Spinlocks are out, SMP-safe queues are in. All sites have been upgraded and have been running for several days without interruption.
Spinlocks, like any semaphore, should be avoided if possible - or so the conventional wisdom goes. And we understand a bit more about why it is considered conventional wisdom now.
SMP-safe queues have mutually exclusive read and write methods so that both operations can occur simultaneously. In our MTP2 implementation there is a single reader and writer for each of four queues. Three of the four queues are exposed to the potential for simultaneous reading and writing.
Thursday, March 16, 2006
Wednesday, March 08, 2006
Announcing the Sangoma Signal Media Gateway
The SS7 ISUP attachment to Linux based softswitches finally has a name: Sangoma Signal Media Gateway. We call it SMG for short. Having a product name makes conversations and documenting so much easier. Here's a link to a product description: http://sangoma.editme.com/wanpipe-linux-asterisk-ss7
Installations in Pakistan, Paraguay, Manilla, Rome, and Michigan are helping us hone the installation process and are getting efficient at finding quality issues (bugs). They are pushing us further and faster along than we would have on our own. Thanks and keep up the good work.
The kernel module MTP2 had a set of nasty bugs that were cured with restructuring the buffer management scheme and surrounding a few critical section in the code with spinlocks.
We have introduced a trouble ticket system to track bugs. We give users access to the system so they can check up on our progress and make sure we are staying focused.
We found that we'll need to support overlap signalling in the early release of the SMG. Several locations cannot mandate en bloc signalling. The specs are a little weak in specifying how overlap signalling should work. Thanks to some attentive and knowledgeable users we captured the information we needed to complete the requirements we need to implement overlap signalling.
We've been told repeatedly that serving multiple E1 installations is required immediately. We identified the key item preventing us from delivering this capability is automated testing. We didn't want to deliver a multiple E1 support until we tested hundreds of simultaneous calls on a lab system. So, we designed and, are currently building, a bulk calling lab. We are going to replicate smaller versions of this lab in several places. The main lab will be able to generate thousands of simultaneous calls, verify connections, and vary hold times.
Installations in Pakistan, Paraguay, Manilla, Rome, and Michigan are helping us hone the installation process and are getting efficient at finding quality issues (bugs). They are pushing us further and faster along than we would have on our own. Thanks and keep up the good work.
The kernel module MTP2 had a set of nasty bugs that were cured with restructuring the buffer management scheme and surrounding a few critical section in the code with spinlocks.
We have introduced a trouble ticket system to track bugs. We give users access to the system so they can check up on our progress and make sure we are staying focused.
We found that we'll need to support overlap signalling in the early release of the SMG. Several locations cannot mandate en bloc signalling. The specs are a little weak in specifying how overlap signalling should work. Thanks to some attentive and knowledgeable users we captured the information we needed to complete the requirements we need to implement overlap signalling.
We've been told repeatedly that serving multiple E1 installations is required immediately. We identified the key item preventing us from delivering this capability is automated testing. We didn't want to deliver a multiple E1 support until we tested hundreds of simultaneous calls on a lab system. So, we designed and, are currently building, a bulk calling lab. We are going to replicate smaller versions of this lab in several places. The main lab will be able to generate thousands of simultaneous calls, verify connections, and vary hold times.
Subscribe to:
Posts (Atom)