it would be appreciated, so I would replace and test those flags with TLP-STRESS. I am thinking about different approaches (G1GC vs ParNew + CMS) How many GB for RAM do you dedicate to the OS in percentage or in an exact number? Can you share the flags for ParNew + CMS that I can play with it and perform a test? Best, Sergio Il giorno lun 21 ott 2019 alle ore 09:27 Reid Pinchback < rpinchb...@tripadvisor.com> ha scritto: > Since the instance size is < 32gb, hopefully swap isn’t being used, so it > should be moot. > > > > Sergio, also be aware that -XX:+CMSClassUnloadingEnabled probably > doesn’t do anything for you. I believe that only applies to CMS, not > G1GC. I also wouldn’t take it as gospel truth that -XX:+UseNUMA is a good > thing on AWS (or anything virtualized), you’d have to run your own tests > and find out. > > > > R > > *From: *Jon Haddad > *Reply-To: *"user@cassandra.apache.org" > *Date: *Monday, October 21, 2019 at 12:06 PM > *To: *"user@cassandra.apache.org" > *Subject: *Re: [EXTERNAL] Re: GC Tuning > https://thelastpickle.com/blog/2018/04/11/gc-tuning.html > > > > *Message from External Sender* > > One thing to note, if you're going to use a big heap, cap it at 31GB, not > 32. Once you go to 32GB, you don't get to use compressed pointers [1], so > you get less addressable space than at 31GB. > > > > [1] > https://blog.codecentric.de/en/2014/02/35gb-heap-less-32gb-java-jvm-memory-oddities/ > <https://urldefense.proofpoint.com/v2/url?u=https-3A__blog.codecentric.de_en_2014_02_35gb-2Dheap-2Dless-2D32gb-2Djava-2Djvm-2Dmemory-2Doddities_=DwMFaQ=9Hv6XPedRSA-5PSECC38X80c1h60_XWA4z1k_R1pROA=OIgB3poYhzp3_A7WgD7iBCnsJaYmspOa2okNpf6uqWc=e9Ahs5XXRBicgUhMZQaboxsqb6jXpjvo48kEojUWaQc=Q7jI4ZEqVMFZIMPoSXTvMebG5fWOUJ6lhDOgWGxiHg8=> > > > > On Mon, Oct 21, 2019 at 11:39 AM Durity, Sean R < > sean_r_dur...@homedepot.com> wrote: > > I don’t disagree with Jon, who has all kinds of performance tuning > experience. But for ease of operation, we only use G1GC (on Java 8), > because the tuning of ParNew+CMS requires a high degree of knowledge and > very repeatable testing harnesses. It isn’t worth our time. As a previous > writer mentioned, there is usually better return on our time tuning the > schema (aka helping developers understand Cassandra’s strengths). > > > > We use 16 – 32 GB heaps, nothing smaller than that. > > > > Sean Durity > > > > *From:* Jon Haddad > *Sent:* Monday, October 21, 2019 10:43 AM > *To:* user@cassandra.apache.org > *Subject:* [EXTERNAL] Re: GC Tuning > https://thelastpickle.com/blog/2018/04/11/gc-tuning.html > <https://urldefense.proofpoint.com/v2/url?u=https-3A__thelastpickle.com_blog_2018_04_11_gc-2Dtuning.html=DwMFaQ=9Hv6XPedRSA-5PSECC38X80c1h60_XWA4z1k_R1pROA=OIgB3poYhzp3_A7WgD7iBCnsJaYmspOa2okNpf6uqWc=e9Ahs5XXRBicgUhMZQaboxsqb6jXpjvo48kEojUWaQc=YFRUQ6Rdb5mcFf6GqguRYCsrcAcP6KzjozIgYp56riE=> > > > > I still use ParNew + CMS over G1GC with Java 8. I haven't done a > comparison with JDK 11 yet, so I'm not sure if it's any better. I've heard > it is, but I like to verify first. The pause times with ParNew + CMS are > generally lower than G1 when tuned right, but as Chris said it can be > tricky. If you aren't willing to spend the time understanding how it works > and why each setting matters, G1 is a better option. > > > > I wouldn't run Cassandra in production on less than 8GB of heap - I > consider it the absolute minimum. For G1 I'd use 16GB, and never 4GB with > Cassandra unless you're rarely querying it. > > > > I typically use the following as a starting point now: > > > > ParNew + CMS > > 16GB heap > > 10GB new gen > > 2GB memtable cap, otherwise you'll spend a bunch of time copying around > memtables (cassandra.yaml) > > Max tenuring threshold: 2 > > survivor ratio 6 > > > > I've also done some tests with a 30GB heap, 24 GB of which was new gen. > This worked surprisingly well in my tests since it essentially keeps > everything out of the old gen. New gen allocations are just a pointer bump > and are pretty fast, so in my (limited) tests of this I was seeing really > good p99 times. I was seeing a 200-400 ms pause roughly once a minute > running a workload that deliberately wasn't hitting a resource limit > (testing real world looking stress vs overwhelming the cluster). > > > &g
Re: [EXTERNAL] Re: GC Tuning https://thelastpickle.com/blog/2018/04/11/gc-tuning.html
Since the instance size is < 32gb, hopefully swap isn’t being used, so it should be moot. Sergio, also be aware that -XX:+CMSClassUnloadingEnabled probably doesn’t do anything for you. I believe that only applies to CMS, not G1GC. I also wouldn’t take it as gospel truth that -XX:+UseNUMA is a good thing on AWS (or anything virtualized), you’d have to run your own tests and find out. R From: Jon Haddad Reply-To: "user@cassandra.apache.org" Date: Monday, October 21, 2019 at 12:06 PM To: "user@cassandra.apache.org" Subject: Re: [EXTERNAL] Re: GC Tuning https://thelastpickle.com/blog/2018/04/11/gc-tuning.html Message from External Sender One thing to note, if you're going to use a big heap, cap it at 31GB, not 32. Once you go to 32GB, you don't get to use compressed pointers [1], so you get less addressable space than at 31GB. [1] https://blog.codecentric.de/en/2014/02/35gb-heap-less-32gb-java-jvm-memory-oddities/<https://urldefense.proofpoint.com/v2/url?u=https-3A__blog.codecentric.de_en_2014_02_35gb-2Dheap-2Dless-2D32gb-2Djava-2Djvm-2Dmemory-2Doddities_=DwMFaQ=9Hv6XPedRSA-5PSECC38X80c1h60_XWA4z1k_R1pROA=OIgB3poYhzp3_A7WgD7iBCnsJaYmspOa2okNpf6uqWc=e9Ahs5XXRBicgUhMZQaboxsqb6jXpjvo48kEojUWaQc=Q7jI4ZEqVMFZIMPoSXTvMebG5fWOUJ6lhDOgWGxiHg8=> On Mon, Oct 21, 2019 at 11:39 AM Durity, Sean R mailto:sean_r_dur...@homedepot.com>> wrote: I don’t disagree with Jon, who has all kinds of performance tuning experience. But for ease of operation, we only use G1GC (on Java 8), because the tuning of ParNew+CMS requires a high degree of knowledge and very repeatable testing harnesses. It isn’t worth our time. As a previous writer mentioned, there is usually better return on our time tuning the schema (aka helping developers understand Cassandra’s strengths). We use 16 – 32 GB heaps, nothing smaller than that. Sean Durity From: Jon Haddad mailto:j...@jonhaddad.com>> Sent: Monday, October 21, 2019 10:43 AM To: user@cassandra.apache.org<mailto:user@cassandra.apache.org> Subject: [EXTERNAL] Re: GC Tuning https://thelastpickle.com/blog/2018/04/11/gc-tuning.html<https://urldefense.proofpoint.com/v2/url?u=https-3A__thelastpickle.com_blog_2018_04_11_gc-2Dtuning.html=DwMFaQ=9Hv6XPedRSA-5PSECC38X80c1h60_XWA4z1k_R1pROA=OIgB3poYhzp3_A7WgD7iBCnsJaYmspOa2okNpf6uqWc=e9Ahs5XXRBicgUhMZQaboxsqb6jXpjvo48kEojUWaQc=YFRUQ6Rdb5mcFf6GqguRYCsrcAcP6KzjozIgYp56riE=> I still use ParNew + CMS over G1GC with Java 8. I haven't done a comparison with JDK 11 yet, so I'm not sure if it's any better. I've heard it is, but I like to verify first. The pause times with ParNew + CMS are generally lower than G1 when tuned right, but as Chris said it can be tricky. If you aren't willing to spend the time understanding how it works and why each setting matters, G1 is a better option. I wouldn't run Cassandra in production on less than 8GB of heap - I consider it the absolute minimum. For G1 I'd use 16GB, and never 4GB with Cassandra unless you're rarely querying it. I typically use the following as a starting point now: ParNew + CMS 16GB heap 10GB new gen 2GB memtable cap, otherwise you'll spend a bunch of time copying around memtables (cassandra.yaml) Max tenuring threshold: 2 survivor ratio 6 I've also done some tests with a 30GB heap, 24 GB of which was new gen. This worked surprisingly well in my tests since it essentially keeps everything out of the old gen. New gen allocations are just a pointer bump and are pretty fast, so in my (limited) tests of this I was seeing really good p99 times. I was seeing a 200-400 ms pause roughly once a minute running a workload that deliberately wasn't hitting a resource limit (testing real world looking stress vs overwhelming the cluster). We built tlp-cluster [1] and tlp-stress [2] to help figure these things out. [1] https://thelastpickle.com/tlp-cluster/ [thelastpickle.com]<https://urldefense.com/v3/__https:/thelastpickle.com/tlp-cluster/__;!OYIaWQQGbnA!ZhiXAdRaL49J8nBlh0F_5MQ97Z1QNTUuTSMvksmEmxan3d65D6ATmQO1ig58W52u_EmQ1GM$> [2] http://thelastpickle.com/tlp-stress [thelastpickle.com]<https://urldefense.com/v3/__http:/thelastpickle.com/tlp-stress__;!OYIaWQQGbnA!ZhiXAdRaL49J8nBlh0F_5MQ97Z1QNTUuTSMvksmEmxan3d65D6ATmQO1ig58W52uuCUZYKw$> Jon On Mon, Oct 21, 2019 at 10:24 AM Reid Pinchback mailto:rpinchb...@tripadvisor.com>> wrote: An i3x large has 30.5 gb of RAM but you’re using less than 4gb for C*. So minus room for other uses of jvm memory and for kernel activity, that’s about 25 gb for file cache. You’ll have to see if you either want a bigger heap to allow for less frequent gc cycles, or you could save money on the instance size. C* generates a lot of medium-length lifetime objects which can easily end up in old gen. A larger heap will reduce the burn of more old-gen collections. There are no magic numbers to just give because it’ll depend on your usage pat
Re: [EXTERNAL] Re: GC Tuning https://thelastpickle.com/blog/2018/04/11/gc-tuning.html
One thing to note, if you're going to use a big heap, cap it at 31GB, not 32. Once you go to 32GB, you don't get to use compressed pointers [1], so you get less addressable space than at 31GB. [1] https://blog.codecentric.de/en/2014/02/35gb-heap-less-32gb-java-jvm-memory-oddities/ On Mon, Oct 21, 2019 at 11:39 AM Durity, Sean R wrote: > I don’t disagree with Jon, who has all kinds of performance tuning > experience. But for ease of operation, we only use G1GC (on Java 8), > because the tuning of ParNew+CMS requires a high degree of knowledge and > very repeatable testing harnesses. It isn’t worth our time. As a previous > writer mentioned, there is usually better return on our time tuning the > schema (aka helping developers understand Cassandra’s strengths). > > > > We use 16 – 32 GB heaps, nothing smaller than that. > > > > Sean Durity > > > > *From:* Jon Haddad > *Sent:* Monday, October 21, 2019 10:43 AM > *To:* user@cassandra.apache.org > *Subject:* [EXTERNAL] Re: GC Tuning > https://thelastpickle.com/blog/2018/04/11/gc-tuning.html > > > > I still use ParNew + CMS over G1GC with Java 8. I haven't done a > comparison with JDK 11 yet, so I'm not sure if it's any better. I've heard > it is, but I like to verify first. The pause times with ParNew + CMS are > generally lower than G1 when tuned right, but as Chris said it can be > tricky. If you aren't willing to spend the time understanding how it works > and why each setting matters, G1 is a better option. > > > > I wouldn't run Cassandra in production on less than 8GB of heap - I > consider it the absolute minimum. For G1 I'd use 16GB, and never 4GB with > Cassandra unless you're rarely querying it. > > > > I typically use the following as a starting point now: > > > > ParNew + CMS > > 16GB heap > > 10GB new gen > > 2GB memtable cap, otherwise you'll spend a bunch of time copying around > memtables (cassandra.yaml) > > Max tenuring threshold: 2 > > survivor ratio 6 > > > > I've also done some tests with a 30GB heap, 24 GB of which was new gen. > This worked surprisingly well in my tests since it essentially keeps > everything out of the old gen. New gen allocations are just a pointer bump > and are pretty fast, so in my (limited) tests of this I was seeing really > good p99 times. I was seeing a 200-400 ms pause roughly once a minute > running a workload that deliberately wasn't hitting a resource limit > (testing real world looking stress vs overwhelming the cluster). > > > > We built tlp-cluster [1] and tlp-stress [2] to help figure these things > out. > > > > [1] https://thelastpickle.com/tlp-cluster/ [thelastpickle.com] > <https://urldefense.com/v3/__https:/thelastpickle.com/tlp-cluster/__;!OYIaWQQGbnA!ZhiXAdRaL49J8nBlh0F_5MQ97Z1QNTUuTSMvksmEmxan3d65D6ATmQO1ig58W52u_EmQ1GM$> > > [2] http://thelastpickle.com/tlp-stress [thelastpickle.com] > <https://urldefense.com/v3/__http:/thelastpickle.com/tlp-stress__;!OYIaWQQGbnA!ZhiXAdRaL49J8nBlh0F_5MQ97Z1QNTUuTSMvksmEmxan3d65D6ATmQO1ig58W52uuCUZYKw$> > > > > Jon > > > > > > > > > > On Mon, Oct 21, 2019 at 10:24 AM Reid Pinchback < > rpinchb...@tripadvisor.com> wrote: > > An i3x large has 30.5 gb of RAM but you’re using less than 4gb for C*. So > minus room for other uses of jvm memory and for kernel activity, that’s > about 25 gb for file cache. You’ll have to see if you either want a bigger > heap to allow for less frequent gc cycles, or you could save money on the > instance size. C* generates a lot of medium-length lifetime objects which > can easily end up in old gen. A larger heap will reduce the burn of more > old-gen collections. There are no magic numbers to just give because it’ll > depend on your usage patterns. > > > > *From: *Sergio > *Reply-To: *"user@cassandra.apache.org" > *Date: *Sunday, October 20, 2019 at 2:51 PM > *To: *"user@cassandra.apache.org" > *Subject: *Re: GC Tuning > https://thelastpickle.com/blog/2018/04/11/gc-tuning.html > [thelastpickle.com] > <https://urldefense.com/v3/__https:/thelastpickle.com/blog/2018/04/11/gc-tuning.html__;!OYIaWQQGbnA!ZhiXAdRaL49J8nBlh0F_5MQ97Z1QNTUuTSMvksmEmxan3d65D6ATmQO1ig58W52uwG_KUYM$> > > > > *Message from External Sender* > > Thanks for the answer. > > This is the JVM version that I have right now. > > openjdk version "1.8.0_161" > OpenJDK Runtime Environment (build 1.8.0_161-b14) > OpenJDK 64-Bit Server VM (build 25.161-b14, mixed mode) > > These are the current flags. Would you change anything in a i3x.large aws > node? > > java -Xloggc:/var/log/cassandra/gc.log > -Dcas
RE: [EXTERNAL] Re: GC Tuning https://thelastpickle.com/blog/2018/04/11/gc-tuning.html
I don’t disagree with Jon, who has all kinds of performance tuning experience. But for ease of operation, we only use G1GC (on Java 8), because the tuning of ParNew+CMS requires a high degree of knowledge and very repeatable testing harnesses. It isn’t worth our time. As a previous writer mentioned, there is usually better return on our time tuning the schema (aka helping developers understand Cassandra’s strengths). We use 16 – 32 GB heaps, nothing smaller than that. Sean Durity From: Jon Haddad Sent: Monday, October 21, 2019 10:43 AM To: user@cassandra.apache.org Subject: [EXTERNAL] Re: GC Tuning https://thelastpickle.com/blog/2018/04/11/gc-tuning.html I still use ParNew + CMS over G1GC with Java 8. I haven't done a comparison with JDK 11 yet, so I'm not sure if it's any better. I've heard it is, but I like to verify first. The pause times with ParNew + CMS are generally lower than G1 when tuned right, but as Chris said it can be tricky. If you aren't willing to spend the time understanding how it works and why each setting matters, G1 is a better option. I wouldn't run Cassandra in production on less than 8GB of heap - I consider it the absolute minimum. For G1 I'd use 16GB, and never 4GB with Cassandra unless you're rarely querying it. I typically use the following as a starting point now: ParNew + CMS 16GB heap 10GB new gen 2GB memtable cap, otherwise you'll spend a bunch of time copying around memtables (cassandra.yaml) Max tenuring threshold: 2 survivor ratio 6 I've also done some tests with a 30GB heap, 24 GB of which was new gen. This worked surprisingly well in my tests since it essentially keeps everything out of the old gen. New gen allocations are just a pointer bump and are pretty fast, so in my (limited) tests of this I was seeing really good p99 times. I was seeing a 200-400 ms pause roughly once a minute running a workload that deliberately wasn't hitting a resource limit (testing real world looking stress vs overwhelming the cluster). We built tlp-cluster [1] and tlp-stress [2] to help figure these things out. [1] https://thelastpickle.com/tlp-cluster/ [thelastpickle.com]<https://urldefense.com/v3/__https:/thelastpickle.com/tlp-cluster/__;!OYIaWQQGbnA!ZhiXAdRaL49J8nBlh0F_5MQ97Z1QNTUuTSMvksmEmxan3d65D6ATmQO1ig58W52u_EmQ1GM$> [2] http://thelastpickle.com/tlp-stress [thelastpickle.com]<https://urldefense.com/v3/__http:/thelastpickle.com/tlp-stress__;!OYIaWQQGbnA!ZhiXAdRaL49J8nBlh0F_5MQ97Z1QNTUuTSMvksmEmxan3d65D6ATmQO1ig58W52uuCUZYKw$> Jon On Mon, Oct 21, 2019 at 10:24 AM Reid Pinchback mailto:rpinchb...@tripadvisor.com>> wrote: An i3x large has 30.5 gb of RAM but you’re using less than 4gb for C*. So minus room for other uses of jvm memory and for kernel activity, that’s about 25 gb for file cache. You’ll have to see if you either want a bigger heap to allow for less frequent gc cycles, or you could save money on the instance size. C* generates a lot of medium-length lifetime objects which can easily end up in old gen. A larger heap will reduce the burn of more old-gen collections. There are no magic numbers to just give because it’ll depend on your usage patterns. From: Sergio mailto:lapostadiser...@gmail.com>> Reply-To: "user@cassandra.apache.org<mailto:user@cassandra.apache.org>" mailto:user@cassandra.apache.org>> Date: Sunday, October 20, 2019 at 2:51 PM To: "user@cassandra.apache.org<mailto:user@cassandra.apache.org>" mailto:user@cassandra.apache.org>> Subject: Re: GC Tuning https://thelastpickle.com/blog/2018/04/11/gc-tuning.html [thelastpickle.com]<https://urldefense.com/v3/__https:/thelastpickle.com/blog/2018/04/11/gc-tuning.html__;!OYIaWQQGbnA!ZhiXAdRaL49J8nBlh0F_5MQ97Z1QNTUuTSMvksmEmxan3d65D6ATmQO1ig58W52uwG_KUYM$> Message from External Sender Thanks for the answer. This is the JVM version that I have right now. openjdk version "1.8.0_161" OpenJDK Runtime Environment (build 1.8.0_161-b14) OpenJDK 64-Bit Server VM (build 25.161-b14, mixed mode) These are the current flags. Would you change anything in a i3x.large aws node? java -Xloggc:/var/log/cassandra/gc.log -Dcassandra.max_queued_native_transport_requests=4096 -ea -XX:+UseThreadPriorities -XX:ThreadPriorityPolicy=42 -XX:+HeapDumpOnOutOfMemoryError -Xss256k -XX:StringTableSize=103 -XX:+AlwaysPreTouch -XX:-UseBiasedLocking -XX:+UseTLAB -XX:+ResizeTLAB -XX:+UseNUMA -XX:+PerfDisableSharedMem -Djava.net.preferIPv4Stack=true -XX:SurvivorRatio=8 -XX:MaxTenuringThreshold=1 -XX:+UseG1GC -XX:G1RSetUpdatingPauseTimePercent=5 -XX:MaxGCPauseMillis=200 -XX:InitiatingHeapOccupancyPercent=45 -XX:G1HeapRegionSize=0 -XX:-ParallelRefProcEnabled -Xms3821M -Xmx3821M -XX:CompileCommandFile=/etc/cassandra/conf/hotspot_compiler -Dcom.sun.management.jmxremote.port=7199 -Dcom.sun.management.jmxremote.rmi.port=7199 -Dcom.sun.management.jmxremote.ssl=false -Dcom.sun.management.jmxremote.
