Questions:
- Are you discarding Gin's log output? Writing stderr/stdout *to a terminal *can be a major bottleneck. - I can't see where you've allowed more ephemeral ports? By default it's around ~28k ephemeral ports on most Linux distros; you may wish to increase this. - On a i7 4770k w/ 16GB RAM (SSD, but there's no disk IO here, so makes no difference) - I can realize about 45k req/s with stdout+stderr re-directed to /dev/null, talking to Redis over a Unix socket instead of TCP. You should profile your application to see where it's spending the most time waiting, otherwise we are both ultimately guessing. On Monday, July 11, 2016 at 12:18:19 AM UTC-7, desaia...@gmail.com wrote: > > Hi Matt > > Thanks for the reply > > Test is done on different server on the same network > Application using "gopkg.in/redis.v3" as plugin so TCP > Pool size = 100000 > > Below is the simple program storing "Hello World" small data compare to > what I use to store in my actual application where Redis not even > supporting 4k reads/sec even after using snapply and probuffing compression > > Redis-benchmark utility shows 80 k reads etc > > //Output - "Hello World" http://server:8001/testkey > > package main > > import ( > "gopkg.in/redis.v3" > "github.com/gin-gonic/gin" > ) > > var redisClient *redis.Client > func main() { > redisClient = redis.NewClient(&redis.Options{ > Addr: "localhost:6379", > Password: "", > DB: 0, > PoolSize: 100000, > }) > > //Gin Code > gin.SetMode(gin.ReleaseMode) > r := gin.Default() > //r.Use(Proton.Gzip(Proton.BestSpeed)) > > r.GET("/:key", func(c *gin.Context) { > key := c.Param("key") > value, _ := redisClient.Get(key).Result() > c.JSON(200, gin.H{"response": value}) > }) > r.Run(":8001") > } > > $ wrk -t1 -c500 http://server:8001/testkey > Running 10s test @ http://server:8001/testkey > 1 threads and 500 connections > Thread Stats Avg Stdev Max +/- Stdev > Latency 15.24ms 16.99ms 296.58ms 90.73% > Req/Sec 28.84k 3.20k 35.90k 87.88% > 288461 requests in 10.08s, 41.81MB read > Requests/sec: 28606.13 > Transfer/sec: 4.15MB > > $ wrk -t4 -c500 http://server:8001/testkey > Running 10s test @ http://server:8001/testkey > 4 threads and 500 connections > Thread Stats Avg Stdev Max +/- Stdev > Latency 19.41ms 17.19ms 306.65ms 68.10% > Req/Sec 7.08k 1.00k 8.91k 73.75% > 281946 requests in 10.03s, 40.87MB read > Requests/sec: 28098.00 > Transfer/sec: 4.07MB > > $ wrk -t10 -c500 http://server:8001/testkey > Running 10s test @ http://server:8001/testkey > 10 threads and 500 connections > Thread Stats Avg Stdev Max +/- Stdev > Latency 19.51ms 18.30ms 316.32ms 66.97% > Req/Sec 2.86k 349.38 3.88k 78.39% > 286221 requests in 10.07s, 41.49MB read > Requests/sec: 28413.52 > Transfer/sec: 4.12MB > > Thanks for your help > > Rgds, > > Abhi > > > On Monday, July 11, 2016 at 4:08:36 AM UTC+5:30, Matt Silverlock wrote: >> >> >> - Are your wrk threads starving your Go program? (why 10 threads? Why >> not 4 for wrk, 4 for Go?) >> - How are you connecting to Redis? (TCP? Unix socket?) >> - What size pool have you set for Redis? >> - Show us your code. >> >> This is likely a classic example of where framework 'benchmarks' are >> completely divorced from reality. >> >> >> On Sunday, July 10, 2016 at 4:55:56 AM UTC-7, desaia...@gmail.com wrote: >>> >>> I am getting only - 29 k req/sec, can you please help me as Redis >>> document says it support 80 k+ req/sec reads not sure if I am doing any >>> wrong >>> >>> Machine - 8 core with 57 gb ram + ssd >>> Go - 1.6.2 linux/amd64 >>> Ubuntu - 15.10 >>> DB - Redis >>> Plugin - gopkg.in/redis.v3 >>> Key - only "hello" value >>> Webservice - Gin framework >>> >>> Sysctl - >>> >>> file-max = 5752905 >>> file-nr = 1888 0 5752905 >>> >>> net.core.rmem_max = 134217728 >>> >>> net.core.wmem_max = 134217728 >>> >>> net.ipv4.tcp_rmem = 4096 87380 67108864 >>> >>> net.ipv4.tcp_wmem = 4096 65536 67108864 >>> >>> net.core.netdev_max_backlog = 250000 >>> >>> net.ipv4.tcp_congestion_control=htcp >>> >>> net.ipv4.tcp_mtu_probing=1 >>> >>> ulimits - 200000 >>> >>> >>> >>> wrk -t10 -c500 http://xx/testkey >>> >>> Running 10s test @ http://xx/testkey >>> >>> 10 threads and 500 connections >>> >>> Thread Stats Avg Stdev Max +/- Stdev >>> >>> Latency 18.84ms 16.90ms 309.98ms 58.24% >>> >>> Req/Sec 2.93k 382.13 4.46k 76.62% >>> >>> 293282 requests in 10.08s, 42.51MB read >>> >>> Requests/sec: 29090.62 >>> >>> Transfer/sec: 4.22MB >>> >>> >>> Thanks, >>> >>> >>> Abhi >>> >>> >>> -- You received this message because you are subscribed to the Google Groups "golang-nuts" group. To unsubscribe from this group and stop receiving emails from it, send an email to golang-nuts+unsubscr...@googlegroups.com. For more options, visit https://groups.google.com/d/optout.
