Hi PVE team
This questions is for PVE developer team:
Currently PVE 3.1 has a kernel with support for DRBD 8.3.13, but DRBD 8.4.4 is much faster and more stable
please see this link:
http://blogs.linbit.com/p/469/843-random-writes-faster/
And I say more stable and faster in my personal exprience because I use these versions in production environments and developer environments since many months ago without any problem or inconveniences.
Then, I want to apply to PVE team one of two things if is possible:
1- PVE and his kernel have support for the latest version of DRBD, or
2- PVE and his kernel have "THE CORRECT KERNEL HEADER" for use it correctly with dkms, since that i have that modify the "kernel header" in each upgrade of PVE + compile DRBD = then, is very annoying because the extra work required in each PVE node
For this support, please consider these:
1- If PVE have doubts in anything, please let me know
2- For use with DKMS the solution is add these lines in the pve Kernel header:
lines are inserted from 45 to 49 in "idr.h" file:
41 #define MAX_ID_SHIFT (sizeof(int)*8 - 1)
42 #define MAX_ID_BIT (1U << MAX_ID_SHIFT)
43 #define MAX_ID_MASK (MAX_ID_BIT - 1)
44
45 /* Leave the possibility of an incomplete final layer */
46 #define MAX_LEVEL (MAX_ID_SHIFT + IDR_BITS - 1) / IDR_BITS
47
48 /* Number of id_layer structs to leave in free list */
49 #define IDR_FREE_MAX MAX_LEVEL + MAX_LEVEL
50
51 struct idr_layer {
52 unsigned long bitmap; /* A zero bit means "space here" */
53 struct idr_layer *ary[1<<IDR_BITS];
54 int count; /* When zero, we can release it */
55 int layer; /* distance from leaf */
56 struct rcu_head rcu_head;
57 };
Please PVE team let me know about of your think about of this topic, and please take into account that linbit certified his development for several linux distributions
Please see this links:
http://www.linbit.com/en/products-and-services/drbd/drbd-rhel
http://www.linbit.com/support/drbd-8.4.4/
Note extra - re edited:
And please consider that DRBD is much faster that gluterfs, ceph or any other storage backend, above all when is used in the same PVE node, it consumes a very bit of processor, and it not have comparation of consumption of hardware resources above all compared with the other storage backends. The performance incomparable of DRBD is the key of success due to his high performance for synchronous replication, especially when databases are used in the VMs (my experoence is with MS-SQl Server).
Then I want to do a third question:
Very respectfully I want to ask: Why not PVE team want to do support for the latest stable version of drbd?... Sheedog not is stable (have many bugs), and PVE team give support for this storage.
Best regards
Cesar
This questions is for PVE developer team:
Currently PVE 3.1 has a kernel with support for DRBD 8.3.13, but DRBD 8.4.4 is much faster and more stable
please see this link:
http://blogs.linbit.com/p/469/843-random-writes-faster/
And I say more stable and faster in my personal exprience because I use these versions in production environments and developer environments since many months ago without any problem or inconveniences.
Then, I want to apply to PVE team one of two things if is possible:
1- PVE and his kernel have support for the latest version of DRBD, or
2- PVE and his kernel have "THE CORRECT KERNEL HEADER" for use it correctly with dkms, since that i have that modify the "kernel header" in each upgrade of PVE + compile DRBD = then, is very annoying because the extra work required in each PVE node
For this support, please consider these:
1- If PVE have doubts in anything, please let me know
2- For use with DKMS the solution is add these lines in the pve Kernel header:
lines are inserted from 45 to 49 in "idr.h" file:
41 #define MAX_ID_SHIFT (sizeof(int)*8 - 1)
42 #define MAX_ID_BIT (1U << MAX_ID_SHIFT)
43 #define MAX_ID_MASK (MAX_ID_BIT - 1)
44
45 /* Leave the possibility of an incomplete final layer */
46 #define MAX_LEVEL (MAX_ID_SHIFT + IDR_BITS - 1) / IDR_BITS
47
48 /* Number of id_layer structs to leave in free list */
49 #define IDR_FREE_MAX MAX_LEVEL + MAX_LEVEL
50
51 struct idr_layer {
52 unsigned long bitmap; /* A zero bit means "space here" */
53 struct idr_layer *ary[1<<IDR_BITS];
54 int count; /* When zero, we can release it */
55 int layer; /* distance from leaf */
56 struct rcu_head rcu_head;
57 };
Please PVE team let me know about of your think about of this topic, and please take into account that linbit certified his development for several linux distributions
Please see this links:
http://www.linbit.com/en/products-and-services/drbd/drbd-rhel
http://www.linbit.com/support/drbd-8.4.4/
Note extra - re edited:
And please consider that DRBD is much faster that gluterfs, ceph or any other storage backend, above all when is used in the same PVE node, it consumes a very bit of processor, and it not have comparation of consumption of hardware resources above all compared with the other storage backends. The performance incomparable of DRBD is the key of success due to his high performance for synchronous replication, especially when databases are used in the VMs (my experoence is with MS-SQl Server).
Then I want to do a third question:
Very respectfully I want to ask: Why not PVE team want to do support for the latest stable version of drbd?... Sheedog not is stable (have many bugs), and PVE team give support for this storage.
Best regards
Cesar
Last edited:
