Re: Supporting large caches ( 4GB) in BIND?
Have you *measured* the hit rate of your current BIND resolvers with different cache sizes? How many queries per second are you trying to support? We do about 3,000 queries/second typically. I haven't measured query -rates vs cache sizes. We've had max-cache-size set to 3GB for a long time, but the process never exceeded 2GB until recent crashes prompted recompilation as 64-bit. We do around 5500 q/s at 85% cache hit rate with a CNS process of just under one Gigabyte. This is not BIND but the statistics might still be relevant. If you feel that more memory is a worthwhile use of resources then by all means go for it. Personally I wouldn't consider it until my hit rate dropped to significantly less than 70%. However, the hit rate is of course dependent on your customers and their query profile, and it is entirely possible that our two cases are significantly different (mine is from the perspective of a commercial ISP). Steinar Haug, Nethelp consulting, sth...@nethelp.no ___ bind-users mailing list bind-users@lists.isc.org https://lists.isc.org/mailman/listinfo/bind-users
Re: Supporting large caches ( 4GB) in BIND?
On Mon, Feb 15, 2010 at 06:47:45PM +0100, sth...@nethelp.no wrote: Have you *measured* the hit rate of your current BIND resolvers with different cache sizes? How many queries per second are you trying to support? We do about 3,000 queries/second typically. I haven't measured query -rates vs cache sizes. We've had max-cache-size set to 3GB for a long time, but the process never exceeded 2GB until recent crashes prompted recompilation as 64-bit. We do around 5500 q/s at 85% cache hit rate with a CNS process of just under one Gigabyte. This is not BIND but the statistics might still be relevant. Thanks, good to know. Looking at a recent snapshot, and if I'm interpreting the bind stats correctly, I'm getting a similar cache hit rate (87%). If you feel that more memory is a worthwhile use of resources then by all means go for it. Personally I wouldn't consider it until my hit rate dropped to significantly less than 70%. However, the hit rate is of course dependent on your customers and their query profile, and it is entirely possible that our two cases are significantly different (mine is from the perspective of a commercial ISP). We have gobs of unused memory , so .. I'm not sure how different our profile is (university vs commercial ISP). But the research project which is querying a very diverse set of names that may not be typically queried will probably affect the cache hit rate. --Shumon. ___ bind-users mailing list bind-users@lists.isc.org https://lists.isc.org/mailman/listinfo/bind-users
Re: Supporting large caches ( 4GB) in BIND?
On Feb 15 2010, Shumon Huque wrote: [...] But there is a hardcoded limit of 4GB (ISC_UINT32_MAX) for the max-cache-size parameter: bin/named/server.c: if (value ISC_UINT32_MAX) { cfg_obj_log(obj, ns_g_lctx, ISC_LOG_ERROR, 'max-cache-size % ISC_PRINT_QUADFORMAT d' is too large, value); result = ISC_R_RANGE; goto cleanup; } max_cache_size = (isc_uint32_t)value; Regardless of Shumon's particular case, this looks like something that is going to bite more nameservers in the next few years, as caches expand to accommodate increasing numbers of RRSIG records (which are not small). -- Chris Thompson Email: c...@cam.ac.uk ___ bind-users mailing list bind-users@lists.isc.org https://lists.isc.org/mailman/listinfo/bind-users