Sunday, January 24, 2010

Hugepage overview

Hugepage overview
==========
Linux / Storage: Memory – Huge Pages Overview

A page is really virtual memory which is managed by the Translation Lookaside Buffers(TLB) in the CPU. The TLB controls the mapping of the virtual memory pages to physical memory addresses. In doing so, it bypasses the kernel virtual memory manager.

Per RedHat,

    The TLB is a limited hardware resource, so utilising a huge amount of physical memory with the default page size consumes the TLB and adds processing overhead – many pages of size 4096 Bytes equates to many TLB resources consumed.

This is where Huge Pages come in. Pages are created at a larger size than the default 4096 bytes, and each page will consume only one TLB resource. So you can see this is a huge benefit. Using Huge Pages decrease the number of TLB resources required.

Side Affect
This is great, depending on what you are trying to accomplish. Once the physical memory is mapped to a Huge Page, it can no longer be used for “normal” memory allocation. This is because the memory is no longer mapped by the kernel virtual memory manager. The applications that you want to dedicate the Huge Pages to have to have support for them.

Benefit
So here is the best part of Huge Pages. It is dedicated memory to be used by only applications that request them. This dedicated memory is stored in physical RAM and will NEVER be swapped out! Thus, guaranteeing a level of performance. When memory is swapped to disk, it’s a lot slower than RAM and decreases the performance of the process(s)/program(s) gets pushed there.

Now knowing that Huge Pages are stored in RAM, this also means that the allocated RAM is dedicated. This is a little bit redundant to the above, but I want to make sure this point is clear.

Example: If a server has 8gigs of RAM and 5gigs are allocated to Huge Pages, that only leaves 3gigs for all other processes, programs, and underlining operating system to use.

Below shows my Linux desktop that has the default page size of 4096 set

user@workstation:~$ cat /proc/meminfo | grep Huge
HugePages_Total:       0
HugePages_Free:        0
HugePages_Rsvd:        0
HugePages_Surp:        0
Hugepagesize:       4096 kB

So as you can see, I have no Huge Pages reserved or in use. The next example is from a production Oracle database server

[root@OracleServer1 ~]# cat /proc/meminfo | grep Huge
HugePages_Total: 12200
HugePages_Free:     85
Hugepagesize:     2048 kB

So to calculate the space dedicated to Huge Pages from above, it is 12,200 x 2048 kB which gives us

===================================




Performance tuning: HugePages in Linux


Recently we quickly and efficiently resolved a major performance issue with one of our New York clients. In this blog, I will discuss about this performance issue and its solution.
Problem statement

The client’s central database was intermittently freezing because of high CPU usage, and their business severely affected. They had already worked with vendor support and the problem was still unresolved.
Symptoms

Intermittent High Kernel mode CPU usage was the symptom. The server hardware was 4 dual-core CPUs, hyperthreading enabled, with 20GB of RAM, running a Red Hat Linux OS with a 2.6 kernel.

During this database freeze, all CPUs were using kernel mode and the database was almost unusable. Even log-ins and simple SQL such as SELECT * from DUAL; took a few seconds to complete. A review of the AWR report did not help much, as expected, since the problem was outside the database.

Analyzing the situation, collecting system activity reporter (sar) data, we could see that at 08:32 and then at 8:40, CPU usage in kernel mode was almost at 70%. It is also interesting to note that, SADC (sar data collection) also suffered from this CPU spike, since SAR collection at 8:30 completed two minutes later at 8:32, as shown below.

A similar issue repeated at 10:50AM:

07:20:01 AM CPU   %user     %nice   %system   %iowait     %idle
07:30:01 AM all    4.85      0.00     77.40      4.18     13.58
07:40:01 AM all   16.44      0.00      2.11     22.21     59.24
07:50:01 AM all   23.15      0.00      2.00     21.53     53.32
08:00:01 AM all   30.16      0.00      2.55     15.87     51.41
08:10:01 AM all   32.86      0.00      3.08     13.77     50.29
08:20:01 AM all   27.94      0.00      2.07     12.00     58.00
08:32:50 AM all   25.97      0.00     25.42     10.73     37.88 <--
08:40:02 AM all   16.40      0.00     69.21      4.11     10.29 <--
08:50:01 AM all   35.82      0.00      2.10     12.76     49.32
09:00:01 AM all   35.46      0.00      1.86      9.46     53.22
09:10:01 AM all   31.86      0.00      2.71     14.12     51.31
09:20:01 AM all   26.97      0.00      2.19      8.14     62.70
09:30:02 AM all   29.56      0.00      3.02     16.00     51.41
09:40:01 AM all   29.32      0.00      2.62     13.43     54.62
09:50:01 AM all   21.57      0.00      2.23     10.32     65.88
10:00:01 AM all   16.93      0.00      3.59     14.55     64.92
10:10:01 AM all   11.07      0.00     71.88      8.21      8.84
10:30:01 AM all   43.66      0.00      3.34     13.80     39.20
10:41:54 AM all   38.15      0.00     17.54     11.68     32.63 <--
10:50:01 AM all   16.05      0.00     66.59      5.38     11.98 <--
11:00:01 AM all   39.81      0.00      2.99     12.36     44.85

Performance forensic analysis

The client had access to a few tools, none of which were very effective. We knew that there is excessive kernel mode CPU usage. To understand why, we need to look at various metrics at 8:40 and 10:10.

Fortunately, sar data was handy. Looking at free memory, we saw something odd. At 8:32, free memory was 86MB; at 8:40 free memory climbed up to 1.1GB. At 10:50 AM free memory went from 78MB to 4.7GB. So, within a range of ten minutes, free memory climbed up to 4.7GB.

07:40:01 AM kbmemfree kbmemused  %memused kbbuffers  kbcached
07:50:01 AM    225968  20323044     98.90    173900   7151144
08:00:01 AM    206688  20342324     98.99    127600   7084496
08:10:01 AM    214152  20334860     98.96    109728   7055032
08:20:01 AM    209920  20339092     98.98     21268   7056184
08:32:50 AM     86176  20462836     99.58      8240   7040608
08:40:02 AM   1157520  19391492     94.37     79096   7012752
08:50:01 AM   1523808  19025204     92.58    158044   7095076
09:00:01 AM    775916  19773096     96.22    187108   7116308
09:10:01 AM    430100  20118912     97.91    218716   7129248
09:20:01 AM    159700  20389312     99.22    239460   7124080
09:30:02 AM    265184  20283828     98.71    126508   7090432
10:41:54 AM     78588  20470424     99.62      4092   6962732  <--
10:50:01 AM   4787684  15761328     76.70     77400   6878012  <--
11:00:01 AM   2636892  17912120     87.17    143780   6990176
11:10:01 AM   1471236  19077776     92.84    186540   7041712

This tells us that there is a correlation between this CPU usage and the increase in free memory. If free memory goes from 78MB to 4.7GB, then the paging and swapping daemons must be working very hard. Of course, releasing 4.7GB of memory to the free pool will sharply increase paging/swapping activity, leading to massive increase in kernel
mode CPU usage. This can lead to massive kernel mode CPU usage.

Most likely, much of SGA pages also can be paged out, since SGA is not locked in memory.
Memory breakdown

The client’s question was, if paging/swapping is indeed the issue, then what is using all my memory? It’s a 20GB server, SGA size is 10GB and no other application is running. It gets a few hundred connections at a time, and PGA_aggregated_target is set to 2GB. So why would it be suffering from memory starvation? If memory is the issue, how can there be 4.7GB of free memory at 10:50AM?

Recent OS architectures are designed to use all available memory. Therefore, paging daemons doesn’t wake up until free memory falls below a certain threshold. It’s possible for the free memory to drop near zero and then climb up quickly as the paging/swapping daemon starts to work harder and harder. This explains why free memory went down to 78MB and rose to 4.7GB 10 minutes later.

What is using my memory though? /proc/meminfo is useful in understanding that, and it shows that the pagetable size is 5GB. How interesting!

Essentially, pagetable is a mapping mechanism between virtual and physical address. For a default OS Page size of 4KB and a SGA size of 10GB, there will be 2.6 Million OS pages just for SGA alone. (Read wikipedia’s entry on page table for more information about page tables.) On this server, there will be 5 million OS pages for 20GB total memory. It will be an enormous workload for the paging/swapping daemon to manage all these pages.

cat /proc/meminfo

MemTotal:     20549012 kB
MemFree:        236668 kB
Buffers:         77800 kB
Cached:        7189572 kB
...
PageTables:    5007924 kB  <--- 5GB!
...
HugePages_Total:     0
HugePages_Free:      0
Hugepagesize:     2048 kB

HugePages

Fortunately, we can use HugePages in this version of Linux. There are couple of important benefits of HugePages:

   1. Page size is set 2MB instead of 4KB
   2. Memory used by HugePages is locked and cannot be paged out.

With a pagesize of 2MB, 10GB SGA will have only 5000 pages compared to 2.6 million pages without HugePages. This will drastically reduce the page table size. Also, HugeTable memory is locked and so SGA can’t be swapped out. The working set of buffers for the paging/swapping daemon will be smaller.

To setup HugePages, the following changes must be completed:

   1. Set the vm.nr_hugepages kernel parameter to a suitable value. In this case, we decided to use 12GB and set the parameter to 6144 (6144*2M=12GB). You can run:

      echo 6144 > /proc/sys/vm/nr_hugepages

      or

      sysctl -w vm.nr_hugepages=6144

      Of course, you must make sure this set across reboots too.
   2. The oracle userid needs to be able to lock a greater amount of memory. So, /etc/securities/limits.conf must be updated to increase soft and hard memlock values for oracle userid.

      oracle          soft    memlock        12582912
      oracle          hard   memlock        12582912

After setting this up, we need to make sure that SGA is indeed using HugePages. The value, (HugePages_Total- HugePages_Free)*2MB will be the approximate size of SGA (or it will equal the shared memory segment shown in the output of ipcs -ma).

cat /proc/meminfo |grep HugePages
HugePages_Total:  6144
HugePages_Free:   1655 <-- Free pages are less than total pages.
Hugepagesize:     2048 kB

Summary

Using HugePages resolved our client’s performance issues. The PageTable size also went down to a few hundred MB. If your database is running in Linux and has HugePages capability, there is no reason not to use it.

2 comments:

Anonymous said...

pleasure to find such a good artical! please keep update!! ........................................

dharmendra said...

My question is: if page size will be more and less data will be inside pages then won't it increase management overhead. Let's take an example, suppose we have approx 50 KB data in lots of pages and if pages size is 2MB (Due to Huge Page setting) then more fragmentation will be there in the RAM resulting into poor performance. Please let me know your thoughts on this. If that's the case indeed then how to control it.