From: Michael Vrable <mvrable@cs.ucsd.edu>
Date: Tue, 5 Apr 2011 20:50:08 +0000 (-0700)
Subject: Merge branch 'master' of git+ssh://niniel.sysnet.ucsd.edu/home/mvrable/local/bluesky
X-Git-Url: http://git.vrable.net/?p=bluesky.git;a=commitdiff_plain;h=5ab537ebebcb0abef91bd032c822d466a2e898e1;hp=d775a59d76e74416f162933dbe39e81ad266da60

Merge branch 'master' of git+ssh://niniel.sysnet.ucsd.edu/home/mvrable/local/bluesky
---

diff --git a/bluesky/cache.c b/bluesky/cache.c
index b20960c..8cd2fb3 100644
--- a/bluesky/cache.c
+++ b/bluesky/cache.c
@@ -433,27 +433,32 @@ static int compute_pressure(BlueSkyFS *fs)
     if (g_atomic_int_get(&fs->cache_dirty)
                 + g_atomic_int_get(&fs->cache_log_dirty)
            > bluesky_watermark_high_dirty)
+    {
+        g_print("pressure: too much dirty data (2)\n");
         return 2;
+    }
 
     /* Too much uncommitted data in the journal on disk, not yet flushed to the
      * cloud? */
-    printf("Dirty journals: %d to %d\n",
-           fs->log->journal_watermark, fs->log->seq_num);
+    /*printf("Dirty journals: %d to %d\n",
+           fs->log->journal_watermark, fs->log->seq_num);*/
     int dirty_limit;
     dirty_limit = bluesky_options.cache_size / (LOG_SEGMENT_SIZE / 1024) / 2;
     int dirty_journals = fs->log->seq_num - fs->log->journal_watermark + 1;
     if (dirty_journals > 1 && dirty_journals >= dirty_limit) {
-        printf("Too many dirty journals (%d >= %d)\n",
+        printf("pressure: too many dirty journals (%d >= %d) (2)\n",
                dirty_journals, dirty_limit);
         return 2;
     }
 
     /* LEVEL 1 */
-    if (g_atomic_int_get(&fs->cache_dirty) < bluesky_watermark_medium_dirty)
+    if (g_atomic_int_get(&fs->cache_dirty) > bluesky_watermark_medium_dirty) {
+        g_print("pressure: too much dirty data (1)\n");
         return 1;
+    }
 
     if (dirty_journals > 1 && dirty_journals > dirty_limit / 2) {
-        printf("Many dirty journals (%d), should start writeback\n",
+        printf("pressure: many dirty journals (%d), should start writeback (1)\n",
                dirty_journals);
         return 1;
     }
diff --git a/bluesky/cleaner.c b/bluesky/cleaner.c
index 5f241d4..a892e72 100644
--- a/bluesky/cleaner.c
+++ b/bluesky/cleaner.c
@@ -115,7 +115,8 @@ static BlueSkyCleanerItem *bluesky_cleaner_find_checkpoint(BlueSkyFS *fs)
     if (last_segment == NULL)
         return NULL;
 
-    g_print("Last cloud log segment: %s\n", last_segment);
+    g_print("Last cloud log segment: %s (processed up to %d)\n",
+            last_segment, fs->log_state->latest_cleaner_seq_seen);
     int seq = atoi(last_segment + 13);
     g_free(last_segment);
 
@@ -156,7 +157,7 @@ static BlueSkyCleanerItem *bluesky_cleaner_find_checkpoint(BlueSkyFS *fs)
 
     BlueSkyCleanerItem *checkpoint = bluesky_cleaner_deserialize(data);
     checkpoint->location.directory = BLUESKY_CLOUD_DIR_CLEANER;
-    checkpoint->location.directory = seq;
+    checkpoint->location.sequence = seq;
     bluesky_string_unref(data);
 
     return checkpoint;
@@ -302,6 +303,12 @@ void bluesky_cleaner_merge(BlueSkyFS *fs)
         return;
     }
 
+    if (checkpoint->type != LOGTYPE_CHECKPOINT) {
+        g_warning("Last cleaner object not a checkpoint; cleaning probably in progress.");
+        bluesky_cleaner_item_free(checkpoint);
+        return;
+    }
+
     /* Iterate over each of the inode map sections in the checkpoint */
     for (int i = 0; i < checkpoint->links->len; i++) {
         BlueSkyCleanerLink *link = &g_array_index(checkpoint->links,
@@ -331,7 +338,7 @@ void bluesky_cleaner_merge(BlueSkyFS *fs)
         bluesky_cleaner_item_free(imap);
     }
 
-    fs->log_state->latest_cleaner_seq_seen = checkpoint->location.directory;
+    fs->log_state->latest_cleaner_seq_seen = checkpoint->location.sequence;
     bluesky_cleaner_item_free(checkpoint);
 }
 
diff --git a/microbench/32kreadbench.py b/microbench/32kreadbench.py
new file mode 100755
index 0000000..1c07cd6
--- /dev/null
+++ b/microbench/32kreadbench.py
@@ -0,0 +1,24 @@
+#!/usr/bin/python
+
+import os, random, sys, time
+
+def read_files(files, rate=1.0):
+    while True:
+        f = random.sample(files, 1)[0]
+
+        start = time.time()
+        fp = open(f, 'r')
+        blocks = (16 << 20) / 32768
+        offset = random.randrange(blocks) * 32768
+        fp.seek(offset)
+        fp.read(32768)
+        print time.time() - start
+        fp.close()
+
+if __name__ == '__main__':
+    all_files = []
+    for (path, dirs, files) in iter(os.walk(".")):
+        for f in files:
+            all_files.append(os.path.join(path, f))
+    print len(all_files), "files total"
+    read_files(all_files)
diff --git a/microbench/run-synread.sh b/microbench/run-synread.sh
index d3d8cc9..b247b5b 100755
--- a/microbench/run-synread.sh
+++ b/microbench/run-synread.sh
@@ -27,10 +27,10 @@ run_synbench() {
         BLUESKY_RUN_NAME=$basename-$(($s / 1024))-c1
         run_cmd $BENCHER run-synread
 
-        SYNREAD_OUTSTANDING=8
-        SYNREAD_PROCS=2
-        BLUESKY_RUN_NAME=$basename-$(($s / 1024))-c16
-        run_cmd $BENCHER run-synread
+        #SYNREAD_OUTSTANDING=8
+        #SYNREAD_PROCS=2
+        #BLUESKY_RUN_NAME=$basename-$(($s / 1024))-c16
+        #run_cmd $BENCHER run-synread
     done
 
     run_cmd $PROXY stop-proxy
@@ -39,10 +39,10 @@ run_synbench() {
 }
 
 SYNREAD_DURATION=120
-for cache in 0 4 12 20 28; do
+for cache in 0 4 8 12 16 20 24 28 32; do
     BLUESKY_CACHE_SIZE=$(($cache * 1024 * 1024))
     if [ $cache -eq 0 ]; then
         BLUESKY_CACHE_SIZE=$((64 * 1024))
     fi
-    run_synbench "syntest2-${cache}G"
+    run_synbench "syntest-${cache}G"
 done
diff --git a/microbench/run-synwrite.sh b/microbench/run-synwrite.sh
index ce66b3e..bc219a8 100755
--- a/microbench/run-synwrite.sh
+++ b/microbench/run-synwrite.sh
@@ -6,7 +6,7 @@ BASEDIR=$(dirname $(which $0))
 BLUESKY_TARGET=s3:mvrable-bluesky-west
 BLUESKY_EXTRA_OPTS="BLUESKY_OPT_NO_CRYPTO=1"
 
-rates="1 2 4 8 10 12 14 16 32"
+rates="2 4 6 8 10 12 14 16 18 20 24 28 32"
 
 run_synbench() {
     basename=$(date +%Y%m%d)-$1
@@ -20,7 +20,7 @@ run_synbench() {
 
     for SYNWRITE_RATE in $rates; do
         run_cmd $BENCHER run-synwrite
-        sleep 30
+        sleep 120
     done
 
     run_cmd $PROXY stop-proxy
@@ -28,7 +28,6 @@ run_synbench() {
     wait $proxy_pid
 }
 
-#for BLUESKY_CACHE_SIZE in $((2048 * 1024)); do
-for BLUESKY_CACHE_SIZE in $((256 * 1024)); do
+for BLUESKY_CACHE_SIZE in $((256 * 1024)) $((2048 * 1024)); do
     run_synbench "write100c-$(($BLUESKY_CACHE_SIZE / 1024))M"
 done
diff --git a/results/20110317/CLEANER b/results/20110317/CLEANER
new file mode 100644
index 0000000..6044f90
--- /dev/null
+++ b/results/20110317/CLEANER
@@ -0,0 +1,83 @@
+Proxy run as:
+
+BLUESKY_CACHE_SIZE=$((8 * 1024 * 1024)) BLUESKY_TARGET=s3:mvrable-bluesky-west BLUESKY_STATS_OUT=/export2/20110317-cleaner.stats /scratch/bluesky.git/nfs3/nfsproxy
+
+Create 1 GB of files:
+for i in {0..1023}; do dd if=/dev/zero of=file-$i bs=1M count=1; done
+
+Thu Mar 17 21:26:36 PDT 2011
+Log files written: up to log-00000000-00000261
+
+Thu Mar 17 21:29:00 PDT 2011
+Dirty 25% of files with cleanbench.py
+Log files written up to log-00000000-00000329
+
+Thu Mar 17 21:29:45 PDT 2011
+Run cleaner, completes around Thu Mar 17 21:32:35 PDT 2011
+max log segments: log-00000000-00000329, log-00000001-00000030
+
+==============================================================================
+
+Proxy run as:
+
+BLUESKY_CACHE_SIZE=$((4 * 1024 * 1024)) BLUESKY_TARGET=s3:mvrable-bluesky-west BLUESKY_STATS_OUT=/export2/20110317-cleaner.stats /scratch/bluesky.git/nfs3/nfsproxy
+
+Create 1 GB of files:
+for i in {0..1023}; do dd if=/dev/zero of=file-$i bs=1M count=1; done
+Fri Mar 18 00:44:58 PDT 2011
+
+Log files written: up to log-00000000-00002066
+
+Fri Mar 18 00:49:14 PDT 2011
+Dirty 25% of files with cleanbench.py
+Log files written up to log-00000000-00002585
+
+Run cleaner
+max log segments: log-00000000-00002585, log-00000001-00000235
+Fri Mar 18 01:15:04 PDT 2011
+
+==============================================================================
+
+Proxy run as:
+
+BLUESKY_CACHE_SIZE=$((4 * 1024 * 1024)) BLUESKY_TARGET=s3:mvrable-bluesky-west BLUESKY_STATS_OUT=/export2/20110318-cleaner.stats /scratch/bluesky.git/nfs3/nfsproxy
+
+Create 1 GB of files:
+for i in {0..8191}; do dd if=/dev/zero of=file-$i bs=1M count=1; done
+Fri Mar 18 01:47:03 PDT 2011
+
+Log files written: up to log-00000000-00002067
+
+Dirty 25% of files with cleanbench.py
+Log files written up to log-00000000-00002586
+Fri Mar 18 01:50:25 PDT 2011
+
+Run cleaner
+Fri Mar 18 01:51:20 PDT 2011
+writeout shortly before Fri Mar 18 02:16:55 PDT 2011
+max log segments: 
+
+==============================================================================
+
+Proxy run as:
+
+BLUESKY_CACHE_SIZE=$((1024 * 1024)) BLUESKY_TARGET=s3:mvrable-bluesky-west BLUESKY_STATS_OUT=/export2/20110318-cleaner.stats /scratch/bluesky.git/nfs3/nfsproxy
+
+Create 1 GB of files:
+for i in {0..1023}; do dd if=/dev/zero of=file-$i bs=1M count=1; done
+Fri Mar 18 01:47:03 PDT 2011
+
+Log files up to: log-00000000-00000265 log-00000001-00000002
+
+Fri Mar 18 09:18:02 PDT 2011
+Dirty 25% of files with cleanbench.py
+Fri Mar 18 09:19:30 PDT 2011
+Log files up to: log-00000000-00000336 log-00000001-00000002
+
+Run cleaner
+Fri Mar 18 09:20:22 PDT 2011
+Fri Mar 18 09:24:23 PDT 2011
+max log segments: log-00000000-00000337 log-00000001-00000036
+
+Re-run cleaner (to check for deletable segments):
+Fri Mar 18 09:25:48 PDT 2011
diff --git a/results/20110317/read-latencies.data b/results/20110317/read-latencies.data
new file mode 100644
index 0000000..57fde23
--- /dev/null
+++ b/results/20110317/read-latencies.data
@@ -0,0 +1,33 @@
+# Four-way comparison of read latencies
+#
+# First column is attempted ops per second, then:
+#   Native NFS
+#   Bluesky NFS
+#   Native Samba
+#   Bluesky CIFS
+50	0.6	0.6	1.12	0.46
+100	2.41	7.59	2.01	7.3
+150	2.91	6.26	2.53	7.33
+200	3.23	8.41	2.96	10.09
+250	4.4	14.58	3.54	13.12
+300	4.28	13.14	3.55	11.38
+350	4.89	12.05	5.39	13.22
+400	6.27	13.81	4.7	14.88
+450	8.11	15.96	4.55	14.47
+500	10.22	21.02	5.32	16.12
+550	12.07	22.01	6.73	18.46
+600	13.39	25.23	7.89	17.12
+650	18.83	27.13	9.22	20.57
+700	18.22	33.35	11.29	24.76
+750	18.31	34.67	10.85	25.48
+800	19.03	37.45	11.14
+850	20.26	37.22	13.21
+900	24.35	41.63	10.58
+950	24.05	45.03	11.74
+1000	22.29	49.88	16.19
+1050	27.05	51.77
+1100	24.07	59.15
+1150	24.67	57.78
+1200	26.07	60.72
+1250	28.53	67.99
+
diff --git a/results/20110318-synread/1024k-c1.data b/results/20110318-synread/1024k-c1.data
new file mode 100644
index 0000000..7b16114
--- /dev/null
+++ b/results/20110318-synread/1024k-c1.data
@@ -0,0 +1,9 @@
+0	4.18348623853	4.18348623853	239
+4	5.08181818182	5.08181818182	197
+8	4.03603603604	4.03603603604	247
+12	4.9009009009	4.9009009009	204
+16	5.52252252252	5.52252252252	180
+20	5.59459459459	5.59459459459	178
+24	8.31818181818	8.31818181818	119
+28	11.6272727273	11.6272727273	86
+32	21.0990990991	21.0990990991	47
diff --git a/results/20110318-synread/128k-c1.data b/results/20110318-synread/128k-c1.data
new file mode 100644
index 0000000..f812fdc
--- /dev/null
+++ b/results/20110318-synread/128k-c1.data
@@ -0,0 +1,9 @@
+0	7.6036036036	0.95045045045	131
+4	13.7207207207	1.71509009009	72
+8	8.97297297297	1.12162162162	111
+12	9.28828828829	1.16103603604	107
+16	12.1981981982	1.52477477477	81
+20	13.1891891892	1.64864864865	75
+24	16.6396396396	2.07995495495	60
+28	26.8378378378	3.35472972973	37
+32	47.036036036	5.8795045045	21
diff --git a/results/20110318-synread/32k-c1.data b/results/20110318-synread/32k-c1.data
new file mode 100644
index 0000000..30e2d07
--- /dev/null
+++ b/results/20110318-synread/32k-c1.data
@@ -0,0 +1,9 @@
+0	13.4324324324	0.419763513514	74
+4	15.027027027	0.469594594595	66
+8	16.2432432432	0.507601351351	61
+12	16.3153153153	0.509853603604	61
+16	21.2522522523	0.664132882883	47
+20	20.954954955	0.654842342342	47
+24	29.7477477477	0.929617117117	33
+28	42.9099099099	1.34093468468	23
+32	73.9459459459	2.31081081081	13
diff --git a/results/figures/specsfs.gnuplot b/results/figures/specsfs.gnuplot
index debc5d3..a085ade 100644
--- a/results/figures/specsfs.gnuplot
+++ b/results/figures/specsfs.gnuplot
@@ -13,29 +13,23 @@ set xtics nomirror
 set x2tics auto
 
 set grid
-set key top left
+set key top right
 
 set output "spec1.eps"
 set ylabel "Achieved Operations per Second"
-plot "../20110311/sfssum.20110312-native" with linespoints title "Local NFS Server", \
-     "../20110307/sfssum.20110307-ec2-west2" with linespoints title "EC2 NFS Server", \
+plot "../20110311/sfssum.20110312-native" with linespoints title "Local NFS", \
      "../20110311/sfssum.20110311-s3-west" with linespoints title "BlueSky", \
      "../20110311/sfssum.20110311-s3-west-4kfs" with linespoints title "BlueSky (4K blocks)", \
      "../20110311/sfssum.20110312-s3-west-fullseg" with linespoints title "BlueSky (full fetches)", \
-     "../20110311/sfssum.20110312-s3-west-noseg" with linespoints title "BlueSky (no segments)", \
-     "../20110311/sfssum.20110315-s3-west-noagg" with linespoints title "BlueSky (no read aggregation)", \
      "../20110311/sfssum.20110313-azure" with linespoints title "Azure"
 
 set output "spec2.eps"
 set ylabel "Operation Latency (ms)"
 set yrange [0:50]
-plot "../20110311/sfssum.20110312-native" using 1:3 with linespoints title "Local NFS Server", \
-     "../20110307/sfssum.20110307-ec2-west2" using 1:3 with linespoints title "EC2 NFS Server", \
+plot "../20110311/sfssum.20110312-native" using 1:3 with linespoints title "Local NFS", \
      "../20110311/sfssum.20110311-s3-west" using 1:3 with linespoints title "BlueSky", \
      "../20110311/sfssum.20110311-s3-west-4kfs" using 1:3 with linespoints title "BlueSky (4K blocks)", \
      "../20110311/sfssum.20110312-s3-west-fullseg" using 1:3 with linespoints title "BlueSky (full fetches)", \
-     "../20110311/sfssum.20110312-s3-west-noseg" using 1:3 with linespoints title "BlueSky (no segments)", \
-     "../20110311/sfssum.20110315-s3-west-noagg" using 1:3 with linespoints title "BlueSky (no read aggregation)", \
      "../20110311/sfssum.20110313-azure" using 1:3 with linespoints title "Azure"
 
 XMAX=1600
@@ -76,3 +70,18 @@ set output "spec-cifs2.eps"
 set ylabel "Operation Latency (ms)"
 plot "../20110227a/sfssum.20110227-samba" using 1:3 with linespoints title "Samba", \
      "../20110317/sfssum.20110317-cifs" using 1:3 with linespoints title "BlueSky"
+
+# Four-way comparison of read latencies among native/BlueSky with NFS/CIFS
+XMAX=750
+set xrange [0:XMAX]
+set x2range [0:XMAX*OPS_WSS_SCALE]
+set xtics nomirror
+set x2tics auto
+set yrange [*:*]
+
+set output "spec-read-latencies.eps"
+set ylabel "Operation Latency (ms)"
+plot "../20110317/read-latencies.data" using 1:2 with linespoints title "Native NFS", \
+     "../20110317/read-latencies.data" using 1:3 with linespoints title "BlueSky NFS", \
+     "../20110317/read-latencies.data" using 1:4 with linespoints title "Samba (CIFS)", \
+     "../20110317/read-latencies.data" using 1:5 with linespoints title "BlueSky CIFS"
diff --git a/results/figures/syn-read.gnuplot b/results/figures/syn-read.gnuplot
index af974bf..26c2795 100644
--- a/results/figures/syn-read.gnuplot
+++ b/results/figures/syn-read.gnuplot
@@ -8,28 +8,28 @@ set xrange [0:100]
 set output "syn-read-1.eps"
 set title "Single-Client Request Stream"
 set key top right
-plot "../20110316-synread/32k-c1.data" using (100*$1/32):4 with linespoints title "32 KB", \
-     "../20110316-synread/128k-c1.data" using (100*$1/32):4 with linespoints title "128 KB", \
-     "../20110316-synread/1024k-c1.data" using (100*$1/32):4 with linespoints title "1024 KB"
+plot "../20110318-synread/32k-c1.data" using (100*$1/32):4 with linespoints title "32 KB", \
+     "../20110318-synread/128k-c1.data" using (100*$1/32):4 with linespoints title "128 KB", \
+     "../20110318-synread/1024k-c1.data" using (100*$1/32):4 with linespoints title "1024 KB"
 
-set output "syn-read-16.eps"
-set title "16 Concurrent Request Streams"
-set key top left
-plot "../20110316-synread/32k-c16.data" using (100*$1/32):4 with linespoints title "32 KB", \
-     "../20110316-synread/128k-c16.data" using (100*$1/32):4 with linespoints title "128 KB", \
-     "../20110316-synread/1024k-c16.data" using (100*$1/32):4 with linespoints title "1024 KB"
+# set output "syn-read-16.eps"
+# set title "16 Concurrent Request Streams"
+# set key top left
+# plot "../20110316-synread/32k-c16.data" using (100*$1/32):4 with linespoints title "32 KB", \
+#      "../20110316-synread/128k-c16.data" using (100*$1/32):4 with linespoints title "128 KB", \
+#      "../20110316-synread/1024k-c16.data" using (100*$1/32):4 with linespoints title "1024 KB"
 
 set ylabel "Read Bandwidth (MB/s)"
 set output "syn-read-1b.eps"
 set title "Single-Client Request Stream"
 set key top left
-plot "../20110316-synread/32k-c1.data" using (100*$1/32):3 with linespoints title "32 KB", \
-     "../20110316-synread/128k-c1.data" using (100*$1/32):3 with linespoints title "128 KB", \
-     "../20110316-synread/1024k-c1.data" using (100*$1/32):3 with linespoints title "1024 KB"
+plot "../20110318-synread/32k-c1.data" using (100*$1/32):3 with linespoints title "32 KB", \
+     "../20110318-synread/128k-c1.data" using (100*$1/32):3 with linespoints title "128 KB", \
+     "../20110318-synread/1024k-c1.data" using (100*$1/32):3 with linespoints title "1024 KB"
 
-set output "syn-read-16b.eps"
-set title "16 Concurrent Request Streams"
-set key top right
-plot "../20110316-synread/32k-c16.data" using (100*$1/32):3 with linespoints title "32 KB", \
-     "../20110316-synread/128k-c16.data" using (100*$1/32):3 with linespoints title "128 KB", \
-     "../20110316-synread/1024k-c16.data" using (100*$1/32):3 with linespoints title "1024 KB"
+# set output "syn-read-16b.eps"
+# set title "16 Concurrent Request Streams"
+# set key top right
+# plot "../20110316-synread/32k-c16.data" using (100*$1/32):3 with linespoints title "32 KB", \
+#      "../20110316-synread/128k-c16.data" using (100*$1/32):3 with linespoints title "128 KB", \
+#      "../20110316-synread/1024k-c16.data" using (100*$1/32):3 with linespoints title "1024 KB"
diff --git a/results/parse-sfsres.py b/results/parse-sfsres.py
index 34c9a09..8d84ac1 100755
--- a/results/parse-sfsres.py
+++ b/results/parse-sfsres.py
@@ -28,6 +28,7 @@ def parse_date(datestr):
     return int(d.strip())
 
 def find_stats(statsdata, timestamp):
+    if statsdata is None: return (0, [0] * len(STATSDATA))
     for s in statsdata:
         if s[0] > timestamp: return (s[0], s[1:])
     return (statsdata[-1][0], statsdata[-1][1:])
@@ -121,6 +122,6 @@ if __name__ == '__main__':
         input_stats = open(sys.argv[2])
         statsdata = parse_stats(input_stats)
     except:
-        statsdata = []
+        statsdata = None
 
     parse_sfsres(input_sfsres, statsdata)