Guest81 | Hello | 05:06 |
---|---|---|
=== Guest81 is now known as Tom | ||
Tom | 有人吗 | 05:07 |
=== Tom is now known as Guest25885 | ||
Guest25885 | hi | 05:10 |
=== Guest81 is now known as yekjhhee | ||
=== Guest81 is now known as yekjhhee | ||
Guest81 | hi | 05:18 |
yekjhhee | hi | 05:19 |
yekjhhee | 有人吗 | 05:20 |
alexxxxey | 有 | 05:21 |
yekjhhee | 居然真的有人 | 05:21 |
yekjhhee | 感觉发现了新天地 | 05:21 |
contrun[m] | 感觉大清要亡了啊 | 05:58 |
imadper | contrun[m]: cherrot: zao. | 06:19 |
contrun[m] | imadper 大清什么时候亡啊 | 06:20 |
imadper | contrun[m]: tomorrow. | 06:20 |
contrun[m] | imadper Great 磨难要结束了 | 06:21 |
imadper | contrun[m]: How? | 06:21 |
contrun[m] | imadper 活着没劲啊 | 06:21 |
imadper | contrun[m]: Try to find a condom-friend? | 06:22 |
contrun[m] | imadper 刁夶一日不完 我一日不找 condom-friend | 06:22 |
imadper | contrun[m]: Why 刁夶 can't be your condom friend? | 06:23 |
* imadper Shitty sogou-pinyin | 06:23 | |
contrun[m] | imadper 刁明泽可以 | 06:23 |
contrun[m] | imadper rime 大法好 | 06:23 |
imadper | contrun[m]: Is it possible that hijack read/write with ptrace to implement a cache system for NFS? | 06:24 |
imadper | contrun[m]: I mean, a local arc for nfs. | 06:24 |
imadper | contrun[m]: How to handle mmap then? | 06:25 |
contrun[m] | imadper 你他妈这是要在错误的道路上越走越远 | 06:25 |
imadper | contrun[m]: More thoughts? | 06:25 |
imadper | contrun[m]: more details? | 06:26 |
contrun[m] | imadper 不是说了用 ceph 吗 | 06:26 |
imadper | contrun[m]: shitty ceph. I don't even know if ceph has a local cache | 06:27 |
contrun[m] | imadper There is no harm in trying | 06:28 |
imadper | contrun[m]: Hard to measure. | 06:28 |
imadper | contrun[m]: cache is always transparent | 06:28 |
contrun[m] | imadper 你想要达到什么效果 | 06:29 |
albet70 | 明明是多难兴邦 | 06:29 |
imadper | contrun[m]: with local zfs, arc hits 99% on our workload. | 06:29 |
imadper | contrun[m]: With 20G arc. | 06:29 |
albet70 | 2.10封路能结束吗? | 06:30 |
imadper | contrun[m]: Which provides 600K iops which 16K recordsize. | 06:30 |
contrun[m] | albet70 你可能还活在梦里面 | 06:30 |
albet70 | contrun 中国梦 | 06:30 |
contrun[m] | imadper 具体使用场景的效果 | 06:31 |
contrun[m] | 等刁夶亲自翠吧 | 06:32 |
imadper | contrun[m]: When runing filecoin action0, there is nearly zero physical disk io with arc. Arcstats shows that arc covers almost all of the read. | 06:32 |
contrun[m] | imadper 访问的文件是网络或本地 缓存在内存? | 06:33 |
imadper | contrun[m]: https://sm.ms/image/DosbV4drQaPtyqw currently, in local. However I need to move it to lan. | 06:34 |
contrun[m] | imadper 缓存的内存不是 kernel 的 buffer 你应该搜 fs-agnostic buffer | 06:35 |
imadper | contrun[m]: I just need ceph or nfs have similar cache. | 06:36 |
imadper | contrun[m]: And if ceph has built-in memory cache, I'd use it directly. | 06:36 |
albet70 | 都啥时候了,还整天来些心灵鸡汤的新闻, | 06:37 |
albet70 | 还有那些没事蹭热点的院士 | 06:37 |
contrun[m] | imadper ceph 和 nfs 的一般缓存是指的缓存到本地硬盘啊 | 06:37 |
contrun[m] | albet70 大清要亡了 | 06:38 |
albet70 | 这次被人实名举报,干脆这个锅就让武汉病毒研究所和中科院那些院士背了吧,不杀不足以平民恨 | 06:38 |
imadper | contrun[m]: that's why I need to hijack read/write with ptrace. | 06:38 |
imadper | contrun[m]: To reduce hardware cost, we have no nvme on ever worker. | 06:39 |
contrun[m] | albet70 这个显然是无稽之谈 | 06:39 |
imadper | s/er/ery/ | 06:39 |
contrun[m] | imadper linux 的 buffer 是怎么触发的?“ | 06:40 |
albet70 | 管它是真是假,总有人需要背锅吧 | 06:40 |
imadper | contrun[m]: readahead? | 06:40 |
contrun[m] | imadper 调 linux buffer 才是正确的方向 | 06:41 |
albet70 | 平时拿皇粮的,此时就是他们为国捐躯的时候 | 06:41 |
contrun[m] | albet70 让亲自指挥 亲自部署的刁夶上 | 06:41 |
imadper | contrun[m]: is linux buffer a LRU? | 06:42 |
contrun[m] | albet70 来一个罪己诏 那是坠吼的 | 06:42 |
albet70 | 网上都说这次问题这么大,就是因为年前那个女的开车碾压了龙脉 | 06:42 |
albet70 | 我感觉这个说法非常科学 | 06:43 |
albet70 | 应该把这个女的全族祭祀了 | 06:43 |
contrun[m] | imadper 母鸡啊 | 06:43 |
contrun[m] | albet70 大帝危矣 | 06:44 |
albet70 | 你想想龙脉那是什么地方,竟然被这种妖精开车碾压,这个国运都被搞没了 | 06:45 |
imadper | albet70: Agree with you. | 06:45 |
contrun[m] | albet70 你国国运不早就没了 刁夶都能上 | 06:47 |
contrun[m] | imadper 你可以看下 数据库是怎么优化的 数据库和你的使用场景很像 | 06:51 |
contrun[m] | imadper 不行 数据库可能没有用比较通用的缓存方案 rocksdb 好像是有 writeahead log 数据本身就存在内存 到了一定大小就 flush 到硬盘 | 06:59 |
imadper | contrun[m]: yep, that's how lsm tree works. | 07:00 |
imadper | contrun[m]: However, I need it for read. | 07:01 |
albet70 | 北京下雪了? | 07:01 |
albet70 | 天津也下雪了? | 07:02 |
imadper | contrun[m]: https://docs.ceph.com/docs/giant/rbd/rbd-config-ref/ | 07:02 |
contrun[m] | albet70 下了 雪地里一个小画家都没有 | 07:02 |
contrun[m] | imadper 不错 | 07:02 |
contrun[m] | imadper 数据库读取硬盘文件也是有缓存的 不知道是怎么做的 | 07:05 |
imadper | contrun[m]: LRU or MFU | 07:06 |
imadper | contrun[m]: arc is the combin of LRU and MFU | 07:06 |
imadper | contrun[m]: it adjust the size of LRU and MFU automaticlly to fix kinds of workload. | 07:07 |
imadper | contrun[m]: https://sm.ms/image/dEsAzRTiabnBMlg in my case, MFU works well. | 07:09 |
contrun[m] | imadper 你需要的是一个 tmpfs 作为前端 nfs or whatever 作为后端的 https://bcachefs.org/ 还有两个问题 bcachefs 有没有 failure recovery tmpfs 要是满了 bcachefs 会怎么办 | 07:16 |
imadper | contrun[m]: IIRC bcachefs works at block layer, | 07:17 |
contrun[m] | https://en.wikipedia.org/wiki/CacheFS | 07:17 |
imadper | contrun[m]: cachefs uses another fs as cache. not memory. | 07:17 |
contrun[m] | tmpfs | 07:18 |
imadper | contrun[m]: https://sm.ms/image/ZNzGfU8Ee7OKstu | 07:19 |
imadper | contrun[m]: https://www.redhat.com/archives/linux-cachefs/2011-October/msg00005.html | 07:20 |
imadper | contrun[m]: :-) | 07:20 |
contrun[m] | imadper 总有一款 cachefs 适合你 | 07:20 |
imadper | contrun[m]: "You cannot use tmpfs because it doesn't support xattr needed for fscahe." | 07:20 |
imadper | contrun[m]: I would give ceph a try. | 07:22 |
imadper | contrun[m]: However all hdds are managed by zfs now. I can't destory zpool now. | 07:23 |
contrun[m] | imadper 是啊 你他妈为什么要瞎几把折腾呢 | 07:23 |
imadper | contrun[m]: just know that ceph has a local memory cache. | 07:23 |
imadper | contrun[m]: and still not sure if it's a mfu. | 07:24 |
contrun[m] | imadper premature optimization | 07:24 |
contrun[m] | imad | 07:24 |
contrun[m] | imadper 又不是不能用 | 07:24 |
imadper | contrun[m]: Just don't want to take too much time on testing ceph/tmpfs/bcachefs | 07:25 |
contrun[m] | imadper 你的问题加内存可以解决? | 07:25 |
imadper | contrun[m]: Nope, The program is alread runing. The job I'm doing now is find out the fastest storage solution. | 07:25 |
imadper | contrun[m]: which means, I'm in optimization phase now. | 07:25 |
imadper | contrun[m]: It's not "premature" | 07:26 |
contrun[m] | imadper 牛逼 | 07:26 |
imadper | contrun[m]: Increasing hardware cost is possible iff I have a sane caching system to use the memory. | 07:26 |
imadper | contrun[m]: So I'm finding the `sane caching system` | 07:27 |
contrun[m] | imadper 改个缓存策略会这么难? | 07:28 |
imadper | contrun[m]: how? | 07:28 |
imadper | contrun[m]: Show me the command. | 07:29 |
contrun[m] | imadper 没有我跳楼 你举证我 跳楼 | 07:29 |
imadper | contrun[m]: you jump, I jump | 07:29 |
contrun[m] | imadper 别 你跳了 你老婆怎么办 你们这些人啊 怎么一点都不顾家呢 | 07:31 |
imadper | contrun[m]: jump jump more health | 07:31 |
contrun[m] | imadper jump jump healthier | 07:32 |
imadper | contrun[m]: .... Okay | 07:32 |
lzkj | kubuntu用tightvncserver,键盘输入乱码是咋回事儿啊 | 08:01 |
lzkj | 完全是乱的😂 | 08:01 |
lzkj | xrdp就更不行,连接上 就显示个鼠标箭头…… | 08:12 |
lzkj | 有么有然滴嘛? | 08:22 |
imadper | ... | 08:32 |
=== sixand1 is now known as sixand | ||
=== sixand1 is now known as sixand | ||
=== sixand1 is now known as sixand | ||
lzkj | 好困啊 | 16:03 |
Generated by irclog2html.py 2.7 by Marius Gedminas - find it at mg.pov.lt!